This page intentionally left blank

This page intentionally left blank

An Introduction to Programming with Mathematica
r
An Introduction to Programming with Mathematica
r
is de-
signed to introduce the Mathematica programming language
to a wide audience. Since the last edition of this book was
published, signicant changes have occurred in Mathemat-
ica and its use worldwide. Keeping pace with these changes,
this substantially larger, updated version includes new and
revised chapters on numerics, procedural, rule-based, and
front end programming, and gives signicant coverage to
the latest features up to, and including, version 5.1 of the
software.
Mathematica notebooks, available fromwww.cambridge.org/
0521846781, contain examples, programs, and solutions to
exercises in the book. Additionally, material to supplement
later versions of the software will be made available. This is
the ideal text for all scientic students, researchers, and pro-
grammers wishing to deepen their understanding of Math-
ematica, or even those keen to program using an interac-
tive language that contains programming paradigms from
all major programming languages: procedural, functional,
recursive, rule-based, and object-oriented.
An Introduction to
Programming
with
Mathematica
r
Third Edition
Paul R. Wellin | Richard J. Gaylord | Samuel N. Kamin
caxniioci uxiviisir\ iiiss
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, So Paulo
Cambridge University Press
The Edinburgh Building, Cambridge cn: :iu, UK
First published in print format
isnx-:, ,;-c-,::-o;-,
isnx-:, ,;-c-,::-cco:-,
Mathematica, Mathlink and Mathsource are registered trademarks of Wolfram
Research, Inc. All other trademarks used herein are the property of their respective
owners. Mathematica is not associated with Mathematica Policy Research, Inc. or
MathTech, Inc.
Wolfram Research is the holder of the copyright to the Mathematica software system
described in this document, including without limitation such aspects of the system as
its code, structure, sequence, organization, look and feel, programming language, and
compilation of command names. Use of the system unless pursuant to the terms of a
license granted by Wolfram Research or an otherwise authorized by law is an
infringement of the copyright.
2005
Information on this title: www.cambridge.org/9780521846783
This book is in copyright. Subject to statutory exception and to the provision of
relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
isnx-:c c-,::-cco:-x
isnx-:c c-,::-o;-:
Cambridge University Press has no responsibility for the persistence or accuracy of
uiis for external or third-party internet websites referred to in this book, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
hardback
eBook (NetLibrary)
eBook (NetLibrary)
hardback
1 An introduction to Mathematica
Mathematica is a very large and seemingly complex system. It contains hundreds of
functions for performing various tasks in science, mathematics, and engineering,
including computing, programming, data analysis, knowledge representation, and
visualization of information. In this introductory chapter, we introduce the elementary
operations in Mathematica and give a sense of its computational and programming
breadth and depth. In addition, we give some basic information that users of Mathemat-
ica need to know, such as how to start Mathematica, how to get out of it, how to enter
simple inputs and get answers, and finally how to use Mathematicas documentation to
get answers to questions about the system.
1.1 A brief overview of Mathematica
Numerical computations
Mathematica has been aptly described as a sophisticated calculator. With it you can enter
mathematical expressions and compute their values.
In[1]:= Sin .86 Log 1
.08
12
12
Out[1]= 0.481899
You can store values in memory.
In[2]:= rent 350
Out[2]= 350
In[3]:= food 175
Out[3]= 175
In[4]:= heat 83
Out[4]= 83
In[5]:= rent food heat
Out[5]= 608
Yet Mathematica differs from calculators and simple computer programs in its ability to
calculate exact results and to compute to an arbitrary degree of precision.
In[6]:=
1
15
1
35
1
63
Out[6]=
1
9
In[7]:= 2
500
Out[7]= 3273390607896141870013189696827599152216642046043064789483291
368096133796404674554883270092325904157150886684127560071009
217256545885393053328527589376
In[8]:= N , 500
Out[8]= 3.14159265358979323846264338327950288419716939937510582097494
459230781640628620899862803482534211706798214808651328230664
709384460955058223172535940812848111745028410270193852110555
964462294895493038196442881097566593344612847564823378678316
527120190914564856692346034861045432664821339360726024914127
372458700660631558817488152092096282925409171536436789259036
001133053054882046652138414695194151160943305727036575959195
309218611738193261179310511854807446237996274956735188575272
48912279381830119491
Symbolic computations
One of the more powerful features of Mathematica is its ability to manipulate and compute
with symbolic expressions. For example, you can factor polynomials and simplify trigono-
metric expressions.
In[9]:= Factor x
5
1
Out[9]= 1 x 1 x x
2
x
3
x
4
In[10]:= TrigReduce Sin
3
Out[10]=
1
4
3 Sin Sin 3
2 An Introduction to Programming with Mathematica
You can simplify expressions using assumptions about variables contained in those expres-
sions. For example, if k is assumed to be an integer, sin 2 k x simplifies to sin x .
In[11]:= Simplify Sin 2 k x , k Integers
Out[11]= Sin x
This computes the conditions for which a general quadratic polynomial will have both
roots equal to each other.
In[12]:= Reduce
x,a x
2
b x c 0 y,a y
2
b y c 0
x y , a, b, c
Out[12]= a 0 && b 0 a 0 && b c 0 a 0 && c
b
2
4 a
You can create functions that are defined piecewise.
In[13]:= Piecewise 1, x 0 , Sin x x
Out[13]=
1 x 0
Sin x
x
True
The knowledge base of Mathematica includes algorithms for solving polynomial equations,
and computing integrals.
In[14]:= Solve x
3
a x 1 0, x
Out[14]= x
2
3
1 3
a
9 3 27 4 a
3
1 3
9 3 27 4 a
3
1 3
2
1 3
3
2 3
,
x
1 3 a
2
2 3
3
1 3
9 3 27 4 a
3
1 3
1 3 9 3 27 4 a
3
1 3
2 2
1 3
3
2 3
,
x
1 3 a
2
2 3
3
1 3
9 3 27 4 a
3
1 3
1 3 9 3 27 4 a
3
1 3
2 2
1 3
3
2 3
In[15]:=
1
1 x
4
x
Out[15]=
1
4 2
2 ArcTan 1 2 x 2 ArcTan 1 2 x
Log 1 2 x x
2
Log 1 2 x x
2
1 An introduction to Mathematica 3
Graphics
The ability to visualize functions or sets of data often allows us greater insight into their
structure and properties. Mathematica provides a wide range of graphing capabilities.
These include two- and three-dimensional plots of functions or data sets, contour and
density plots of functions of two variables, bar charts, histograms and pie charts of data
sets, and many packages designed for specific graphical purposes. In addition, the Mathemat-
ica programming language allows you to construct graphical images from the ground up
using primitive elements, as we will see in Chapter 9.
Here is a simple two-dimensional plot of the function sin x 2 sin x
2
.
In[16]:= Plot Sin x 2 Sin x
2
, x, ,
3 2 1 1 2 3
1
0.5
0.5
1
Out[16]= Graphics
You can combine two or more plots in a single graphic by enclosing them inside curly
braces.
In[17]:= Plot Sin x , Sin 2 x , x, 0, 2 ;
1 2 3 4 5 6
1
0.5
0.5
1
4 An Introduction to Programming with Mathematica
Here is a plot of the sinc function, given in the previous section.
In[18]:= Plot Piecewise 1, x 0 , Sin x x , x, 2 , 2 ;
6 4 2 2 4 6
0.2
0.2
0.4
0.6
0.8
1
Here is a surface of constant negative curvature, represented parametrically by the three
functions , , and . This surface is often referred to as Dinis surface.
In[19]:= Cos Sin ;
Sin Sin ;
0.2 Cos Log Tan
2
;
In[22]:= ParametricPlot3D , , , , 0, 4 , , .05, 1 , Axes False,
Boxed False, PlotPoints 30, AspectRatio 1.75 ;
1 An introduction to Mathematica 5
Working with data
The ability to plot and visualize data is extremely important in engineering and all of the
social, natural, and physical sciences. Mathematica can import and export data from other
applications, plot the data in a variety of forms, and be used to perform numerical analysis
on the data.
The file dataset.m contains pairs of data points, in this case representing body
mass vs. heat production for 13 different animals. The data are given as (m, r), where m
represents the mass of the animal and r the heat production in kcal per day. First we set up
a platform independent path to the file and then import that file.
In[23]:= datafile ToFileName $BaseDirectory,
"Applications", "IPM3", "DataFiles" , "dataset.m"
Out[23]= C:\Documents and Settings\All Users\Application Data\
Mathematica\Applications\IPM3\DataFiles\dataset.m
In[24]:= data Import datafile, "Table"
Out[24]= 0.06099, 6.95099 , 0.403, 28.189 ,
0.62199, 41.1 , 2.50999, 120.799 ,
2.95999, 147.9 , 3.33, 182.8 , 8.19999, 368.8 ,
28.1999, 981.299 , 57.4, 1303.29 , 72.2999, 1512.5 ,
340.199, 7100.29 , 711, 10101.1 , 5000., 29894.9
You can immediately plot the data using the ListPlot function.
In[25]:= ListPlot data, PlotStyle PointSize .02 ;
200 400 600 800
2500
5000
7500
10000
12500
15000
17500
6 An Introduction to Programming with Mathematica
This plots the data on loglog axes.
In[26]:= logplot ListPlot Log data , PlotStyle PointSize .02 ;
2 2 4 6 8
4
6
8
10
You can then fit a straight line to the log-data by performing a linear least squares fit. In
this example, we are fitting to the model a mx, where a and m are the parameters to be
determined in the model with variable x.
In[27]:= f FindFit Log data , a m x, a, m , x
Out[27]= a 4.15437, m 0.761465
Here is a plot of the linear fit function.
In[28]:= fplot Plot a m x . f, x, 3, 9 ;
2 2 4 6 8
4
6
8
10
1 An introduction to Mathematica 7
Finally, you can see how well the fitted function approximates the log plot by combining
these last two graphics.
In[29]:= Show fplot, logplot ;
2 2 4 6 8
4
6
8
10
Programming
With a copy of The Mathematica Book (Wolfram 2003) or one of the many tutorial books
(see, for example, Glynn and Gray 1999) describing the vast array of computational tasks
that can be performed with Mathematica, it would seem you can compute just about
anything you might want. But that impression is mistaken. There are simply more kinds of
calculations than could possibly be included in a single program. Whether you are inter-
ested in computing bowling scores or finding the mean square distance of a random walk
on a torus, Mathematica does not have a built-in function to do everything that a user could
possibly want. What it does have and what really makes it the amazingly useful tool it is
is the capability for users to define their own functions. This is called programming, and it
is what this book is all about.
Sometimes, the programs you create will be succinct and focused on a very specific
task. Mathematica possesses a rich set of tools that enable you to quickly and naturally
translate the statement of a problem into a program. For example, the following program
defines a test for perfect numbers, numbers that are equal to the sum of their proper
divisors.
In[30]:= PerfectQ n_ : Apply Plus, Divisors n 2 n
We then define another function that selects those numbers from a range of integers that
pass this PerfectQ test.
In[31]:= PerfectSearch n_ : Select Range n , PerfectQ
8 An Introduction to Programming with Mathematica
This then finds all perfect numbers less than 1,000,000.
In[32]:= PerfectSearch 10
6
Out[32]= 6, 28, 496, 8128
Here are two functions for representing regular polygons. The first defines the
vertices of a regular n-gon, while the second uses those vertices to create a polygon graph-
ics object that can then be displayed with the built-in Show function.
In[33]:= vertices n_Integer, r_: 1 :
Table r Cos
2
n
, r Sin
2
n
, , 0, n 1
In[34]:= RegularPolygon n_ :
Graphics Line vertices n . a_, b__ a, b, a ,
AspectRatio Automatic
In[35]:= Show RegularPolygon 5
Out[35]= Graphics
As another example of a succinct program, here is an iterative function that imple-
ments the well-known Newton method for root finding.
In[36]:= NewtonZero f_, xi_ : NestWhile #
f #
f' #
&, xi, Unequal, 2
In[37]:= g x_ : x
3
2 x
2
1
In[38]:= NewtonZero g, 2.0
Out[38]= 1.61803
Of course, sometimes the task at hand requires a more involved program, stretching
across several lines (or even pages) of code. For example, here is a slightly longer program
to compute the score of a game of bowling, given a list of the number of pins scored by
each ball.
1 An introduction to Mathematica 9
In[39]:= BowlingScore pins_ :
Module score , score x_, y_, z_ : x y z;
score 10, y_, z_, r___ : 10 y z score y, z, r ;
score x_, y_, z_, r___ :
x y z score z, r ; x y 10;
score x_, y_, r___ : x y score r ; x y 10;
score If pins 2 pins 1 10, pins, Append pins, 0
Here is the computation for a perfect game 12 strikes in a row.
In[40]:= BowlingScore 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10
Out[40]= 300
These examples use a variety of programming styles: functional programming,
rule-based programming, the use of anonymous functions, and more. We do not expect
you to understand the examples in this section at this point that is why we wrote this
book! What you should understand is that in many ways Mathematica is designed to be as
broadly useful as possible and that there are many calculations for which Mathematica does
not have a built-in function, so, to make full use of its many capabilities, you will some-
times need to program. The main purpose of this book is to show you how.
Another purpose is to teach you the basic principles of programming. These princi-
ples making assignments, defining rules, using conditionals, recursion, and iteration are
applicable (with great differences in detail, to be sure) to all other programming languages.
Symbolic and interactive documents
In addition to the computational tools that Mathematica provides for what many profession-
als associate with technical computing, it also contains tools for creating and modifying the
user interface to such tasks. These tools include hyperlinks for jumping to other locations
within a document or across files, buttons to perform tasks that you might normally
associate with a command-line interface, and tools to modify and manipulate the appear-
ance and functionality of your Mathematica notebooks directly. In this section we will give
a few short examples of what is possible, waiting until Chapter 10 for a methodical look at
how to program these elements.
The first example takes the code necessary to display a polyhedron and puts it in a
button. The two lines of code that could be evaluated normally in a notebook first load a
package and then display an icosahedron in the notebook.
In[41]:= Needs "Graphics`Polyhedra`"
10 An Introduction to Programming with Mathematica
In[42]:= Show Stellate Polyhedron Icosahedron
Out[42]= Graphics3D
Here is a short program that creates a button containing the above two expressions.
Cell[BoxData[
ButtonBox[
RowBox[{"Stellate", " ", "Icosahedron"}],
ButtonFunction:>CompoundExpression[
Needs[ "Graphics`Polyhedra`"],
Show[Stellate[Polyhedron[Icosahedron]]]
],
ButtonEvaluator->Automatic],
"Input",
Active->True]
The formatted version of the above cell can be displayed by choosing Show Expression
from the Format menu. When you do that, it will look like the following:
Stellate Icosahedron
Clicking the button will cause the Mathematica code in the ButtonFunction to be
immediately evaluated and the following graphics will then be displayed in your notebook.
Functions are available to jump around to different parts of a Mathematica notebook
and perform various actions. Here is a short piece of code that creates a button which,
upon being clicked, moves the selection to the next cell and then evaluates that cell.
1 An introduction to Mathematica 11
Cell[TextData[{
Cell[BoxData[
ButtonBox["EVALUATE",
ButtonFunction:>FrontEndExecute[ {
FrontEnd`SelectionMove[
ButtonNotebook[ ], All, ButtonCell],
FrontEnd`SelectionMove[
ButtonNotebook[ ], Next, Cell],
FrontEnd`SelectionEvaluate[
ButtonNotebook[ ]]}],
Active->True]]],
StyleBox[" MATHEMATICA INPUT"]
}], "Text"]
The formatted version of the above cell can be displayed by choosing Show Expression
from the Format menu. When you do that, it will look somewhat like the following
(although we have removed some of the text formatting above to improve readability of the
code). Clicking the EVALUATE button will cause the input cell immediately following to
be selected and then evaluated.
EVALUATE MATHEMATICAINPUT
In[43]:= 3 4 5
Out[43]= 27
The following example demonstrates how you can use Mathematica functions to
perform some of the user interface actions that you would normally associate with key-
board and mouse events. By using such techniques, you can create a specific set of actions
that will follow certain evaluations. For example, if you were creating an electronic quiz for
your students, you could include hint buttons within your class notebooks that would
open a new notebook with hints and suggestions upon clicking.
This creates a new notebook that contains three cells a Section cell, a Text cell,
and an Input cell. Upon evaluation, the NotebookPut command below will cause a new
notebook to appear, containing the three specified cells. The screen shots below show
what appears in the user interface after evaluating each of the preceding inputs.
12 An Introduction to Programming with Mathematica
In[44]:= nb NotebookPut
Notebook
Cell "Symbolic and Interactive Documents", "Section" ,
Cell "Cells and notebooks are Mathematica expressions.",
"Text" ,
Cell "Integrate Sin x Cos x ,x ", "Input"
Out[44]= NotebookObject Untitled 1
This moves the selection bar past the last cell in the above notebook.
In[45]:= SelectionMove nb, Next, Cell, 4
1 An introduction to Mathematica 13
We then select the most previous cell.
In[46]:= SelectionMove nb, Previous, Cell
Finally, we evaluate the selected cell.
In[47]:= SelectionEvaluate nb
In Chapter 10 we will give a detailed discussion of how to modify and manipulate the
user interface through the use of the symbolic programming techniques that are discussed
throughout this book.
14 An Introduction to Programming with Mathematica
1.2 Using Mathematica
Before you can do any serious work, you will need to know how to get a Mathematica
session started, how to stop it, and how to get out of trouble when you get into it. These
procedures depend somewhat on the system you are using. You should read the system-spe-
cific information that came with your copy of Mathematica; and you may need to consult a
local Mathematica guru if our advice here is not applicable to your system.
Getting into and out of Mathematica
The most commonly used interface is often referred to as a notebook interface in which
the user creates and works in interactive documents. Personal computers running Win-
dows, Macintosh operating systems, Linux, and most flavors of Unix all support this
graphical user interface, which normally starts up automatically when you begin your
Mathematica session.
There are some situations where you may want to start up Mathematica from a
command prompt and issue commands directly through that interface, bypassing the
notebook interface entirely. For example, you may have a very long computation that you
need to run in batch mode. Typically, Mathematica is started up on these systems by typing
math at a command prompt. We will not discuss using Mathematica through a command
prompt any further. If you are interested in this mode you should consult the documenta-
tion that came with your copy of Mathematica.
Starting Mathematica and first computations
To start Mathematica you will have to find and then double-click on the Mathematica icon
on your computer, which will look something like this:
The computer will then load parts of Mathematica into its memory and soon a blank
window will appear on the screen. This window is the visual interface to a Mathematica
notebook and it has many features that are useful to the user.
Notebooks allow you to write text, perform computations, write and run programs,
and create graphics all in one document. Notebooks also have many of the features of
common word processors, so those familiar with word processing will find the notebook
interface easy to learn. In addition, the notebook provides features for outlining material
which you may find useful for giving talks and demonstrations.
1 An introduction to Mathematica 15
When a blank notebook first appears on the screen (either from just starting Mathe-
matica or from selecting New in the File menu), you can start typing immediately. For
example, if you type N[Pi,200] press (hold down the Shift key while pressing
the Enter key) to evaluate an expression. Mathematica will evaluate the result and print the
200-decimal digit approximation to on the screen.
Notice that when you evaluate an expression in a notebook, Mathematica adds input
and output prompts. In the example notebook above, these are denoted In[1]:= and
Out[1]=. These prompts can be thought of as markers (or labels) that you can refer to
during your Mathematica session.
16 An Introduction to Programming with Mathematica
You should also note that when you started typing Mathematica placed a bracket on
the far right side of the window that enclosed the cell that you were working in. These cell
brackets are helpful for organizational purposes within the notebook. Double-clicking on
cell brackets will open any collapsed cells, or close any open cells as can be seen in the
previous screen shot.
Double-clicking on the cell bracket containing the 1.1 A Brief Overview of Mathe-
matica cell will open the cell to display its contents:
Using cell brackets in this manner allows you to organize your work in an orderly
manner, as well as to outline material. For a complete description of cell brackets and
many other interface features, you should consult the documentation that came with your
version of Mathematica.
For information on other features such as saving, printing, and editing notebooks,
consult the manuals that came with your version of Mathematica.
Entering input
New input can be entered whenever there is a horizontal line that runs across the width of
the notebook. If one is not present where you wish to place an input cell, move the cursor
up and down until it changes to a horizontal bar and then click the mouse once. A horizon-
tal line should now appear across the width of the window. You can immediately start
typing and an input cell will be created.
1 An introduction to Mathematica 17
Input can be entered exactly as it appears in this book. To get Mathematica to evalu-
ate an expression that you have entered, press - ; that is, hold down the Shift key
and then press the Enter key.
You can enter mathematical expressions in a traditional looking two-dimensional
format using either palettes for quick entry of template expressions, or keyboard equiva-
lents. For example, the following expression can be entered by using the Basic Input
palette, or through a series of keystrokes. For details of inputting mathematical expres-
sions, read your user documentation or read the section on 2D Expression Input in the
Help Browser.
In[1]:=
1
1 x
3
x
Out[1]=
ArcTan
1 2 x
3
3
1
3
Log 1 x
1
6
Log 1 x x
2
As noted previously, Mathematica enters the In and Out prompts for you. You do
not type these prompts. You will see themafter you evaluate your input.
You can refer to the result of the previous calculation using the symbol %.
In[2]:= 2
64
Out[2]= 18446744073709551616
In[3]:= % 1
Out[3]= 18446744073709551617
You can also refer to the result of any earlier calculation using its Out[i] label or,
equivalently, %i.
In[4]:= Out 1
Out[4]=
ArcTan
1 2 x
3
3
1
3
Log 1 x
1
6
Log 1 x x
2
In[5]:= %2
Out[5]= 18446744073709551616
Ending a Mathematica session
To end your Mathematica session, choose Exit from the File menu. You will be prompted
to save any unsaved open notebooks.
18 An Introduction to Programming with Mathematica
Getting out of trouble
From time to time, you will type an input which will cause Mathematica to misbehave in
some way, perhaps by just going silent for a long time (if, for example, you have inadvert-
ently asked it to do something very difficult) or perhaps by printing out screen after screen
of not terribly useful information. In this case, you can try to interrupt the calculation.
How you do this depends on your computers operating system:
Macintosh: type . (the Command key and the period) and then type a
Windows 95/98/NT/2000/XP: type . (the Alt key and the period)
Unix: type - . and then type a and then
These attempts to stop the computation will sometimes fail. If after waiting a reason-
able amount of time (say, a few minutes), Mathematica still seems to be stuck, you will have
to kill the kernel. (Before attempting to kill the kernel, try to convince yourself that the
computation is really in a loop from which it will not return and that it is not just an
intensive computation that requires a lot of time.) Killing the kernel is accomplished by
selecting Quit Kernel from the Kernel menu. The kernel can then be restarted without
killing the front end by first selecting Start Kernel Local under the Kernel menu, or you
can simply evaluate a command in a notebook and a new kernel should start up
automatically.
The syntax of inputs
You can enter mathematical expressions in a linear syntax using arithmetic operators
common to almost all computer languages.
In[6]:= 39 13
Out[6]= 3
Alternately, you can enter this expression in the traditional form by typing 39, / , then
13.
In[7]:=
39
13
Out[7]= 3
The caret (^) is used for exponentiation.
In[8]:= 2^5
Out[8]= 32
1 An introduction to Mathematica 19
You can enter this expression in a more traditional typeset form by typing 2, ^ , and
then 5.
In[9]:= 2
5
Out[9]= 32
Mathematica includes several different ways of entering typeset expressions, either
directly from the keyboard as we did above, or via palettes available from the File menu.
Below is a brief table showing some of the more commonly used typeset expressions and
how they are entered through the keyboard. You should read your documentation and
become comfortable using these input interfaces so that you can easily enter the kinds of
expressions in this book.
Expression FullForm Keyboardshortcut
x
2
SuperscriptBox x, 2 x 6 , 2
x
i
SubscriptBox x, i x , i
x
y
FractionBox x, y x , y
x SqrtBox x 2 , x
x y GreaterEqual x, y x , y
Table 1.1: Entering typeset expressions
You can indicate multiplication by simply putting a space between the two factors, as
in mathematics. You can also use the asterisk (*) for that purpose, as is traditional in most
computer languages.
In[10]:= 2 5
Out[10]= 10
In[11]:= 2 5
Out[11]= 10
Mathematica also gives operations the same precedence as in mathematics. In particu-
lar, multiplication and division have a higher precedence than addition and subtraction, so
that 3 + 4 * 5 equals 23 and not 35.
In[12]:= 3 4 5
Out[12]= 23
20 An Introduction to Programming with Mathematica
Functions are also written as they are in mathematics books, except that function
names are capitalized and their arguments are enclosed in square brackets.
In[13]:= Factor x
5
1
Out[13]= 1 x 1 x x
2
x
3
x
4
Almost all of the built-in functions are spelled out in full, as in the above example.
The exceptions to this rule are well-known abbreviations such as D for differentiation,
Sqrt for square roots, Log for logarithms, and Det for the determinant of a matrix.
Spelling out the name of a function in full is quite useful when you are not sure whether a
function exists to perform a particular task. For example, if we wanted to compute the
conjugate of a complex number, an educated guess would be:
In[14]:= Conjugate 3 4
Out[14]= 3 4
Whereas square brackets [ and ] are used to enclose the arguments to functions,
curly braces { and } are used to indicate a list or range of values. Lists are a basic data type
in Mathematica and are used to represent vectors and matrices (and tensors of any dimen-
sion), as well as additional arguments to functions such as in Plot and Integrate.
In[15]:= a, b, c . x, y, z
Out[15]= a x b y c z
In[16]:= Plot Sin x 2 Sin x , x, 2 , 2 ;
6 4 2 2 4 6
1
0.5
0.5
1
In[17]:= Integrate[Cos[x], {x, a, b}]
Out[17]= Sin a Sin b
In the Plot example, the list {x,-2 ,2 } indicates that the function
sin x 2 sin x is to be plotted over an interval as x takes on values from 2 to 2 .
The Integrate expression above is equivalent to the mathematical expression
a
b
cos x x.
1 An introduction to Mathematica 21
Mathematica has very powerful list-manipulating capabilities that will be explored in
detail in Chapter 3.
When you end an expression with a semicolon (;), Mathematica computes its value
but does not display it. This is very helpful when the result of the expression would be very
long and you do not need to see it. In the following example, we first create a list of the
integers from 1 to 10,000, suppressing their display with the semicolon, and then compute
their sum and average.
In[18]:= nums Range 10000 ;
In[19]:= Apply[Plus, nums]
Out[19]= 50005000
In[20]:=
%
Length nums
Out[20]=
10001
2
An expression can be entered on multiple lines, but only if Mathematica can tell that
it is not finished after the first line. For example, you can enter 3* on one line and 4 on the
next.
In[21]:= 3 *
4
Out[21]= 12
But you cannot enter 3 on the first line and *4 on the second.
In[22]:= 3
*4
Out[22]= 3
If you use parentheses, you can avoid this problem.
In[23]:= (3
*4)
Out[23]= 12
With the notebook interface, you can input as many lines as you like within an input
cell; Mathematica will evaluate them all when you enter still obeying the rules
stated above for any incomplete lines.
22 An Introduction to Programming with Mathematica
Finally, you can enter a comment some words that are not evaluated by entering
the words between (* and *).
In[24]:= D[Sin[x], (* differentiate Sin[x] *)
{x, 1}] (* with respect to x once *)
Out[24]= Cos x
Alternate input syntax
There are several different ways to write expressions in Mathematica. Usually, you will
simply use the traditional notation, fun[x], for example. But you should be aware of
several alternatives to this syntax that are widely used.
Here is an example using the standard function notation for writing a function with
one argument.
In[25]:= N
Out[25]= 3.14159
This uses a prefix operator.
In[26]:= N
Out[26]= 3.14159
Here is a postfix operator notation.
In[27]:= N
Out[27]= 3.14159
For functions with two arguments, you can use an infix notation. The following
expression is identical to N[ ,30].
In[28]:= N 30
Out[28]= 3.14159265358979323846264338328
Finally, many people prefer to use a more traditional syntax when entering and
working with mathematical expressions. You can compute an integral using standard
Mathematica syntax.
In[29]:= Integrate 1 Sin x , x
Out[29]= Log Cos
x
2
Log Sin
x
2
1 An introduction to Mathematica 23
The same integral, represented in a more traditional manner, can be entered from palettes
or keyboard shortcuts.
In[30]:=
1
Sin x
x
Out[30]= Log Cos
x
2
Log Sin
x
2
Many mathematical functions have traditional symbols associated with their opera-
tions and when available these can be used instead of the fully spelled-out names. For
example, you can compute the intersection of two sets using the Intersection function.
In[31]:= Intersection a, b, c, d, e , b, f, a, z
Out[31]= a, b
Or you can do the same computation using more traditional notation.
In[32]:= a, b, c, d, e b, f, a, z
Out[32]= a, b
To learn how to enter these and other notations quickly, either from palettes or
directly from the keyboard using shortcuts, refer to the 2D Expression Input section in the
Front End category of the Help Browser.
The front end and the kernel
When you work in Mathematica you are actually working with two separate programs.
They are referred to as the front end and the kernel. The front end is the user interface. It
consists of the notebooks that you work in together with the menu system, palettes (which
are really just notebooks), and any element that accepts input from the keyboard or mouse.
The kernel is the program that does the calculations. So a typical operation between the
user (you) and Mathematica consists of the following steps, where the program that is
invoked in each step is indicated in parentheses:
enter input in the notebook (front end)
send input to the kernel to be evaluated by pressing - (front end)
kernel does the computation and sends it back to the front end (kernel)
result is displayed in the notebook (front end)
There is one remaining piece that we have not yet mentioned; that is MathLink.
Since the kernel and front end are two separate programs, a means of communication is
24 An Introduction to Programming with Mathematica
necessary for these two programs to talk to each other. That communication protocol is
called MathLink and it comes bundled with Mathematica. It operates behind the scenes,
completely transparent to the user.
MathLink is a very general communications protocol that is not limited to communi-
cation between the front end and the kernel, but can also be used to set up communication
between the front end and other programs on your computer, programs like compiled C
and Fortran code. It can also be used to connect a kernel to a word processor or spread-
sheet or many other programs.
MathLink programming is beyond the scope of this book, but if you are interested,
there are several books and articles that discuss it (see the References at the end of this
book).
Errors
In the course of using and programming in Mathematica, you will encounter various sorts
of errors, some obvious, some very subtle, some easily rectified, and others not. We have
already mentioned that it is possible to send Mathematica into an infinite loop from which
it cannot return. In this section, we discuss those situations where Mathematica does finish
the computation, but without giving you the answer you expected.
Perhaps the most frequent error you will make is misspelling the name of a function.
Here is an illustration of the kind of thing that will usually happen in this case.
In[33]:= Sine 1.5
General::spell :
Possible spelling error: new symbol name "Sine" is
similar to existing symbols Line, Sin, Sinh . More
Out[33]= Sine 1.5
Whenever you type a name that is close to an existing name, Mathematica will print a
warning message like the one above. You may often use such names intentionally, in which
case these messages can be annoying. In that case, it is best to turn off the warnings.
In[34]:= Off General::spell
Now, Mathematica will not report that function names might be misspelled; and,
when it cannot find a definition associated with a misspelled function, it returns your input
unevaluated.
In[35]:= Intergate x
2
, x
Out[35]= Intergate x
2
, x
1 An introduction to Mathematica 25
You can turn these spell warnings back on by evaluating On[General::spell].
In[36]:= On General::spell
Having your original expression returned unevaluated as if this were perfectly
normal is a problem you will often run into. Aside from misspelling a function name, or
simply using a function that does not exist, another case where this occurs is when you give
the wrong number of arguments to a function, especially to a user-defined function. For
example, the BowlingScore function takes a single list argument; if we accidentally leave
out the list braces, then we are actually giving BowlingScore 12 arguments.
In[37]:= BowlingScore 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10
Out[37]= BowlingScore 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10
Of course, some kinds of inputs cause genuine error messages. Syntax errors, as
shown above, are one example. The built-in functions are designed to usually warn you of
such errors in input. In the first example below, we have supplied the Log function with an
incorrect number of arguments (it expects one or two arguments only). In the second
example, FactorInteger operates on integers only and so the real number argument
causes the error condition.
In[38]:= Log 2, 16, 3
Log::argt : Log called with 3
arguments; 1 or 2 arguments are expected. More
Out[38]= Log 2, 16, 3
In[39]:= FactorInteger 12.5
FactorInteger::facn : Argument 12.5` in
FactorInteger 12.5 is not an exact number. More
Out[39]= FactorInteger 12.5
Getting help
Mathematica contains a vast array of documentation that you can access in a variety of
ways. It is also designed so that you can create new documentation for your own functions
and program in such a way that users of your programs can get help in exactly the same
way as they would for Mathematicas built-in functions.
If you are aware of the name of a function but are unsure of its syntax or what it does,
the easiest way to find out about it is to evaluate ?function. For example, here is the
usage message for ParametricPlot.
26 An Introduction to Programming with Mathematica
In[40]:= ?ParametricPlot
ParametricPlot fx, fy , u, umin, umax produces a parametric
plot of a curve with x and y coordinates fx and fy generated
as a function of t. ParametricPlot fx, fy , gx, gy , ... ,
u, umin, umax plots several parametric curves. More
Also, if you were not sure of the spelling of a command (Integrate, for example),
you could type the following to display all built-in functions that start with Integ.
In[41]:= ?Integ*
System`
Integer IntegerExponent IntegerQ Integrate
IntegerDigits IntegerPart Integers
Clicking on one of these links will produce a short usage statement about that
function. For example, if you were to click on the Integrate link, here is what would be
displayed in your notebook.
Integrate f, x gives the indefinite integral of f with respect
to x. Integrate f, x, xmin, xmax gives the definite
integral of f with respect to x from xmin to xmax. Integrate
f, x, xmin, xmax , y, ymin, ymax gives a multiple
definite integral of f with respect to x and y. More
Clicking the More hyperlink would take you directly to the Help Browser where a
much more detailed explanation of this function can be found.
You can also get help by highlighting any Mathematica function and pressing the F1
key on your keyboard. This will take you directly into the documentation for that function
in the Help Browser.
The Help Browser
Mathematica contains a very useful addition to the help system called the Help Browser.
The Help Browser allows you to search for functions easily and it provides extensive
documentation and examples.
To start the Help Browser, select Help Browser under the Help menu. You should
quickly see something like the following:
1 An introduction to Mathematica 27
Notice the eight category tabs near the top of the Help Browser window. Choosing
the Add-ons & Links tab will give you access to all of the packages that come with each
implementation of Mathematica. Similarly, choosing The Mathematica Book tab will give
you access to the entire Mathematica book that ships with each professional version of
Mathematica.
Suppose you were looking for information about three-dimensional parametric
graphics. First click the Built-in Functions tab, then select Graphics and Sound on the left,
then 3D Plots and finally ParametricPlot3D. The Help Browser should look like this:
Notice that in the main window, the Help Browser has displayed information about
the ParametricPlot3D function. This is identical to the usage message you would get
if you entered ?ParametricPlot3D.
28 An Introduction to Programming with Mathematica
Alternatively, you could have clicked the Master Index tab and searched for Paramet-
ricPlot3D or even simply parametric and then browsed through the index to find what
you were looking for.
Many additional features are available in the Help Browser and you are advised to
consult your documentation for a complete list and description.
1 An introduction to Mathematica 29
2 The Mathematica language
Expressions are the basic building blocks from which everything is built. Their
structure, internal representation, and how they are evaluated are essential to under-
standing Mathematica. In this chapter we focus on the Mathematica language with
particular emphasis on the structure and syntax of expressions. We will also look at
how to define and name new expressions, how to combine them using logical opera-
tors, and how to control properties of your expressions through the use of attributes.
2.1 Expressions
Introduction
Although it may appear different at first, everything that you will work with in Mathematica
has a similar underlying structure. This means things like a simple computation, a data
object, a graphic, the cells in your Mathematica notebook, even your notebook itself, all
have a similar structure they are all expressions, and an understanding of expressions is
essential to mastering Mathematica.
Internal forms of expressions
When doing a simple arithmetic operation such as 3 4 5, you are usually not concerned
with exactly how a system such as Mathematica actually performs the additions or multiplica-
tions. Yet you will find it extremely useful to be able to see the internal representation of
such expressions as this will allow you to manipulate them in a consistent and powerful
manner.
Internally, Mathematica groups the objects that it operates on into different types:
integers are distinct from real numbers; lists are distinct from numbers. One of the reasons
that it is useful to identify these different data types is that specialized algorithms can be
used on certain classes of objects that will help to speed up the computations involved.
The Head function can be used to identify types of objects. For numbers, it will
report whether the number is an integer, a rational number, a real number, or a complex
number.
In[1]:= Head 7 , Head
1
7
, Head 7.0 , Head 7 2
Out[1]= Integer, Rational, Real, Complex
In fact, every Mathematica expression has a Head that gives some information about
that type of expression.
In[2]:= Head a b
Out[2]= Plus
In[3]:= Head 1, 2, 3, 4, 5
Out[3]= List
Atoms
The three basic building blocks of Mathematica the atoms from which every expression
is ultimately constructed are, symbols, numbers, and strings.
A symbol consists of a letter followed without interruption by letters and numbers.
For example, both f and the built-in Integrate are symbols.
In[4]:= Head f
Out[4]= Symbol
In[5]:= Head Integrate
Out[5]= Symbol
In Mathematica, built-in constants all are Symbols.
In[6]:= Head , Head , Head EulerGamma
Out[6]= Symbol, Symbol, Symbol
Symbols can consist of one or more concatenated characters so long as they do not begin
with a number.
In[7]:= Head myfunc
Out[7]= Symbol
32 An Introduction to Programming with Mathematica
The four kinds of numbers integers, real numbers, complex numbers and rational
numbers are shown in the list below.
In[8]:= Head 4 , Head
5
7
, Head 5.201 , Head 3 4
Out[8]= Integer, Rational, Real, Complex
A string is composed of characters and is enclosed in quotes. They will be discussed
in detail in Section 3.5.
In[9]:= Head "Mathematica"
Out[9]= String
The structure of expressions
As mentioned earlier, everything in Mathematica is an expression. Expressions are either
atomic, as described in the previous section, or they are normal expressions, which have a
head and contain zero or more elements. Normal expressions are of the following form,
where h is the head of the expression and the e
i
are the elements which may themselves be
atomic or normal expressions.
h e
1
, e
2
, , e
n
Although we indicated that you can use Head to determine the type of atomic
expressions, this is entirely general. For normal expressions, Head simply gives the head of
that expression.
In[10]:= Head a b c
Out[10]= Plus
To see the full internal representation of an expression, use FullForm.
In[11]:= FullForm a b c
Out[11]//FullForm=
Plus a, b, c
In[12]:= FullForm a, b, c
Out[12]//FullForm=
List a, b, c
The important thing to notice is that both of these objects (the sum and the list) have
very similar internal representations. Each is made up of a function (Plus and List,
2 The Mathematica language 33
respectively), each encloses its arguments in square brackets, and each separates its argu-
ments with commas.
Regardless of how an atomic or normal expression may appear in your notebook, its
structure is uniquely determined by its head and parts as seen using FullForm. This is
important for understanding the Mathematica evaluation mechanism which depends on the
matching of patterns based on their FullForm representation, a subject we will turn to in
detail in Chapter 6.
The number of elements in any expression is given by its length.
In[13]:= Length a b c
Out[13]= 3
Here is a more complicated expression.
In[14]:= expr Sin x a x
2
b x c
Out[14]= c b x a x
2
Sin x
Its head is Times because it is composed of the product of Sin[x] and the quadratic
polynomial.
In[15]:= Head expr
Out[15]= Times
Its length is 2 since it only contains two factors.
In[16]:= Length expr
Out[16]= 2
Although the FullForm of this expression is a little harder to decipher, if you look
carefully you should see that it is composed of the product of Plus[c,Times[b,x],
Times[a,Power[x,2]]] and Sin[x].
In[17]:= FullForm expr
Out[17]//FullForm=
Times Plus c, Times b, x , Times a, Power x, 2 , Sin x
There are several important differences between atomic expressions and nonatomic
expressions. While the heads of all expressions are extracted in the same way using the
Head function the head of an atom provides different information than the head of other
expressions. For example, the head of a symbol or string is the kind of atom that it is.
In[18]:= Head Integrate
Out[18]= Symbol
34 An Introduction to Programming with Mathematica
In[19]:= Head "hello"
Out[19]= String
The head of a number is the specific kind of number that it is, its data type.
In[20]:= Head 2
Out[20]= Integer
In[21]:= Head 5.21
Out[21]= Real
The FullForm of an atom (except a complex or rational number) is the atom itself.
In[22]:= FullForm f
Out[22]//FullForm=
f
In[23]:= FullForm
5
7
Out[23]//FullForm=
Rational 5, 7
Atoms have no parts (which of course is why they are called atoms). In contrast,
nonatomic expressions do have parts. To extract different parts of an expression, use the
Part function. For example, the first part of the expression a+b is a.
In[24]:= Part a b, 1
Out[24]= a
The second part is b.
In[25]:= Part a b, 2
Out[25]= b
This should be clearer from looking at the internal representation of this expression.
In[26]:= FullForm a b
Out[26]//FullForm=
Plus a, b
So Part[a+b,1] is another way of asking for the first element of Plus[a,b], which is
just a. In general Part[expr,n] gives the nth element of expr.
2 The Mathematica language 35
It is worth noting that the 0th part is the Head of the expression.
In[27]:= Part a b, 0
Out[27]= Plus
As we stated above, atomic expressions have no parts.
In[28]:= Part "read my lips", 1
Part::partd : Part specification
read my lips 1 is longer than depth of object. More
Out[28]= read my lips 1
This error message indicates that the string "read my lips" has no first part, since it is
atomic. The expression expr[[1]] is shorthand for Part[expr,1].
Similarly, complex numbers are atomic and hence have no parts.
In[29]:= 3 4 1
Part::partd : Part specification
3 4 1 is longer than depth of object. More
Out[29]= 3 4 1
Because everything in Mathematica has the common structure of an expression, most
of the built-in functions that are used for list manipulation, such as Part, can also be used
to manipulate the arguments of any other kind of expression (except atoms).
In[30]:= Append w x y, z
Out[30]= w x y z
This result can best be understood by looking at the FullForm of the following two
expressions.
In[31]:= FullForm w x y
Out[31]//FullForm=
Plus w, Times x, y
In[32]:= FullForm w x y z
Out[32]//FullForm=
Plus w, Times x, y , z
Appending z to w+x y is equivalent to adding z as an argument to the Plus function.
More generally:
In[33]:= Append f a, b , c
Out[33]= f a, b, c
36 An Introduction to Programming with Mathematica
Finally, for more complicated expressions, you might find it useful to display the
internal representation with the TreeForm function, which shows the tree structure of
an expression. In the following example, the root node of the tree is Plus, which then
branches three times at c, bx, and at ax
2
, the latter two branching further.
In[34]:= TreeForm a x
2
b x c
Out[34]//TreeForm=
Plus c,
Times b, x
,
Times a,
Power x, 2
Exercises
1. Give the full (internal) form of the expression a(b+c).
2. What is the traditional representation of Times[a,Power[Plus[b,c],-1]].
3. What do you expect to be the result of the following operations? Use the FullForm
of the expressions to understand what is going on.
a. ((x^2 + y) z/w) [[2, 1, 2]]
b. (a/b)[[2, 2]]
4. What is the part specification of the b in the expression a x^2 + b x + c?
2.2 Definitions
Defining variables and functions
One of the most common tasks in any programming environment is to define functions,
constants, and procedures to perform various tasks. Sometimes a particular function that
you need is not part of the built-in set of functions. Other times, you may need to use an
expression over and over again and so it would be useful to define it once and have it
available for future reference. Because you want your newly defined expressions to work
with all the built-in functions seamlessly, by defining your own functions and constants
you essentially expand the range of Mathematicas capabilities.
2 The Mathematica language 37
For example, you might define a constant a to have a certain numeric value.
In[1]:= a N 2
Out[1]= 6.28319
Then, whenever a is used in a subsequent computation, Mathematica will find a rule
associated with a and will substitute that value wherever a occurs.
In[2]:= Cos a
Out[2]= 1.
To check what definitions are associated with a, use ?a.
In[3]:= ?a
Global`a
a 6.28319
To define a rule for a function f, enclose its arguments in square brackets and use x_
to indicate the variable that will be substituted for x on the right-hand side.
In[4]:= f x_
1
1 x
Out[4]=
1
1 x
The expression f[x_] on the left side of this assignment is a pattern. It indicates the class
of expressions for which this definition should be used. We will have much more to say
about patterns and pattern matching in Mathematica in Chapter 6, but, for now, it is
enough to say that the pattern f[x_] matches f[any expression].
You can evaluate f at different values by replacing x with the value you wish to use.
These values can be numeric, exact, or symbolic.
In[5]:= f .1
Out[5]= 0.909091
In[6]:= f 1
Out[6]=
1
2
In[7]:= f
2
Out[7]=
1
1
2
38 An Introduction to Programming with Mathematica
We clear the symbols that are no longer needed.
In[8]:= Clear a, f
Immediate vs. delayed assignments
When you make an assignment to a variable, you are only interested in giving that variable
a specific value and then using the variable name to represent that value in subsequent
computations. But oftentimes, when you set up definitions for functions, those functions
may depend upon the values of previously defined functions or constants. In such instances
it us useful to delay the assignment until the function is actually used in a computation.
This is the basic difference between immediate and delayed assignments.
An immediate assignment is written Set[lhs,rhs] or, more commonly:
lhs = rhs
where lhs is an abbreviation for left-hand side and rhs abbreviates right-hand side.
As an example, consider defining rand1 to be an immediate assignment that gener-
ates a uniformly distributed random number between 0 and 1.
In[9]:= rand1 Random
Out[9]= 0.668693
Notice that the output of this assignment is the value of the right-hand side and that
Mathematica evaluates the right-hand side immediately; that is, when the assignment is made.
A delayed assignment is set up with the SetDelayed function and is written Set
Delayed[lhs,rhs] or, in its standard input form:
lhs := rhs
As an example, consider rand2 to be defined similarly to rand1, but with a delayed
assignment.
In[10]:= rand2 : Random
Notice that the delayed assignment does not return a value when the assignment is
made. In fact, the right-hand side will not be evaluated until the function rand2 is called.
2 The Mathematica language 39
Let us call the function rand1 five times.
In[11]:= Table rand1, 5
Out[11]= 0.668693, 0.668693, 0.668693, 0.668693, 0.668693
Because the right-hand side of rand1 was evaluated when the definition was made, rand1
was assigned the value 0.668693. Each subsequent call to rand1 returns that value.
In[12]:= ?rand1
Global`rand1
rand1 0.668693
On the other hand, creating a table of values using rand2 produces a very different result.
In[13]:= Table rand2, 5
Out[13]= 0.8312, 0.781807, 0.124634, 0.934537, 0.600252
Each of the five times that rand2 is called in the Table, Mathematica looks up the
definition of rand2 (which does not have a numeric value), and sees that it should evaluate
Random[]. It does this each time it is called, generating a new random number each time.
In[14]:= ?rand2
Global`rand2
rand2 : Random
When a SetDelayed function is entered, nothing is returned. When a Set func-
tion is entered, the value resulting from evaluating the right-hand side is returned. This
difference in output is indicative of a more fundamental difference in what happens when
the two kinds of functions are entered and rewrite rules are thereby created. To see this,
we need to look at the global rule base, wherein reside rewrite rules.
The global rule base
The global rule base is composed of two kinds of rewrite rules: the built-in functions,
which are part of every Mathematica session, and the user-defined rewrite rules, which are
entered during a particular session.
We can get information about both kinds of rules in the global rule base by entering
?name. In the case of a built-in function, the resulting usage message gives information
about the syntax for using the function and a brief statement explaining what the function
does.
40 An Introduction to Programming with Mathematica
In[15]:= ? Apply
Apply f, expr or f expr replaces the head
of expr by f. Apply f, expr, levelspec replaces
heads in parts of expr specified by levelspec. More
In the case of a user-defined rewrite rule, the rule itself is printed. For the simple
examples above, the crucial difference between rewrite rules created with the SetDe
layed and Set functions becomes apparent by querying the rule base for the rewrite
rules associated with the symbols rand1 and rand2.
In[16]:= ?rand1
Global`rand1
rand1 0.668693
A rewrite rule created using the Set function has the same left-hand side as the
function that created it but the right-hand side of the rule may differ from the right-hand
side of the function. This is because the right-hand side of the rule was evaluated at the
moment the rule was first evaluated.
In[17]:= ?rand2
Global`rand2
rand2 : Random
Comparing this with the original SetDelayed function, we see that a rewrite rule
created using the SetDelayed function looks exactly like the function that created it.
This is because both the left-hand side and right-hand side of a SetDelayed function are
placed in the rule base without being evaluated.
In view of this difference between the SetDelayed and Set functions, the question
is when should you use one or the other function to create a rewrite rule?
When you define a function, you usually do not want either the left-hand side or the
right-hand side to be evaluated; you just want to make it available for use when the appro-
priate function call is made. This is precisely what occurs when a SetDelayed function is
entered, so the SetDelayed function is commonly used in writing function definitions.
When you make a value declaration, you do not want the left-hand side to be evalu-
ated; you just want to make it a nickname to serve as a shorthand for a value. This is what
happens when a Set function is entered and so the Set function is commonly used to
make value declarations, such as assigning a numeric value to a constant or variable.
A new rewrite rule overwrites, or replaces, an older rule with the same left-hand side.
However, keep in mind that if two left-hand sides are the same except for the names of
2 The Mathematica language 41
their pattern variables, they are considered different by Mathematica. Clear[name] can be
used to remove a rewrite rule from the global rule base.
Piecewise-defined functions
You can set up several definitions for a function and Mathematica will apply the definition
that applies. In the following example we give a piecewise-defined function g, whose values
depend upon whether x is less than 0, between 0 and 1, or greater than 1. We specify the
conditions on x by means of the /; symbol.
In[18]:= g x_ : x
3
; x 0
In[19]:= g x_ : x ; 0 x 1
In[20]:= g x_ : Sin x ; x 1
In[21]:= Plot g x , x, 2, 3 ;
2 1 1 2 3
1.5
1
0.5
0.5
1
Defining the function above is more easily accomplished using the new (in Version
5.1) Piecewise function as follows.
In[22]:= Piecewise x
3
, x 0 , x, 0 x 1 , Sin x , x 1
Out[22]=
x
3
x 0
x 0 x 1
Sin x x 1
You could plot this expression directly or define a function with this Piecewise
object on the right-hand side of your definition and then use that function like any other.
We will look at further uses of piecewise-defined objects in later chapters, in particular in
the chapter on procedural programming.
42 An Introduction to Programming with Mathematica
Functions with multiple definitions
When you make an assignment, the symbol associated with the evaluation rule is called an
assignment tag. Assignment tags are used to specify the structure of expressions. So, for
example, the expression {a,b,c} is represented internally by List[a,b,c]. Its assign-
ment tag is List. List does not really do anything except serve as a wrapper to specify
the structure of this expression. Similarly, the expression 1+2 is represented internally by
Plus[1,2]; its assignment tag is Plus.
Occasionally you will encounter the Tag expression when you try to evaluate some
incorrect input.
In[23]:= 1 2 4
Set::write : Tag Plus in 1 2 is Protected. More
Out[23]= 4
For user-defined functions, the tag basically refers to the name of the function. So,
for example, the following assignment associates the rule 1 x x
2
with the tag f.
In[24]:= f x_ : 1 x x
2
There can be many evaluation rules associated with one tag. The following assign-
ments all associate rules with the symbol f.
In[25]:= f x_, y_ : x y
In[26]:= f x_, y_, z_ :
1
x y z
To view all of the rules associated with f, use ?f.
In[27]:= ?f
Global`f
f x_ : 1 x x
2
f x_, y_ : x y
f x_, y_, z_ :
1
x y z
The advantages of this structure is that you can use one name for a function that will
behave differently depending upon the number or form of arguments you give to that
function. Using a different symbol for each of these tasks would require you and those who
use your programs to have to remember multiple function names when one might be
sufficient.
2 The Mathematica language 43
Let us clear symbols that are no longer needed.
In[28]:= Clear f, g
Exercises
1. What rewrite rules do each of the following functions create? Check your predictions
by entering them and then querying the rule base.
a. randLis1[n_]:= Table[Random[], {n}]
b. randLis2[n_]:= (x=Random[]; Table[x, {n}])
c. randLis3[n_]:= (x:=Random[]; Table[x, {n}])
d. randLis4[n_] = Table[Random[], {n}]
2. Consider the two functions f and g, which are identical except that one is written
using an immediate assignment and the other using a delayed assignment.
In[1]:= f n_ Sum 1 x
j
, j, 1, n
Out[1]=
1 x 1 1 x
n
x
In[2]:= g n_ : Sum 1 x
j
, j, 1, n
Explain why the output of these two functions look so different. Are they in fact
different?
In[3]:= f 2
Out[3]=
1 x 1 1 x
2
x
In[4]:= g 2
Out[4]= 1 x 1 x
2
3. Create a piecewise-defined function g x based on the following and then plot the
function from 2 to 0.
g x
1 x 2
2
2 x 1
1 x
2
x 0
44 An Introduction to Programming with Mathematica
2.3 Predicates and Boolean operations
Predicates
When working with data sets, you are often presented with the problem of extracting those
data points that meet certain criteria. Similarly, when you write programs, oftentimes what
to do next at any particular point in your program will depend upon some test or condition
being met. Every programming language has constructs for testing data or conditions.
Some of the most useful such constructs are called predicates. A predicate is a function that
returns a value of true or false depending upon whether its argument passes a test. For
example, the predicate PrimeQ tests for the primality of its argument.
In[1]:= PrimeQ 144
Out[1]= False
Other predicates are available for testing numbers to see whether they are even, odd,
integral, and so on.
In[2]:= OddQ[21]
Out[2]= True
In[3]:= EvenQ[21]
Out[3]= False
In[4]:= IntegerQ[5/9]
Out[4]= False
The NumericQ predicate tests whether its argument is a numeric quantity. Essen-
tially, NumericQ[x] gives True whenever N[x] evaluates to an explicit number.
In[5]:= NumericQ
Out[5]= True
In[6]:= NumericQ
Out[6]= False
This is distinct from a related function, NumberQ, which evaluates to True whenever its
argument is an explicit number (that is, has head one of Integer, Rational, Real,
Complex).
2 The Mathematica language 45
In[7]:= NumberQ 3.2
Out[7]= True
In[8]:= NumberQ
Out[8]= False
Many other predicates are available for testing if an expression is an atom, a list, a
matrix, a polynomial, and much more.
In[9]:= AtomQ "string"
Out[9]= True
In[10]:= ListQ a, b, c
Out[10]= True
In[11]:= MatrixQ
1 0 0
0 1 0
0 0 1
Out[11]= True
In[12]:= PolynomialQ
1
x
1
x
2
1
x
3
, x
Out[12]= False
In[13]:= IntervalMemberQ Interval 3, 4 ,
Out[13]= True
Relational and logical operators
Another type of predicate that is commonly used in programming are relational operators.
These are used to compare two or more expressions and return a value of True or False.
The relational operators in Mathematica are Equal ( ), Unequal ( ), Greater (>),
Less (<), GreaterEqual( ), and LessEqual ( ). They can be used to compare num-
bers or arbitrary expressions.
In[14]:= 7 5
Out[14]= False
46 An Introduction to Programming with Mathematica
In[15]:= Equal 3, 7 4,
6
2
Out[15]= True
In[16]:= x
2
1
x
4
1
x
2
1
Simplify
Out[16]= True
Note that the relational operators have lower precedence than arithmetic operators.
The second example above is interpreted as 3 (7-4) and not as (3 7)-4. Table 2.1
lists the relational operators and their various input forms.
StandardForm Functional form Meaning
x y Equal x, y equal
x y Unequal x, y unequal
x y Greater x, y greater than
x y Less x, y less than
x y GreaterEqual x, y greater thanor equal
x y LessEqual x, y less thanor equal
Table 2.1: Relational operators
The logical operators (sometimes known as Boolean operators) determine the truth
of an expression based on Boolean arithmetic. For example, the conjunction of two true
statements is always true.
In[17]:= 4 5 && 8 1
Out[17]= True
The Boolean operation and is represented in Mathematica by And, with shorthand
notation && or . Here is a table that gives all the possible values for the And operator.
(The function TruthTable is developed in Chapter 10.)
In[18]:= TruthTable A B, A, B
Out[18]//DisplayForm=
A B A B
T T T
T F F
F T F
F F F
2 The Mathematica language 47
The logical or operator, represented by Or and with shorthand notation || (or ),
is true when either of its arguments is true.
In[19]:= 4 3 3
6
2
Out[19]= True
In[20]:= 0 0.0001
22
7
Out[20]= False
Note the difference between this Boolean or and the common notion of or. A
phrase such as, It is cold or it is hot, uses the word or in an exclusive sense; that is, it
excludes the possibility that it is both cold and hot. The logical Or is inclusive in the sense
that if A and B are both true, then A||B is also true.
In[21]:= True True
Out[21]= True
Mathematica also contains an operator for the exclusive or, Xor.
In[22]:= Xor True, True
Out[22]= False
In[23]:= Xor True, False
Out[23]= True
Table 2.2 shows the logical operators and their input forms.
StandardForm Functional form Meaning
x Not x not
x y Unequal x, y unequal
x && y And x, y and
x y Or x, y or
x y && x && y Xor x, y exclusive or
Table 2.2: Logical operators
Introduced in Version 4 of Mathematica are the bitwise logical operators. These func-
tions operate on integers as binary bits. For example, BitOr[x,y] gives the integer whose
binary representation has 1s wherever the binary representation of x or y has 1s. Here is
the bitwise OR of 21 and 19, given in binary form.
48 An Introduction to Programming with Mathematica
In[24]:= BaseForm BitOr 2^^10101, 2^^10011 , 2
Out[24]//BaseForm=
10111
2
Similarly, BitXor[x,y] gives the integer with 1s at positions where either x or y
have 1s, but not both.
In[25]:= BaseForm BitXor 2^^10101, 2^^10011 , 2
Out[25]//BaseForm=
110
2
Functional form Meaning
BitAnd x, y bitwise ANDof x andy
BitOr x, y bitwise ORof x and y
BitNot x bitwise NOTof x
BitXor x, y bitwise XORof x andy
Table 2.3: Bitwise operators
In Chapter 4 we will look at an application of bitwise operators to an example
involving error-correcting codes: the computation of Hamming distance.
Exercises
1. Create a predicate function that returns a value of True if its argument is between
1 and 1.
2. Write a predicate function NaturalQ[n] that returns a value of True if n is a
natural number and False otherwise; that is, NaturalQ[n] is True if n is among
0, 1, 2, 3, .
3. Create a predicate function SubsetQ[lis
1
, lis
2
] that returns a value of True if lis
1
is
a subset of lis
2
. Remember: the empty set {}, is a subset of every set.
2 The Mathematica language 49
2.4 Attributes
All functions in Mathematica have certain properties, called attributes. These attributes can
make a function commutative or associative, or they may give the function the ability to be
threaded over a list. The attributes of any function are displayed with the Attributes
function.
In[1]:= Attributes Plus
Out[1]= Flat, Listable, NumericFunction,
OneIdentity, Orderless, Protected
The Flat attribute indicates that this function (Plus) is associative. That is, given
three elements to add, it does not matter which two are added first. In mathematics, this is
known as associativity and is written as a b c a b c. In Mathematica this could be
indicated by saying that the two expressions Plus[a, Plus[b, c]] and Plus[
Plus[a, b], c] are equivalent to the flattened form Plus[a, b, c]. When Mathe-
matica knows that a function has the attribute Flat, it writes it in flattened form.
In[2]:= Plus Plus a, b , c
Out[2]= a b c
The Orderless attribute indicates that the function is commutative; that is,
a b b a. This allows Mathematica to write such an expression in an order that is useful
for computation. It does this by sorting the elements into a canonical order. For expressions
consisting of letters and words, this ordering is alphabetic.
In[3]:= t h i n
Out[3]= h i n t
Sometimes a canonical order is readily apparent.
In[4]:= x
3
x
5
x
4
x
2
1 x
Out[4]= 1 x x
2
x
3
x
4
x
5
Other times, it is not so apparent.
In[5]:= x
3
y
2
y
7
x
5
y x
4
y
9
x
2
1 x
Out[5]= 1 x x
4
y x
3
y
2
x
5
y
7
x
2
y
9
When a symbol has the attribute Protected, the user is prevented from modifying
the function in any significant way. All built-in operations have this attribute.
Functions with the attribute OneIdentity have the property that repeated applica-
tion of the function to the same argument will have no effect. For example, the expression
50 An Introduction to Programming with Mathematica
Plus[Plus[a, b]] is equivalent to Plus[a, b], hence only one addition is
performed.
In[6]:= FullForm Plus Plus a b
Out[6]//FullForm=
Plus a, b
The other attributes for the Plus function, (Listable and NumericFunction)
will be discussed in later chapters. Consult the manual (Wolfram 2003) for a complete list
of the Attributes that symbols can have.
Although it is unusual to want to alter the attributes of a built-in function, it is fairly
common to change the default attributes of a user-defined function. For example, suppose
you had a function which you wanted to inherit the Orderless attribute. Without
explicitly setting that attribute, the function does not reorder its arguments.
In[7]:= f x, a, m
Out[7]= f x, a, m
The SetAttributes function is used to change the attributes of a function. Explicitly
setting f to have the Orderless attribute causes its arguments to be automatically sorted.
In[8]:= SetAttributes f, Orderless
In[9]:= f x, a, m
Out[9]= f a, m, x
We will see a practical use of SetAttributes in Section 5.3.
2 The Mathematica language 51
3 Lists
The list is the fundamental data structure used in Mathematica to group objects
together. A very extensive set of built-in functions is provided by Mathematica to
manipulate lists in a variety of ways, ranging from simple operations, such as moving
list elements around, to more sophisticated operations, such as applying a function to a
list. We also discuss working with strings, as their structure and manipulation is so
similar to lists.
3.1 Introduction
Many computations involve working with a collection of objects. For example, abstract
mathematics deals with operations on arbitrary sets, represented notationally, but also
conceptually, as lists.
In[1]:= a, b, c c, d, e
Out[1]= a, b, c, d, e
In[2]:= a, b, c c, d, e
Out[2]= c
Data, in Mathematica, is represented using lists. A large collection of functions is
available for manipulating and analyzing lists of data. For example, you can sort any set of
data.
In[3]:= Sort 4, 16, 1, 77, 23
Out[3]= 1, 4, 16, 23, 77
You can extract elements of a dataset based on some criteria. Here we select those
numbers from a list that are greater than 0.
In[4]:= Select 4.9239, 1.24441, 0.80388, 3.27761 , Positive
Out[4]= 4.9239, 3.27761
Working with such collections of objects requires that the objects (also called data
objects) be gathered together in some way. There are a variety of structures that can be used
to store data objects in a computer. The most often used data structure in Mathematica is
the list. This is created using the built-in List function which has the standard input form
of a sequence of arguments separated by commas and enclosed in braces.
arg
1
, arg
2
, , arg
n
Lists are used throughout Mathematica, not only to represent a collection of data
elements, but also to delineate a range of values for some variable or iterator. For example,
the second argument to the Table function is a list that specifies the iterator variable and
the values that it should range over.
In[5]:= Table i
2
, i, 1, 5
Out[5]= 1, 4, 9, 16, 25
Similarly, the plotting functions use lists to specify the range over which a variable
should be evaluated.
In[6]:= Plot Sin x , x, 0, 2 ;
1 2 3 4 5 6
1
0.5
0.5
1
Internally, lists are stored in the functional form using the List function with some
arbitrary number of arguments.
List arg
1
, arg
2
, , arg
n
For example, using FullForm we can view the internal representation of the list
{a,b,c}.
In[7]:= FullForm a, b, c
Out[7]//FullForm=
List a, b, c
54 An Introduction to Programming with Mathematica
The arguments of the List function (the list elements) can be any type of expression,
including numbers, symbols, functions, character strings, and even other lists.
In[8]:= 2.4, f, Sin, "ossifrage", 5, 3 , ,
Out[8]= 2.4, f, Sin, ossifrage, 5, 3 , ,
Elements in lists can be rearranged, sorted, removed, new elements added, and
operations performed on select elements or on the list as a whole. In fact, lists are such
general objects in Mathematica that they can be used to represent a vast array of objects.
In this chapter, we will demonstrate the use of built-in Mathematica functions to
manipulate lists in various ways. In cases where the operation of a function is relatively
straightforward, we will simply demonstrate its use without explanation (the on-line Help
system and the The Mathematica Book (Wolfram 2003) should be consulted for more
detailed explanations of all of the built-in functions). The underlying message here is that
almost anything you might wish to do to a list can be accomplished using built-in func-
tions. It is important to have as firm a handle on these functions as possible, since a key to
good, efficient programming in Mathematica is to use the built-in functions whenever
possible to manipulate list structures.
3.2 Creating and measuring lists
List construction
In addition to using the List function to collect data objects, you can also generate lists
from scratch by creating the objects and then placing them in a list.
Range[imin,imax, di] generates a list of ordered numbers starting from imin and
going up to, but not exceeding, imax in increments of di.
In[1]:= Range 4, 7, 3
Out[1]= 4, 1, 2, 5
If di is not specified, a value of one is used.
In[2]:= Range 4, 8
Out[2]= 4, 5, 6, 7, 8
3 Lists 55
If neither imin nor di is specified, then both are given the value of 1.
In[3]:= Range 4
Out[3]= 1, 2, 3, 4
It is not necessary for imin, imax, or di to be integers.
In[4]:= Range 1.5, 6.3, .75
Out[4]= 1.5, 2.25, 3., 3.75, 4.5, 5.25, 6.
Table[expr,{i,imin,imax,di}] generates a list by evaluating expr a number of
times.
In[5]:= Table 3 k, k, 1, 10, 2
Out[5]= 3, 9, 15, 21, 27
The first argument, 3k in the above example, is the expression that is evaluated to produce
the elements in the list. The second argument to the Table function, {i,imin,imax,di},
is called the iterator. It is a list that specifies the number of times the expression is evaluated
and hence the number of elements in the list. The iterator variable may or may not appear
in the expression being evaluated. The value imin is the value of i used in the expression to
create the first list element. The value di is the incremental increase in the value of i used in
the expression to create additional list elements. The value imax is the maximum value of i
used in the expression to create the last list element (if incrementing i by di gives a value
greater than imax, that value is not used).
In[6]:= Table i, i, 1.5, 6.3, .75
Out[6]= 1.5, 2.25, 3., 3.75, 4.5, 5.25, 6.
Table[i,{i,imin,imax,di}] is equivalent to Range[imin,imax,di]. As with the
Range function, the arguments to Table can be simplified when the iterator increment is
one.
In[7]:= Table 3 i, i, 2, 5
Out[7]= 6, 9, 12, 15
Similarly, both imin and di can be omitted and are then assumed to be 1.
In[8]:= Table i
2
, i, 4
Out[8]= 1, 4, 9, 16
56 An Introduction to Programming with Mathematica
If the iterator variable does not appear in the expression being evaluated, it may be omitted
as well. The expression will then simply be evaluated that many times.
In[9]:= Table Random , 3
Out[9]= 0.155408, 0.0408563, 0.62081
The expression that the Table function evaluates can be completely arbitrary. In the
following computation, it is used to create tables of formulas.
In[10]:= Table Expand 1
i
, i, 1, 3
Out[10]= 1 , 1 2
2
, 1 3 3
2 3
Table can be used to create a nested list; that is, a list containing other lists as ele-
ments. This can be done by using additional iterators.
In[11]:= Table i j, j, 1, 4 , i, 1, 3
Out[11]= 2, 3, 4 , 3, 4, 5 , 4, 5, 6 , 5, 6, 7
When there is more than one iterator, their order of appearance is important, because the
value of the outer iterator is varied for each value of the inner iterator In the above exam-
ple, for each value of j (the inner iterator), i was varied from 1 to 3, producing a three-ele-
ment list for each of the four values of j. If you reverse the iterator order, you will get an
entirely different list.
In[12]:= Table i j, i, 1, 3 , j, 1, 4
Out[12]= 2, 3, 4, 5 , 3, 4, 5, 6 , 4, 5, 6, 7
You will often find it useful to display nested lists in a matrix or tabular form.
In[13]:= Table i j, i, 1, 4 , j, 1, 3 TableForm
Out[13]//TableForm=
2 3 4
3 4 5
4 5 6
5 6 7
In[14]:= Table i j, i, 1, 4 , j, 1, 3 MatrixForm
Out[14]//MatrixForm=
2 3 4
3 4 5
4 5 6
5 6 7
3 Lists 57
The value of the outer iterator may depend on the value of the inner iterator, which can
result in a nonrectangular list.
In[15]:= Table i j, i, 1, 3 , j, 1, i
Out[15]= 2 , 3, 4 , 4, 5, 6
In[16]:= TableForm %
Out[16]//TableForm=
2
3 4
4 5 6
However, the inner iterator may not depend on the outer iterator because, as we have seen,
the inner iterator is fixed as the outer one varies.
In[17]:= Table i j, i, 1, j , j, 1, 3
Table::iterb :
Iterator i, 1, j does not have appropriate bounds. More
Out[17]= Table i j, i, 1, j , j, 1, 3
Measuring lists
Recall from Chapter 2 that Length[expr] is used to give the number of elements in expr.
For a simple unnested (linear) list, the Length function tells us how many elements are in
the list.
In[18]:= Length a, b, c, d, e, f
Out[18]= 6
In a nested list, each inner list is an element of the outer list. Therefore, the Length
of a nested list indicates the number of inner lists, and not their sizes.
In[19]:= Length 1, 2 , 3, 4 , 5, 6 , a, b , c, d , e, f
Out[19]= 2
To find out more about the inner lists, use the Dimensions function.
In[20]:= Dimensions 1, 2 , 3, 4 , 5, 6 , a, b , c, d , e, f
Out[20]= 2, 3, 2
This indicates that there are two inner lists, that each inner list contains three lists, and
that the innermost lists each have two elements. MatrixForm may help to see the struc-
ture better.
58 An Introduction to Programming with Mathematica
In[21]:= MatrixForm 1, 2 , 3, 4 , 5, 6 , a, b , c, d , e, f
Out[21]//MatrixForm=
1
2
3
4
5
6
a
b
c
d
e
f
The number of dimensions of a (possibly nested) list, is given by ArrayDepth.
In[22]:= ArrayDepth 1, 2 , 3, 4 , 5, 6 , a, b , c, d , e, f
Out[22]= 3
This is identical to the number of levels in that expression, as displayed by TreeForm.
In[23]:= TreeForm a, b, c
Out[23]//TreeForm=
List a,
List b,
List c
Exercises
1. Generate the list {{0},{0,2},{0,2,4},{0,2,4,6},{0,2,4,6,8}} in two
different ways using the Table function.
2. A table containing ten random 1s and 0s can easily be created using Table[
Random[Integer],{10}]. Create a ten-element list of random 1s, 0s and 1s.
3. Create a ten-element list of random 1s and 1s. This table can be viewed as a list of
the steps taken in a random walk along the x-axis, where a step can be taken in either
the positive x direction (corresponding to 1) or the negative x direction
(corresponding to 1) with equal likelihood.
The random walk in one, two, three (and even higher) dimensions is used in science
and engineering to represent phenomena that are probabilistic in nature. We will use
a variety of random walk models throughout this book to illustrate specific program-
ming points.
4. From a mathematical point of view, a list can be viewed as a vector and a nested list
containing inner lists of equal length can be viewed as a matrix (or an array). Mathe-
matica has another built-in function Array which creates lists. We can use an
undefined function f to see how Array works.
3 Lists 59
In[1]:= Array f, 5
Out[1]= f 1 , f 2 , f 3 , f 4 , f 5
In[2]:= Array f, 3, 4
Out[2]= f 1, 1 , f 1, 2 , f 1, 3 , f 1, 4 ,
f 2, 1 , f 2, 2 , f 2, 3 , f 2, 4 ,
f 3, 1 , f 3, 2 , f 3, 3 , f 3, 4
Generate both of these lists using the Table function.
5. Predict the dimensions of the list {{{1,a},{4,d}},{{2,b},{3,c}}}. Use the
Dimensions function to check your answer.
3.3 Manipulating lists
Testing a list
The locations of specific elements in a list can be determined using the Position
function.
In[1]:= Position 5, 7, 5, 2, 1, 4 , 5
Out[1]= 1 , 3
This result indicates that the number 5 occurs in the first and third positions in the list.
The extra braces are used to avoid confusion with the case when elements are nested within
a list.
In[2]:= Position a, b, c , d, e, f , f
Out[2]= 2, 3
The expression f occurs once, in the third position within the second inner list.
There is also a function that picks out the elements in a list that return True when a
predicate is applied to them. For example, this finds all of the even numbers in a list.
In[3]:= Select 1, 4, 1, 5, 9, 2 , EvenQ
Out[3]= 4, 2
Other functions exist to select or count the number of elements in a list that match a
certain pattern. We will look at these in detail in Chapter 6.
60 An Introduction to Programming with Mathematica
Extracting elements
Elements can easily be extracted from a specific location in a list. For example, this extracts
the third element in the list vec.
In[4]:= vec 2, 3, 7, 8, 1, 4 ;
In[5]:= Part vec, 3
Out[5]= 7
The Part function can be abbreviated using a standard input form.
In[6]:= vec 3
Out[6]= 7
If you are interested in the elements from more than one location, you can extract
them using a list. For example, this picks out the second and fourth elements of vec.
In[7]:= vec 2, 4
Out[7]= 3, 8
For multi-dimensional lists, you have to specify both the sublist and the position of
the element in that sublist that you are interested in.
Here is a sample 3 3 matrix that we will work with.
In[8]:= mat Table a
i,j
, i, 3 , j, 3 MatrixForm
Out[8]//MatrixForm=
a
1,1
a
1,2
a
1,3
a
2,1
a
2,2
a
2,3
a
3,1
a
3,2
a
3,3
This picks out the first part of the second sublist.
In[9]:= mat 2, 1
Out[9]= a
2,1
For multi-dimensional lists, several options are available to extract subsections of the
list. A common operation involves extracting rows or columns from a matrix.
This extracts the entire second column of mat.
In[10]:= mat All, 2 MatrixForm
Out[10]//MatrixForm=
a
1,2
a
2,2
a
3,2
3 Lists 61
And here is the third row of this matrix.
In[11]:= mat 3, All
Out[11]= a
3,1
, a
3,2
, a
3,3
If you only specify one argument, the second is assumed to be All.
In[12]:= mat 3
Out[12]= a
3,1
, a
3,2
, a
3,3
In addition to being able to extract elements from specific locations in a list, you can
extract consecutively placed elements within the list. You can take elements from either the
front or the back of a list.
In[13]:= Take 1, 4, 1, 5, 9, 2 , 2
Out[13]= 1, 4
In[14]:= Take 1, 4, 1, 5, 9, 2 , 2
Out[14]= 9, 2
If you take consecutive elements from a list other than from the front and the back, you
need to remember that the numbering of positions is different front-to-back and
back-to-front.
In[15]:= Take 1, 4, 1, 5, 9, 2 , 2, 4
Out[15]= 4, 1, 5
In[16]:= Take 1, 4, 1, 5, 9, 2 , 5, 3
Out[16]= 4, 1, 5
You can mix both positive and negative indices.
In[17]:= Take 1, 4, 1, 5, 9, 2 , 5, 4
Out[17]= 4, 1, 5
You can also take elements in steps. This takes the first through sixth element in incre-
ments of 2; that is, it takes every other element.
In[18]:= Take 1, 4, 1, 5, 9, 2 , 1, 6, 2
Out[18]= 1, 1, 9
62 An Introduction to Programming with Mathematica
You can discard elements from a list, keeping the rest. Elements can be removed
from either end of the list or from consecutive locations.
In[19]:= Drop 1, 4, 1, 5, 9, 2 , 2
Out[19]= 1, 5, 9, 2
In[20]:= Drop 1, 4, 1, 5, 9, 2 , 1
Out[20]= 1, 4, 1, 5, 9
In[21]:= Drop 1, 4, 1, 5, 9, 2 , 3, 5
Out[21]= 1, 4, 2
You can remove elements at specific locations as well.
In[22]:= Delete 1, 4, 1, 5, 9, 2 , 1
Out[22]= 4, 1, 5, 9, 2
In[23]:= Delete 1, 4, 1, 5, 9, 2 , 3 , 4
Out[23]= 1, 4, 9, 2
Certain extractions are used so often that they are given their own functions.
In[24]:= First 1, 4, 1, 5, 9, 2
Out[24]= 1
In[25]:= Last 1, 4, 1, 5, 9, 2
Out[25]= 2
In[26]:= Rest 1, 4, 1, 5, 9, 2
Out[26]= 4, 1, 5, 9, 2
Rearranging lists
Every list can be sorted into a canonical order. For lists of numbers or letters, this ordering
is usually obvious.
In[27]:= Sort 3, 1.7, , 4,
22
7
Out[27]= 4, 1.7, 3,
22
7
,
Mathematica uses the following canonical orderings: numbers are ordered by numeri-
cal value, with complex numbers first ordered by real part and then by absolute value of
3 Lists 63
the imaginary part; symbols and strings are ordered alphabetically, powers and products
are ordered in a manner corresponding to the terms in a polynomial; expressions are
ordered depth-first with shorter expressions coming first.
You can also sort lists according to an ordering function that you can specify.
In[28]:= Sort 3, 1.7, , 4,
22
7
, Greater
Out[28]=
22
7
, , 3, 1.7, 4
When applied to a nested list, Sort will use the first element of each nested list to
determine the order.
In[29]:= Sort 2, c , 7, 9 , e, f, g , 1, 4.5 , x, y, z
Out[29]= 1, 4.5 , 2, c , 7, 9 , e, f, g , x, y, z
The order of the elements in a list can be reversed.
In[30]:= Reverse 1, 2, 3, 4, 5
Out[30]= 5, 4, 3, 2, 1
All of the elements can be rotated a specified number of positions to the right or the
left. By default RotateLeft (and RotateRight) shifts the list one position to the left
(right).
In[31]:= RotateLeft 1, 2, 3, 4, 5
Out[31]= 2, 3, 4, 5, 1
This rotates every element two positions to the right.
In[32]:= RotateRight 1, 2, 3, 4, 5 , 2
Out[32]= 4, 5, 1, 2, 3
Partition rearranges list elements to form a nested list. It may use all of the
elements and simply divvy up a list. Here we partition the list into nonoverlapping sublists
of length two.
In[33]:= Partition 1, 4, 1, 5, 9, 2 , 2
Out[33]= 1, 4 , 1, 5 , 9, 2
You might be interested in only using some of the elements from a list. For example,
this takes one-element sublists, with an offset of two; that is, every other one-element
sublist.
64 An Introduction to Programming with Mathematica
In[34]:= Partition 1, 4, 1, 5, 9, 2 , 1, 2
Out[34]= 1 , 1 , 9
You can also create overlapping inner lists, consisting of ordered pairs (two-element
sublists) whose second element is the first element of the next ordered pair.
In[35]:= Partition 1, 4, 1, 5, 9, 2 , 2, 1
Out[35]= 1, 4 , 4, 1 , 1, 5 , 5, 9 , 9, 2
The Transpose function pairs off the corresponding elements of the inner lists. Its
argument is a single list consisting of nested lists.
In[36]:= Transpose 1, 2, 3, 4 , a, b, c, d
Out[36]= 1, a , 2, b , 3, c , 4, d
In[37]:= Transpose 1, 2, 3, 4 , a, b, c, d , i, ii, iii, iv
Out[37]= 1, a, i , 2, b, ii , 3, c, iii , 4, d, iv
For rectangular lists, you might think of Transpose as exchanging the rows and
columns of the corresponding matrix.
Elements can be added to the front, the back, or to any specified position in a given
list.
In[38]:= Append 1, 2, 3, 4 , 5
Out[38]= 1, 2, 3, 4, 5
In[39]:= Prepend 1, 2, 3, 4 , 5
Out[39]= 5, 1, 2, 3, 4
In[40]:= Insert 1, 2, 3, 4 , 5, 3
Out[40]= 1, 2, 5, 3, 4
Elements at specific locations in a list can be replaced with other elements. Here, 5
replaces the element in the second position of the list.
In[41]:= ReplacePart a, b, c, d, e , 5, 2
Out[41]= a, 5, c, d, e
You can flatten a nested list to various extents. You can remove all of the inner
braces, creating a linear list of elements.
In[42]:= Flatten 3, 1 , 2, 4 , 5, 3 , 7, 4
Out[42]= 3, 1, 2, 4, 5, 3, 7, 4
3 Lists 65
You can limit the degree of flattening, removing only some of the inner lists. For
example, two inner lists, each having two ordered pairs, can be turned into a single list of
four ordered pairs by only flattening down one level deep.
In[43]:= Flatten 3, 1 , 2, 4 , 5, 3 , 7, 4 , 1
Out[43]= 3, 1 , 2, 4 , 5, 3 , 7, 4
List component assignment
The capability to alter elements of lists merits detailed consideration. The general syntax
for modifying a list is:
name[[integer-valued-expression]] = expr
The name must be the name of a list. The integer-valued-expression must evaluate to a
legal subscript, that is a number whose absolute value is less than or equal to the length of
the list. The assignment returns the value of expr (as assignments always do), but has the
effect of changing the list to which name is bound.
Here is a list with five elements.
In[44]:= L 0, 1, 2, 3, 4
Out[44]= 0, 1, 2, 3, 4
This replaces the value of the first element of L with the value 10.
In[45]:= L 1 10
Out[45]= 10
We see now that L has changed.
In[46]:= L
Out[46]= 10, 1, 2, 3, 4
Components of nested lists can be modified as well.
name expr
1
, expr
2
]] = expr
expr
1
and expr
2
are expressions that must evaluate to integers. expr
1
chooses the sublist of
name, and expr
2
the element of that sublist.
66 An Introduction to Programming with Mathematica
Here is a 2 3 nested list.
In[47]:= A 1, 2, 3 , 4, 5, 6
Out[47]= 1, 2, 3 , 4, 5, 6
This assigns the third element in the second sublist the value 20.
In[48]:= A 2, 3 20
Out[48]= 20
In[49]:= A
Out[49]= 1, 2, 3 , 4, 5, 20
However, note that assigning one array name to another one makes a copy of the
first. In this way, component assignments to either one will not affect the other.
In[50]:= B A
Out[50]= 1, 2, 3 , 4, 5, 20
In[51]:= B 1, 2 30
Out[51]= 30
In[52]:= B
Out[52]= 1, 30, 3 , 4, 5, 20
In[53]:= A
Out[53]= 1, 2, 3 , 4, 5, 20
In[54]:= A 2, 1 40
Out[54]= 40
In[55]:= B
Out[55]= 1, 30, 3 , 4, 5, 20
This behavior is in distinction to languages such as C where aliasing can allow one
list to point to another; with pointers, changing one array will have an affect on any array
that points to it.
3 Lists 67
Exercises
1. Predict where the 9s are located in the following list.
2, 1, 10 , 9, 5, 7 , 2, 10, 4 , 10, 1, 9 , 6, 1, 6
Confirm your prediction using Position.
2. Given a list of x, y data points
x
1
, y
1
, x
2
, y
2
, x
3
, y
3
, x
4
, y
4
, x
5
, y
5
separate the x and y components to get:
x
1
, x
2
, x
3
, x
4
, x
5
, y
1
, y
2
, y
3
, y
4
, y
5
3. Consider a two-dimensional random walk on a square lattice. (A square lattice can be
envisioned as a two-dimensional grid, just like the lines on graph paper.) Each step
can be in one of four directions: 1, 0 , 0, 1 , 1, 0 , 0, 1 , corresponding to
steps in the east, north, west and south directions, respectively. Use the list
{{1,0},{0,1},{-1,0},{0,-1}} to create a list of the steps of a ten-step
random walk.
4. In three steps, make a list of the elements in even-numbered locations in the list
{a,b,c,d,e,f,g}.
5. Suppose you are given a list S of length n, and a list P containing n different numbers
between 1 and n (that is, P is a permutation of Range[n]). Compute the list T such
that for all k between 1 and n, T[[k]]=S[[P[[k]]]]. For example, if
S={a,b,c,d} and P={3,2,4,1}, then T={c,b,d,a}.
6. Given the lists S and P in the previous exercise, compute the list U such that for all k
between 1 and n, U[[P[[k]]]] = S[[k]] (that is, S[[i]] takes the value from
position P[[i]] in U). Thus, for S={a,b,c,d} and P={3,2,4,1},
U={d,b,a,c}. Think of it as moving S[[1]] to position P[[1]], S[[2]] to
position P[[2]], and so on. Hint: Start by pairing the elements of P with the
elements of S.
68 An Introduction to Programming with Mathematica
3.4 Working with several lists
A number of the functions described earlier in this chapter, such as Transpose, work
with several lists if they are inside a nested list structure. We can also work directly with
multiple lists.
Join concatenates two lists.
In[1]:= Join 2, 5, 7, 3 , d, a, e, j
Out[1]= 2, 5, 7, 3, d, a, e, j
Here is the union of these two lists.
In[2]:= 4, 1, 2 5, 1, 2
Out[2]= 1, 2, 4, 5
In[3]:= Union 4, 1, 2 , 5, 1, 2
Out[3]= 1, 2, 4, 5
When the Union function is used either on a single list or a number of lists, a list is
formed consisting of the original elements in canonical order with all duplicate elements
removed. The Complement function gives all those elements in the first list that are not
in the other list or lists. Intersection[list
1
, list
2
,] finds all those elements common
to the list
i
. Complement and Intersection also remove duplicates and sort the ele-
ments that remain.
In[4]:= 4, 1, 2 5, 1, 2
Out[4]= 1, 2
In[5]:= Complement 4, 1, 2 , 5, 1, 2
Out[5]= 4
These last three functions, Union, Complement, and Intersection, treat lists
somewhat like sets in that there are no duplicates and the order of elements in the lists is
not respected.
Exercises
1. How would you perform the same task as Prepend[{x,y},z] using the Join
function?
2. Starting with the lists {1,2,3,4} and {a,b,c,d}, create the list {2,4,b,d}.
3 Lists 69
3. Given two lists, find all those elements that are not common to the two lists. For
example, starting with the lists, {a,b,c,d} and {a,b,e,f}, your answer would
return the list {c,d,e,f}.
3.5 Strings and characters
Characters are the objects that appear on the computer screen like a, 3, or !.
Uppercase and lowercase letters, numbers, punctuation marks, and spaces form the basic
set of characters. A sequence of characters enclosed in double quotes is called a string.
In[1]:= Head "The magic words are squeamish ossifrage."
Out[1]= String
When Mathematica prints out a string, it appears without the quotes.
In[2]:= "The magic words are squeamish ossifrage."
Out[2]= The magic words are squeamish ossifrage.
You can use the InputForm function to see these quotes.
In[3]:= InputForm "The magic words are squeamish ossifrage."
Out[3]//InputForm=
"The magic words are squeamish ossifrage."
A string is a value and, like other values (such as numbers and lists), there are built-in
functions available to manipulate strings, similar to those for lists. Their operations are
indicated by their names.
In[4]:= StringLength "The magic words are squeamish ossifrage."
Out[4]= 40
In[5]:= StringReverse "abcde"
Out[5]= edcba
In[6]:= StringTake "abcde", 3
Out[6]= abc
In[7]:= StringDrop "abcde", 1
Out[7]= abcd
70 An Introduction to Programming with Mathematica
In[8]:= StringPosition "abcde", "bc"
Out[8]= 2, 3
In[9]:= StringInsert "abcde", "t", 3
Out[9]= abtcde
In[10]:= StringReplace "abcde", "cd" "uv"
Out[10]= abuve
New in Version 5.1, you can use regular expressions in the functions you use to
manipulate strings.
In[11]:= StringMatchQ "all in good time", RegularExpression "a. "
Out[11]= True
In[12]:= StringCases "abc1, abd2, bcd3", RegularExpression "a. ?\\d"
Out[12]= abc1, abd2
In addition to using built-in functions to manipulate a string, you can convert a
string to a list of characters with the built-in Characters function.
In[13]:= Characters "abcde"
Out[13]= a, b, c, d, e
You can then use the list manipulating functions to alter the list or extract elements
from the list.
In[14]:= Take %, 2, 3
Out[14]= b, c
Finally, you can change the resulting list back into a string using the built-in String
Join function.
In[15]:= StringJoin %
Out[15]= bc
Another way to manipulate a string is to convert it to a list of character codes and
then operate on the codes using mathematical functions. Each character in a computers
character set is assigned a number, called its character code. Moreover, by general agree-
ment, almost all computers use the same character codes, called the ASCII codes. In this
code, the uppercase letters A, B, , Z are assigned the numbers 65, 66, , 90 while the
lowercase letters a, b, , z have the numbers 97, 98, , 122 (note that the number of an
uppercase letter is 32 less than its lowercase version). The numbers 0, 1, , 9 are coded as
3 Lists 71
48, 49, , 57 while the punctuation marks period, comma, and exclamation point have the
codes 46, 44, and 33, respectively. The space character is represented by the code 32.
Table 3.1 shows the characters and their codes.
Characters ASCII codes
A, B, , Z 65, 66, , 90
a, b, , z 97, 98, , 122
0, 1, , 9 48, 49, , 57
. period 46
, comma 44
exclamation 33
space 32
Table 3.1: ASCII character codes
Using the character code representation of characters, the following series of compu-
tations changes a word from lowercase to uppercase.
In[16]:= ToCharacterCode "darwin"
Out[16]= 100, 97, 114, 119, 105, 110
In[17]:= % 32
Out[17]= 68, 65, 82, 87, 73, 78
In[18]:= FromCharacterCode %
Out[18]= DARWIN
This can be accomplished more succinctly using StringReplace.
In[19]:= StringReplace "darwin", x_ ToUpperCase x
Out[19]= DARWIN
Or simply:
In[20]:= ToUpperCase "darwin"
Out[20]= DARWIN
72 An Introduction to Programming with Mathematica
Exercises
1. Convert the first character in a string (which you may assume to be a lowercase
letter) to uppercase.
2. Given a string containing two digits, convert it to its integer value; so the string
"73" produces the number 73.
3. Given a string containing two digits, convert it to its value as an integer in base 8; for
example, the string "73" will produce the number 59.
4. Given a string of digits of arbitrary length, convert it to its integer value. (Hint: You
may find that the Dot function is helpful.)
5. Create a Boolean function OrderedWordQ that returns True or False depending
upon whether its argument is in alphabetic order. So OrderedWordQ["best"]
would return True but OrderedWordQ["brag"] would return False. Then find
all those words in the file dictionary.dat that are ordered according to this
function.
Here is a platform-independent path to the dictionary file.
In[1]:= wordfile ToFileName $InstallationDirectory, "Documentation",
"English", "Demos", "DataFiles" , "dictionary.dat"
Out[1]= C:\Program Files\Wolfram Research\Mathematica\5.1\
Documentation\English\Demos\DataFiles\dictionary.dat
This reads in the file using ReadList, specifying the type of data being read in as a
Word.
In[2]:= words ReadList wordfile, Word ;
6. Create a function PalindromeQ[str] that returns a value of True if its argument
str is a palindrome; that is, if the string str is the same forward and backward. For
example, refer is a palindrome.
3 Lists 73
4 Functional programming
Programming in Mathematica is essentially a matter of writing user-defined functions
that work like mathematical functions; when applied to specific values, they perform
computations producing results. In fact, these functions can operate on arbitrary
expressions, including other functions. This functional style of programming distin-
guishes Mathematica from more traditional procedural languages like C and Fortran,
and a facility at functional programming is essential for taking full advantage of
Mathematicas powerful language to solve your computational tasks.
4.1 Introduction
Functions are objects that operate on expressions and output unique expressions for each
input. We can think of functions as mathematicians do. For example, here is a definition
for a function of two variables.
In[1]:= f x_, y_ : Cos x Sin y
You can evaluate the function for numeric or symbolic values.
In[2]:= f , 1.6
Out[2]= 0.000426397
In[3]:= f ,
Out[3]= Cos Sin
Functions can be significantly more complicated objects. Below is a function that
operates on functions. Like the function f above it takes two arguments, but, in this case,
its arguments are a function or expression, and a list containing the variable of integration
and the integration limits.
In[4]:= Integrate Exp I x , x, a, b
Out[4]=
a b
This particular function can be also be called with a function and a variable.
In[5]:= Integrate Exp I x , x
Out[5]=
x
Here is a function that also takes two arguments and operates on functions, but it
returns a graphical object as its value.
In[6]:= Plot Sin x 2 Sin x , x, 0, 2
1 2 3 4 5 6
1
0.5
0.5
1
Out[6]= Graphics
Programming involves writing a set of instructions to be applied for some appropri-
ate input. Whereas procedural programs provide a step-by-step set of instructions, func-
tional programming involves the application of functions to their arguments. For example,
here is a traditional procedural approach to switching the elements in a list of pairs.
In[7]:= lis , 1 , , 2 , , 3
Out[7]= , 1 , , 2 , , 3
In[8]:= temp lis;
Do temp i, 1 , temp i, 2 lis i, 2 , lis i, 1 ,
i, 1, Length lis ;
temp
Out[10]= 1, , 2, , 3,
We first allocate an empty array temp, of the same size as lis; then we put elements into
temp one by one as we loop over lis; finally we return the value of temp.
Here is a simpler procedure using a structured iteration.
In[11]:= Table lis i, 2 , lis i, 1 , i, 1, 3
Out[11]= 1, , 2, , 3,
76 An Introduction to Programming with Mathematica
And here is a functional approach to solving the same problem.
In[12]:= Map Reverse, lis
Out[12]= 1, , 2, , 3,
This simple example illustrates several of the key features of functional program-
ming. A functional approach often allows for a more direct implementation of the solution
to many problems, especially when list manipulations are involved. Notice that the proce-
dural approach required setting up a list structure and then looping over the list as i takes
on successive values, whereas the functional approach simply applied the Reverse func-
tion to the list directly.
Up to this point, we have described fairly simple functions and stayed focused on the
built-in functions present in Mathematica. In this chapter we will first take a look at some
of the most powerful and useful functional programming constructs in Mathematica and
then discuss the creation of our own functions, using many of the list and string manipulat-
ing functions discussed earlier. It is well worthwhile to spend time familiarizing yourself
with these functions by playing around with them; for example, create various lists and
apply built-in functions to them. Having a larger vocabulary of built-in functions will not
only make it easier to follow the programs and do the exercises here, but will enhance your
own programming skills as well.
4.2 Functions for manipulating expressions
Three of the most powerful and commonly used functions by experienced Mathematica
programmers are Map, Apply, and Thread. They provide very sophisticated ways of
manipulating expressions in Mathematica. Becoming familiar with them is essential to
functional programming in Mathematica. In this section we will discuss their syntax and
look at some simple examples of their use. We will also briefly look at some related func-
tions (Inner and Outer), which will prove useful in manipulating the structure of your
expressions. These higher-order functions will be used throughout the rest of this book.
4 Functional programming 77
Map
Map applies a function to each element in a list.
In[1]:= Map Head, 3,
22
7
,
Out[1]= Integer, Rational, Symbol
This is illustrated using an undefined function f and a simple linear list.
In[2]:= Map f, a, b, c
Out[2]= f a , f b , f c
More generally, mapping a function f over the expression g[a,b,c] essentially
wraps the function f around each of the elements of g.
In[3]:= Map f, g a, b, c
Out[3]= g f a , f b , f c
So this general computation is identical to Map[f,{a,b,c}], except in that example g is
replaced with List (remember that FullForm[{a,b,c}] is List[a,b,c]).
The real power of the Map function is that you can map any function across any
expression for which that function makes sense. Using the Reverse function with Map,
you can reverse the order of elements in each list of a nested list.
In[4]:= Map Reverse, a, b , c, d , e, f
Out[4]= b, a , d, c , f, e
The elements in each of the inner lists in a nested list can be sorted.
In[5]:= Map Sort, 2, 6, 3, 5 , 7, 4, 1, 3
Out[5]= 2, 3, 5, 6 , 1, 3, 4, 7
Often, you will need to define your own function to perform some computation on
every element of a list. This is the sort of computation that Map is expressly designed for.
Here is a list of three elements.
In[6]:= vec 2, , ;
If we wished to square each element and add 1, we could first define a function that per-
forms this computation on its arguments.
In[7]:= f x_ : x
2
1
78 An Introduction to Programming with Mathematica
Mapping this function over vec, will then wrap f around each element and evaluate f of
those elements.
In[8]:= Map f, vec
Out[8]= 5, 1
2
, 1
2
Later in this chapter we will look at even simpler ways of performing such
computations.
Thread and MapThread
The Thread function exchanges operations with arguments that are lists.
In[9]:= Thread g a, b, c , x, y, z
Out[9]= g a, x , g b, y , g c, z
You can accomplish something quite similar with MapThread. It differs from
Thread in that it takes two arguments the function that you are mapping and a list of
two (or more) lists as arguments of the function. It creates a new list in which the corre-
sponding elements of the old lists are paired (or zipped together).
In[10]:= MapThread g, a, b, c , x, y, z
Out[10]= g a, x , g b, y , g c, z
With Thread, you can fundamentally change the structure of the expressions you
are working with. For example, this threads the Equal function over the two lists given as
its arguments.
In[11]:= Thread Equal a, b, c , x, y, z
Out[11]= a x, b y, c z
In[12]:= Map FullForm, %
Out[12]= Equal a, x , Equal b, y , Equal c, z
Here is another example of the use of Thread. We start off with a list of variables
and a list of values.
In[13]:= vars x
1
, x
2
, x
3
, x
4
, x
5
;
In[14]:= values 1.2, 2.5, 5.7, 8.21, 6.66 ;
From these two lists, we create a list of rules.
In[15]:= Thread Rule vars, values
Out[15]= x
1
1.2, x
2
2.5, x
3
5.7, x
4
8.21, x
5
6.66
4 Functional programming 79
Notice how we started with a rule of lists and Thread produced a list of rules. In this way,
you might think of Thread as a generalization of Transpose.
Here are a few more examples of MapThread. This raises each element in the first
list to the power given by the corresponding element in the second list.
In[16]:= MapThread Power, 2, 6, 3 , 5, 1, 2
Out[16]= 32, 6, 9
Using Trace, you can view some of the intermediate steps that Mathematica performs in
doing this calculation.
In[17]:= MapThread Power, 2, 6, 3 , 5, 1, 2 Trace
Out[17]= MapThread Power, 2, 6, 3 , 5, 1, 2 ,
2
5
, 6
1
, 3
2
, 2
5
, 32 , 6
1
, 6 , 3
2
, 9 , 32, 6, 9
Using the List function, the corresponding elements in the three lists are placed in a list
structure (note that Transpose would do the same thing).
In[18]:= MapThread List, 5, 3, 2 , 6, 4, 9 , 4, 1, 4
Out[18]= 5, 6, 4 , 3, 4, 1 , 2, 9, 4
The Listable attribute
Many of the built-in functions that take a single argument have the property that, when a
list is the argument, the function is automatically applied to all of the elements in the list.
In other words, these functions are automatically mapped on to the elements of the list.
For example, the Log function has this attribute.
In[19]:= Log a, E, 1
Out[19]= Log a , 1, 0
This is the same result you get using the Map function.
In[20]:= Map Log, a, E, 1
Out[20]= Log a , 1, 0
Many of the built-in functions that take two or more arguments have the property
that, when multiple lists are the arguments, the function is automatically applied to all of
the corresponding elements in the list. In other words, these functions are automatically
threaded on to the elements of the list.
In[21]:= 4, 6, 3 5, 1, 2
Out[21]= 9, 7, 5
80 An Introduction to Programming with Mathematica
This gives the same result as using the Plus function with MapThread.
In[22]:= MapThread Plus, 4, 6, 3 , 5, 1, 2
Out[22]= 9, 7, 5
Functions that are either automatically mapped or threaded on to the elements of list
arguments are said to be Listable. Many of Mathematicas built-in functions have this
Attribute.
In[23]:= Attributes Log
Out[23]= Listable, NumericFunction, Protected
In[24]:= Attributes Plus
Out[24]= Flat, Listable, NumericFunction,
OneIdentity, Orderless, Protected
By default, functions that you define do not have any attributes associated with them. So,
for example, if you define a function g, say, it will not automatically be threaded over a list.
In[25]:= g a, b , c, d
Out[25]= g a, b , c, d
If you want your function to have the ability to thread over lists, give it the Listable
attribute using SetAttributes.
In[26]:= SetAttributes g, Listable
In[27]:= g a, b , c, d
Out[27]= g a , g b , g c , g d
Note that clearing a symbol only clears values associated with that symbol. It does not
clear any attributes associated with the symbol.
In[28]:= Clear g
In[29]:= ?g
Global`g
Attributes g Listable
To clear attributes, you need to use Remove.
In[30]:= Remove g
4 Functional programming 81
Now there is no remaining information associated with g.
In[31]:= ?g
Information::notfound : Symbol g not found. More
Apply
Whereas Map is used to perform the same operation on each element of an expression,
Apply is used to change the structure of an expression.
In[32]:= Apply h, g a, b, c
Out[32]= h a, b, c
The function h was applied to the expression g[a,b,c] and Apply replaced the head of
g[a,b,c] with h.
If the second argument is a list, applying h to that expression simply replaces its head
(List) with h.
In[33]:= Apply h, a, b, c
Out[33]= h a, b, c
The following computation shows the same thing, except we are using the internal represen
tation of the list {a,b,c} here to better see how the structure is changed.
In[34]:= Apply h, List a, b, c
Out[34]= h a, b, c
We see that the elements of List are now the arguments of h. Essentially, you should
think of Apply[h,expr] as replacing the head of expr with h.
In[35]:= Apply Plus, 1, 2, 3, 4
Out[35]= 10
Here, List[1,2,3,4] has been changed to Plus[1,2,3,4] or, in other words, the
head List has been replaced by Plus.
Plus[a,b,c,d] is the internal representation of the sum of these four symbols
that you would normally write a+b+c+d.
In[36]:= Plus a, b, c, d
Out[36]= a b c d
82 An Introduction to Programming with Mathematica
This list conversion can be applied to an entire list.
In[37]:= Apply h, 1, 2, 3 , 5, 6, 7
Out[37]= h 1, 2, 3 , 5, 6, 7
This is just vector addition.
In[38]:= Apply Plus, 1, 2, 3 , 5, 6, 7
Out[38]= 6, 8, 10
One important distinction between Map and Apply that you should be aware of
concerns the level of the expression at which each operate. By default, Map operates at
level 1. That is, in Map[h, expr], h will be applied to each element at the top level of expr.
So, for example, if expr consists of a nested list, h will be applied to each of the sublists, but
not deeper, by default.
In[39]:= Map h, a, b , c, d
Out[39]= h a, b , h c, d
If you wish to apply h at a deeper level, then you have to specify that explicitly using a
third argument to Map.
In[40]:= Map h, a, b , c, d , 2
Out[40]= h a , h b , h c , h d
Apply, on the other hand, operates at level 0. That is, in Apply[h,expr], Apply
looks at the part 0 of expr (that is, its Head) and replaces it with h.
In[41]:= Apply f, a, b , c, d
Out[41]= f a, b , c, d
Again, if you wish to apply h at a different level, then you have to specify that explic-
itly using a third argument to Apply.
In[42]:= Apply h, a, b , c, d , 1
Out[42]= h a, b , h c, d
For example, to apply Plus to each of the inner lists, you need to specify that Apply will
operate at level 1.
In[43]:= Apply Plus, 1, 2, 3 , 5, 6, 7 , 1
Out[43]= 6, 18
4 Functional programming 83
If you are a little unsure of what has just happened, consider the following example and,
instead of h, think of Plus.
In[44]:= Apply h, 1, 2, 3 , 5, 6, 7 , 1
Out[44]= h 1, 2, 3 , h 5, 6, 7
Inner and Outer
The Outer function applies a function to all of the combinations of the elements in
several lists. This is a generalization of the mathematical outer product.
In[45]:= Outer f, a, b , 2, 3, 4
Out[45]= f a, 2 , f a, 3 , f a, 4 , f b, 2 , f b, 3 , f b, 4
Using the List function as an argument, you can create lists of ordered pairs that com-
bine the elements of several lists.
In[46]:= Outer List, a, b , 2, 3, 4
Out[46]= a, 2 , a, 3 , a, 4 , b, 2 , b, 3 , b, 4
Using Inner, you can thread a function on to several lists and then use the result as the
argument to another function.
In[47]:= Inner f, a, b, c , d, e, f , g
Out[47]= g f a, d , f b, e , f c, f
This function lets you carry out some interesting operations.
In[48]:= Inner Times, x
1
, y
1
, z
1
, x
2
, y
2
, z
2
, Plus
Out[48]= x
1
x
2
y
1
y
2
z
1
z
2
In[49]:= Inner List, a, b, c , d, e, f , Plus
Out[49]= a b c, d e f
Looking at these two examples, you can see that Inner is really a generalization of the
mathematical dot product.
In[50]:= Dot x
1
, y
1
, z
1
, x
2
, y
2
, z
2
Out[50]= x
1
x
2
y
1
y
2
z
1
z
2
84 An Introduction to Programming with Mathematica
Exercises
1. Write a function addPair[{x,y}] that adds the elements in a pair. Then use your
addPair function to sum each pair from the following.
data 1, 2 , 2, 3 , 3, 4 , 4, 5 , 5, 6 ;
Your output should look like {3,5,7,9,11}.
2. Use Apply to add the elements in each pair from a list of pairs of numbers such as in
the previous exercise.
3. A matrix can be rotated by performing a number of successive operations. Rotate the
matrix {{1,2,3},{4,5,6}} clockwise by 90 degrees, obtaining
{{4,1},{5,2},{6,3}}, in two steps. Use TableForm to display the results.
4. While matrices can easily be added using Plus, matrix multiplication is more
complicated. The Dot function, written as a single period, can be used.
In[1]:= 1, 2 , 3, 4 . x, y
Out[1]= x 2 y, 3 x 4 y
Perform matrix multiplication on {{1,2},{3,4}} and {x,y} without using Dot.
(This can be done in two or three steps.)
5. FactorInteger[n] returns a nested list of prime factors and their exponents for
the number n.
In[2]:= FactorInteger 3628800
Out[2]= 2, 8 , 3, 4 , 5, 2 , 7, 1
Use Apply to reconstruct the number from this nested list.
6. Repeat the above exercise but instead use Inner to construct the original number n
from the factorization given by FactorInteger[n].
7. Using Inner, write a function div[vecs,vars] that computes the divergence of an
n-dimensional vector field vecs e
1
, e
2
, , e
n
dependent upon n variables
vars v
1
, v
2
, , v
n
. The divergence is given by the sum of the pairwise partial
derivatives.
e
1
v
1
e
2
v
2

e
n
v
n
4 Functional programming 85
4.3 Iterating functions
A commonly performed task in computer science and mathematics is to repeatedly apply a
function to some expression. Iterating functions has a long and rich tradition in the history
of computing. Perhaps the most famous example is Newtons method for root finding.
Chaos theory rests on studying how iterated functions behave under small perturbations of
their initial conditions or starting values. In this section, we will introduce several func-
tions available in Mathematica for function iteration. In later chapters we will apply these
and other programming constructs to look at some applications of iteration, including
Newtons method.
The Nest function is used to iterate functions. Here, g is iterated (or applied to a)
four times.
In[1]:= Nest g, a, 4
Out[1]= g g g g a
The NestList function displays all of the intermediate values of the Nest operation.
In[2]:= NestList g, a, 4
Out[2]= a, g a , g g a , g g g a , g g g g a
Using a starting value of 0.85, this generates a list of ten iterates of the Cos function.
In[3]:= NestList Cos, 0.85, 10
Out[3]= 0.85, 0.659983, 0.790003, 0.703843, 0.76236, 0.723208,
0.749687, 0.731902, 0.743904, 0.73583, 0.741274
The list elements above are the values of 0.85, Cos[0.85], Cos[Cos[0.85]], and so
on.
In[4]:= 0.85, Cos 0.85 , Cos Cos 0.85 , Cos Cos Cos 0.85
Out[4]= 0.85, 0.659983, 0.790003, 0.703843
In fact, the iterates of the cosine function tend towards a fixed point which can be
obtained with FixedPoint. This function is particularly useful when you do not know
how many iterations to perform on a function whose iterations eventually settle down.
In[5]:= FixedPoint Cos, 0.85
Out[5]= 0.739085
Whereas Nest and NestList operate on functions of one variable, Fold and
FoldList generalize this notion by iterating a function of two arguments. In the follow-
ing example, the function f is first applied to a starting value x and the first element from a
86 An Introduction to Programming with Mathematica
list, then this result is used as the first argument of the next iteration, with the second
argument coming from the second element in the list, and so on.
In[6]:= Fold f, x, a, b, c
Out[6]= f f f x, a , b , c
If FoldList is used, then you will see all of the intermediate results of the Fold
operation.
In[7]:= FoldList f, x, a, b, c
Out[7]= x, f x, a , f f x, a , b , f f f x, a , b , c
It is easy to see what is going on with the FoldList function by working with an arith-
metic operator. This generates running sums.
In[8]:= FoldList Plus, 0, a, b, c, d
Out[8]= 0, a, a b, a b c, a b c d
In[9]:= FoldList Plus, 0, 1, 2, 3, 4, 5
Out[9]= 0, 1, 3, 6, 10, 15
Exercises
1. Determine the locations after each step of a ten-step one-dimensional random walk.
(Recall that you have already generated the step directions in Exercise 3 at the end of
Section 3.2.)
2. Create a list of the step locations of a ten-step random walk on a square lattice.
3. Using Fold, create a function fac[n] that takes in an integer n as argument and
returns the factorial of n; that is, n n 1 n 2 3 2 1.
4 Functional programming 87
4.4 Programs as functions
A computer program is a set of instructions (a recipe) for carrying out a computation.
When a program is evaluated with appropriate inputs, the computation is performed and
the result is returned. In this sense, a program is a mathematical function and the inputs to
a program are the arguments of the function. Executing a program is equivalent to apply-
ing a function to its arguments or, as it is often referred, making a function call.
User-defined functions
While there are a great many built-in functions in Mathematica that can be used to carry
out computations, we invariably find ourselves needing customized functions. For exam-
ple, once we have written a program to compute some values for some particular inputs,
we might want to perform the same set of operations on different inputs. We would
therefore like to create our own user-defined functions that we could then apply in the same
way as we call a built-in function by entering the function name and specific argument
values. We will start with the proper syntax (or grammar) to use when writing a function
definition.
The function definition looks very much like a mathematical equation: a left-hand
side and a right-hand side separated by a colon-equal sign.
name[arg
1
_,arg
2
_,,arg
n
_]:= body
The left-hand side starts with a symbol. This symbol is referred to as the function
name (or sometimes just as the function, as in the sine function). The function name is
followed by a set of square brackets, inside of which are a sequence of symbols ending with
blanks. These symbols are referred to as the function argument names, or just the function
arguments.
The right-hand side of a user-defined function definition is called the body of the
function. The body can be either a single expression (a one-liner), or a series of expressions
(a compound function), both of which will be discussed in detail shortly. Argument names
from the left-hand side appear on the right-hand side without blanks. Basically, the right-
hand side is a formula stating what computations are to be done when the function is
called with specific values of the arguments.
When a user-defined function is defined with a delayed assignment (:=), nothing is
returned. Thereafter, calling the function by entering the left-hand side of the function
definition with specific values of the arguments causes the body of the function to be
88 An Introduction to Programming with Mathematica
computed with the specific argument values substituted where the argument names occur.
In other words, when using delayed assignments, the body of your function is only evalu-
ated when the function is called, not when it is first defined.
A simple example of a user-defined function is square which squares a value (it is a
good idea to use a function name that indicates the purpose of the function).
In[1]:= square x_ : x
2
After entering a function definition, you call the function in the same way that a
built-in function is applied to an argument.
In[2]:= square 5
Out[2]= 25
Building up programs
The ability to use the output of one function as the input of another is one of the keys to
functional programming. A mathematician would call this composition of functions. In
Mathematica, this sequential application of several functions is known as a nested function
call. Nested function calls are not limited to using a single function repeatedly, such as with
the built-in Nest and Fold functions.
In[3]:= Cos Sin Tan 4.0
Out[3]= 0.609053
To see the above computation more clearly, we can step through the computation.
In[4]:= Tan 4.0
Out[4]= 1.15782
In[5]:= Sin %
Out[5]= 0.915931
In[6]:= Cos %
Out[6]= 0.609053
Wrapping the Trace function around the computation lets us see all of the intermediate
expressions that are used in this evaluation.
In[7]:= Trace Cos Sin Tan 4.0
Out[7]= Tan 4. , 1.15782 , Sin 1.15782 , 0.915931 ,
Cos 0.915931 , 0.609053
4 Functional programming 89
You can read nested functions in much the same way that they are created, starting
with the innermost functions and working towards the outermost functions. For example,
the following expression determines whether all of the elements in a list are even numbers.
In[8]:= Apply And, Map EvenQ, 2, 4, 6, 7, 8
Out[8]= False
Let us step through the computation much the same as Mathematica does, from the
inside out.
1. Map the predicate EvenQ to every element in the list {2,4,6,7,8}.
In[9]:= Map EvenQ, 2, 4, 6, 7, 8
Out[9]= True, True, True, False, True
2. Apply the logical function And to the result of the previous step.
In[10]:= Apply And, %
Out[10]= False
Finally, here is a definition that can be used on arbitrary lists.
In[11]:= setEvenQ lis_ : Apply And, Map EvenQ, lis
In[12]:= setEvenQ 11, 5, 1, 18, 16, 6, 17, 6
Out[12]= False
Another, more complicated, example returns the elements in a list of positive num-
bers that are bigger than all of the preceding numbers in the list.
In[13]:= Union Rest FoldList Max, 0, 3, 1, 6, 5, 4, 8, 7
Out[13]= 3, 6, 8
The Trace of the function call shows the intermediate steps of the computation.
In[14]:= Trace Union Rest FoldList Max, 0, 3, 1, 6, 5, 4, 8, 7
Out[14]= FoldList Max, 0, 3, 1, 6, 5, 4, 8, 7 ,
Max 0, 3 , 3 , Max 3, 1 , Max 1, 3 , 3 ,
Max 3, 6 , 6 , Max 6, 5 , Max 5, 6 , 6 ,
Max 6, 4 , Max 4, 6 , 6 , Max 6, 8 , 8 ,
Max 8, 7 , Max 7, 8 , 8 , 0, 3, 3, 6, 6, 6, 8, 8 ,
Rest 0, 3, 3, 6, 6, 6, 8, 8 , 3, 3, 6, 6, 6, 8, 8 ,
Union 3, 3, 6, 6, 6, 8, 8 , 3, 6, 8
90 An Introduction to Programming with Mathematica
This computation can be described as follows:
The FoldList function is first applied to the function Max, 0, and the list
{3,1,6,5,4,8,7} (look at the Trace of this computation to see what Fold
List is doing here).
In[15]:= FoldList Max, 0, 3, 1, 6, 5, 4, 8, 7
Out[15]= 0, 3, 3, 6, 6, 6, 8, 8
The Rest function is then applied to the result of the previous step to remove the
first element of the list.
In[16]:= Rest %
Out[16]= 3, 3, 6, 6, 6, 8, 8
Finally, the Union function is applied to the result of the previous step to remove
duplicates.
In[17]:= Union %
Out[17]= 3, 6, 8
Here is the function definition.
In[18]:= maxima x_ : Union Rest FoldList Max, 0, x
Applying maxima to a list of numbers produces a list of all those numbers that are
larger than any number that comes before it.
In[19]:= maxima 4, 2, 7, 3, 4, 9, 14, 11, 17
Out[19]= 4, 7, 9, 14, 17
Notice that in each of the nested functions described here, the argument of the first
function was explicitly referred to, but the expressions that were manipulated in the
succeeding function calls were not identified other than as the results of the previous steps
(that is, as the results of the preceding function applications).
Here is an interesting application of building up a program with nested functions
the creation of a deck of cards. (Hint: The suit icons are entered by typing in \[ClubSuit
], \[DiamondSuit], etc.)
4 Functional programming 91
In[20]:= cardDeck Flatten
Outer List, , , , , Join Range 2, 10 , J, Q, K, A , 1
Out[20]= , 2 , , 3 , , 4 , , 5 , , 6 , , 7 , , 8 , , 9 , , 10 ,
, J , , Q , , K , , A , , 2 , , 3 , , 4 , , 5 , , 6 ,
, 7 , , 8 , , 9 , , 10 , , J , , Q , , K , , A , , 2 ,
, 3 , , 4 , , 5 , , 6 , , 7 , , 8 , , 9 , , 10 ,
, J , , Q , , K , , A , , 2 , , 3 , , 4 , , 5 , , 6 ,
, 7 , , 8 , , 9 , , 10 , , J , , Q , , K , , A
You might think of cardDeck as a name for the expression given on the right-hand side
of the immediate definition, or you might think of cardDeck as defining a function with
zero arguments.
To understand what is going on here, we will build up this program from scratch.
First we form a list of the number and face cards in a suit by combining a list of the num-
bers 2 through 10, Range[2,10], with a four-element list representing the jack, queen,
king, and ace, {J,Q,K,A}.
In[21]:= Join Range 2, 10 , J, Q, K, A
Out[21]= 2, 3, 4, 5, 6, 7, 8, 9, 10, J, Q, K, A
Now we pair each of the 13 elements in this list with each of the four elements in the list
representing the card suits { , , , }. This produces a list of 52 ordered pairs represent-
ing the cards in a deck, where the king of clubs, for example, is represented by { ,K}).
In[22]:= Outer List, , , , , %
Out[22]= , 2 , , 3 , , 4 , , 5 , , 6 , , 7 ,
, 8 , , 9 , , 10 , , J , , Q , , K , , A ,
, 2 , , 3 , , 4 , , 5 , , 6 , , 7 , , 8 ,
, 9 , , 10 , , J , , Q , , K , , A ,
, 2 , , 3 , , 4 , , 5 , , 6 , , 7 , , 8 ,
, 9 , , 10 , , J , , Q , , K , , A ,
, 2 , , 3 , , 4 , , 5 , , 6 , , 7 , , 8 ,
, 9 , , 10 , , J , , Q , , K , , A
While we now have all of the cards in the deck, they are grouped by suit in a nested list.
We therefore un-nest the list:
In[23]:= Flatten %, 1
Out[23]= , 2 , , 3 , , 4 , , 5 , , 6 , , 7 , , 8 , , 9 , , 10 ,
, J , , Q , , K , , A , , 2 , , 3 , , 4 , , 5 , , 6 ,
, 7 , , 8 , , 9 , , 10 , , J , , Q , , K , , A , , 2 ,
, 3 , , 4 , , 5 , , 6 , , 7 , , 8 , , 9 , , 10 ,
, J , , Q , , K , , A , , 2 , , 3 , , 4 , , 5 , , 6 ,
, 7 , , 8 , , 9 , , 10 , , J , , Q , , K , , A
Voila!
92 An Introduction to Programming with Mathematica
The step-by-step construction that we used here, applying one function at a time,
checking each function call separately, is a very efficient way to prototype your programs in
Mathematica. We will use this technique again in the next example.
We will perform what is called a perfect shuffle, consisting of cutting the deck in half
and then interleaving the cards from the two halves. Rather than working with the large
list of 52 ordered pairs during the prototyping, we will use a short made-up list. A short list
of an even number of ordered integers is a good choice for the task.
In[24]:= d Range 6
Out[24]= 1, 2, 3, 4, 5, 6
We first divide the list into two equal-sized lists.
In[25]:= Partition d, Length d 2
Out[25]= 1, 2, 3 , 4, 5, 6
We now want to interleave these two lists to form {1,4,2,5,3,6}. The first step is to
pair the corresponding elements in each of the two lists above. This can be done using the
Transpose function.
In[26]:= Transpose %
Out[26]= 1, 4 , 2, 5 , 3, 6
We now un-nest the interior lists using the Flatten function. We could flatten our
simple list using Flatten[], but, since we know that ultimately we will be dealing with
ordered pairs rather than integers, we will use Flatten[,1] as we did in creating the
card deck.
In[27]:= Flatten %, 1
Out[27]= 1, 4, 2, 5, 3, 6
That does the job. Given this prototype, it is easy to write the actual function to
perform a perfect shuffle on a deck of cards. Notice we have generalized this shuffle to lists
of arbitrary length.
In[28]:= shuffle lis_ :
Flatten Transpose Partition lis, Length lis 2 , 1
In[29]:= shuffle cardDeck
Out[29]= , 2 , , 2 , , 3 , , 3 , , 4 , , 4 , , 5 , , 5 , , 6 ,
, 6 , , 7 , , 7 , , 8 , , 8 , , 9 , , 9 , , 10 ,
, 10 , , J , , J , , Q , , Q , , K , , K , , A , , A ,
, 2 , , 2 , , 3 , , 3 , , 4 , , 4 , , 5 , , 5 , , 6 ,
, 6 , , 7 , , 7 , , 8 , , 8 , , 9 , , 9 , , 10 ,
, 10 , , J , , J , , Q , , Q , , K , , K , , A , , A
4 Functional programming 93
Let us take this example one step further and construct a function that deals cards
from a card deck. We will construct this function in stages using the prototyping method
we showed earlier.
First we need to define a function that removes a single element from a randomly chosen
position in a list.
In[30]:= removeRand lis_ :
Delete lis, Random Integer, 1, Length lis
The function removeRand first uses the Random function to randomly choose an integer
k between 1 and the length of the list, and then uses the Delete function to remove the
kth element of the list. For example, if a list has 10 elements, an integer between 1 and 10,
say 6, is randomly determined and the element in the sixth position in the list is then
removed from the list.
In[31]:= lis 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 ;
removeRand lis
Out[32]= 2, 3, 4, 5, 6, 7, 8, 9, 10
Now we want to make a function call that applies the removeRand function to the
cardDeck list, then applies the removeRand function to the resulting list, then applies
the removeRand function to the resulting list, and so on, a total of n times. The way to
carry out this operation is with the Nest function.
Nest[removeRand, cardDeck, n]
Lastly, we want the cards that are removed fromcardDeck rather than those that remain.
Complement[cardDeck, Nest[removeRand, cardDeck, n]]
Now, we write this up formally into the user-defined deal function.
In[33]:= deal n_ : Complement cardDeck, Nest removeRand, cardDeck, n
Let us try it out.
In[34]:= deal 5
Out[34]= , 3 , , K , , 2 , , K , , J
Not a bad hand!
94 An Introduction to Programming with Mathematica
Exercises
1. One of the games in the Illinois State Lottery is based on choosing n numbers, each
between 0 and 9, with duplicates allowed; in practice, a selection is made from
containers of numbered ping pong balls. We can model this game using a simple
user-defined function, which we will call pick (after the official lottery names of Pick
3 and Pick 4).
In[1]:= pick n_ : Table Random Integer, 0, 9 , n
In[2]:= pick 4
Out[2]= 0, 9, 0, 4
This program can be generalized to performrandom sampling with replacement on any
list. Write a function chooseWithReplacement[lis,n], where lis is the list, n is
the number of elements being chosen and the following is a typical result.
In[3]:= chooseWithReplacement a, b, c, d, e, f, g, h , 3
Out[3]= h, b, f
2. Write your own user-defined functions using the ToCharacterCode and From
CharacterCode functions to perform the same operations as StringInsert and
StringDrop.
3. Create a function distance[a,b] that finds the distance between two points a and
b in the plane.
4. Write a user-defined function interleave2 that interleaves the elements of two
lists of unequal length. (You have already seen how to interleave lists of equal length
using Partition earlier in this section.) Your function should take the lists
{1,2,3} and {a,b,c,d} as inputs and return {1,a,2,b,3,c,d}.
5. Write a nested function call that creates a deck of cards and performs a perfect
shuffle on it.
6. Write nested function calls using the ToCharacterCode and FromCharacter
Code functions to perform the same operations as the built-in StringJoin and
StringReverse functions.
4 Functional programming 95
4.5 Auxiliary functions
There are several major drawbacks to the deal function created in the previous section. In
order to use deal, the definition of removeRand and the value of cardDeck must be
entered before calling deal. It would be much more convenient if we could incorporate
these functions within the deal function definition itself. In the next section, we will show
how this can be done.
Compound functions
The left-hand side of a compound function is the same as that of any user-defined function.
The right-hand side consists of consecutive expressions enclosed in parentheses and
separated by semicolons.
name arg
1
_, arg
2
_, , arg
n
_ : expr
1
; expr
2
; ; expr
m
The expressions can be user-defined functions (also known as auxiliary functions),
value declarations, and function calls. When a compound function is evaluated with
particular argument values, these expressions are evaluated in order and the result of the
evaluation of the last expression is returned (by adding a semicolon after expr
n
, the display
of the final evaluation result can also be suppressed).
We will work with the deal function to illustrate how a compound function is
created. We need the following three expressions.
In[1]:= cardDeck Flatten Outer List,
, , , , Join Range 2, 10 , J, Q, K, A , 1 ;
In[2]:= removeRand lis_ :
Delete lis, Random Integer, 1, Length lis
In[3]:= deal n_ : Complement cardDeck, Nest removeRand, cardDeck, n
The conversion to a compound function is easily done. We will first remove the old
definitions.
In[4]:= Clear deal, cardDeck, removeRand
96 An Introduction to Programming with Mathematica
Now we can create and enter the new definition.
In[5]:= deal n_ :
cardDeck Flatten Outer List,
, , , , Join Range 2, 10 , J, Q, K, A , 1 ;
removeRand lis_ : Delete lis,
Random Integer, 1, Length lis ;
Complement cardDeck, Nest removeRand, cardDeck, n
Let us check that this works.
In[6]:= deal 5
Out[6]= , 3 , , 2 , , 3 , , 4 , , Q
A couple of things should be pointed out about the right-hand side of a compound
function definition. Since the expressions on the right-hand side are evaluated in order,
value declarations and auxiliary function definitions should be given before they are used
and the argument names used on the left-hand side of auxiliary function definitions must
differ from the argument names used by the compound function itself.
Finally, when we enter a compound function definition, we are entering not only the
function but also the auxiliary functions and the value declarations. If we then remove the
function definition using Clear, the auxiliary function definitions and value declarations
remain. This can cause a problem if we subsequently try to use the names of these auxiliary
functions and values elsewhere.
So how does the global rule base treat compound functions? When a compound
function definition is entered, a rewrite rule corresponding to the entire definition is
created. Each time the compound function is subsequently called, rewrite rules are created
from the auxiliary function definitions and value declarations within the compound
function.
In[7]:= ?cardDeck
Global`cardDeck
cardDeck , 2 , , 3 , , 4 , , 5 , , 6 , , 7 , , 8 ,
, 9 , , 10 , , J , , Q , , K , , A , , 2 , , 3 , , 4 ,
, 5 , , 6 , , 7 , , 8 , , 9 , , 10 , , J , , Q , , K ,
, A , , 2 , , 3 , , 4 , , 5 , , 6 , , 7 , , 8 , , 9 ,
, 10 , , J , , Q , , K , , A , , 2 , , 3 , , 4 , , 5 ,
, 6 , , 7 , , 8 , , 9 , , 10 , , J , , Q , , K , , A
It is considered bad programming practice to leave auxiliary definitions in the global
rule base that are not explicitly needed by the user of your function. In fact, it could
interfere with a users workspace and cause unintended problems.
4 Functional programming 97
To prevent these additional rewrite rules from being placed in the global rule base,
you can localize their names by using the Module construct in the compound function
definition. This is what we discuss next.
Localizing names: Module
When a user-defined function is written, it is generally a good idea to isolate the names of
values and functions defined on the right-hand side from the outside world in order to
avoid any conflict with the use of a name elsewhere in the session (for example, cardDeck
might be used elsewhere to represent a pinochle deck). This can be done by wrapping the
right-hand side of the function definition in the built-in Module function.
name arg
1
_, arg
2
_, , arg
n
_ : Module name
1
, name
2
value, ,
expr
The first argument of the Module function is a list of the names we want to localize. If we
wish, we can assign values to these names, as is shown with name
2
above (the assigned value
is only an initial value and can be changed subsequently). The list is separated from the
right-hand side by a comma and so the parentheses enclosing the right-hand side of a
compound function are not needed.
We can demonstrate the use of Module with the deal function.
In[8]:= Clear deal
In[9]:= deal n_ : Module cardDeck, removeRand ,
cardDeck Flatten Outer List,
, , , , Join Range 2, 10 , J, Q, K, A , 1 ;
removeRand lis_ : Delete lis,
Random Integer, 1, Length lis ;
Complement cardDeck, Nest removeRand, cardDeck, n
Briefly, when Module is encountered, the symbols that are being localized (card
Deck and removeRand in the above example) are temporarily given new and unique
names, and all occurrences of those symbols in the body of the Module are given those
new names as well. In this way, these unique and temporary names, which are local to the
function, will not interfere with any functions outside of the Module.
98 An Introduction to Programming with Mathematica
It is generally a good idea to wrap the right-hand side of all compound function
definitions in the Module function. Another way to avoid conflicts in the use of names of
auxiliary function definitions is to use a function that can be applied without being given a
name. Such functions are called pure functions, which we discuss in Section 4.6.
Localizing values: Block
Occasionally, you will need to localize a value associated with a symbol without localizing
the symbol name itself. For example, you may have a recursive computation that requires
you to temporarily reset the system variable $RecursionLimit. You can do this with
Block, thereby only localizing the value of $RecursionLimit during the evaluation
inside the Block.
In[10]:= Block $RecursionLimit 20 ,
x g x
$RecursionLimit::reclim :
Recursion depth of 20 exceeded. More
Out[10]= g g
g g g g g g g g g g g g g g g g Hold g x
Notice the global value of $RecursionLimit is unchanged.
In[11]:= $RecursionLimit
Out[11]= 256
This construct is similar to what is done for the iterators in Table, Do, Sum, and Prod
uct.
Module, on the other hand, would create an entirely new symbol, $Recursion
Limit$nn that would have nothing to do with the global variable $RecursionLimit,
and so Module would be inappropriate for this particular task.
Localizing constants: With
Another scoping construct is available when you simply need to localize constants. If, in
the body of your function, you use a variable that is assigned a constant once and never
changes, then With is the preferred means to localize that constant.
This sets the global variable y to have the value 5.
In[12]:= y 5;
4 Functional programming 99
Here is a simple function that initializes y as a local constant.
In[13]:= f x_ : With y x 1 ,
y
We see the global symbol is unchanged and it does not interfere with the local symbol y
inside the With.
In[14]:= y
Out[14]= 5
In[15]:= f 2
Out[15]= 3
Using With, you can initialize local constants with the values of global symbols. For
example:
In[16]:= With y y ,
g x_ : x y
This shows that the global value for y was inserted inside g.
In[17]:= ?g
Global`g
g x$_ : x$ 5
Resetting the global value of y has no effect on the localized y inside the With.
In[18]:= y 1;
In[19]:= g 5
Out[19]= 10
Exercises
1. Write a compound function definition for the location of steps taken in an n-step
random walk on a square lattice. Hint: Use the definition for the step increments of
the walk as an auxiliary function.
2. The PerfectSearch function defined in Section 1.1 is impractical for checking
large numbers because it has to check all numbers from 1 through n. If you already
100 An Introduction to Programming with Mathematica
know the perfect numbers below 500, say, it is inefficient to check all numbers from
1 to 1,000 if you are only looking for perfect numbers in the range 500 to 1,000.
Modify searchPerfect so that it accepts two numbers as input and computes all
perfect numbers between the inputs. For example, PerfectSearch[a,b] will
produce a list of all perfect numbers in the range froma to b.
3. Overload the PerfectSearch function to compute all 3-perfect numbers. A 3-per-
fect number is such that the sum of its divisors equals three times the number. For
example, 120 is 3-perfect since it is equal to three times the sum of its divisors.
In[1]:= Apply Plus, Divisors 120
Out[1]= 360
Find the only other 3-perfect number under 1,000.
You can overload PerfectSearch as defined in Exercise 2 above by defining a
three-argument version PerfectSearch[a,b,3].
4. Overload PerfectSearch to find the three 4-perfect numbers less than 2,200,000.
5. Redefine PerfectSearch so that it accepts as input a number k, and two numbers
a and b, and computes all k-perfect numbers in the range froma to b. For example,
PerfectSearch[1,30,2] would compute all 2-perfect numbers in the range
from 1 to 30 and, hence, would output {6,28}.
6. If n is defined to be the sum of the divisors of n, then n is called superperfect if
n 2 n. Write a function SuperPerfectSearch[a,b] that finds all super-
perfect numbers in the range froma to b.
7. Often in processing files you will be presented with expressions that need to be
converted into a format that can be more easily manipulated inside Mathematica. For
example, a file may contain dates in the form 20030515 to represent May 15, 2003.
Mathematica represents its dates as a list {year,month,day,hour,minutes,seconds}.
Write a function convertToDate[n] to convert a number consisting of eight
digits such as 20030515 into a list of the form {2003,5,15}.
In[2]:= convertToDate 20030515
Out[2]= 2003, 5, 15
4 Functional programming 101
4.6 Pure functions
A pure function is a function that does not have a name and that can be used on the spot;
that is, at the moment it is created. This is often convenient, especially if the function is
only going to be used once or as an argument to a higher-order function, such as Map,
Fold, or Nest. The built-in function Function is used to create a pure function.
The basic form of a pure function is Function[x,body] for a pure function with a
single variable x (any symbol can be used for the variable), and
Function[{x,y,},body] for a pure function with more than one variable. The body
looks like the right-hand side of a user-defined function definition, with the variables x, y,
, where argument names would be.
As an example, the square function we created earlier can be written as a pure
function.
In[1]:= Function z, z
2
Out[1]= Function z, z
2
There is also a standard input form that can be used in writing a pure function which is
easier to write than the Function notation but can be a bit cryptic to read. The right-
hand side of the function definition is rewritten by replacing the variable by the pound
symbol (#) and ending the expression with the ampersand symbol (&) to indicate that this
is a pure function.
#
2
&
If there is more than one variable, #1, #2, and so on are used.
A pure function can be used exactly like more conventional looking functions, by
following the function with the argument values enclosed in square brackets. First we show
the pure function using Function.
In[2]:= Function z, z
2
6
Out[2]= 36
Here is the same thing, but using the more cryptic shorthand notation (the parentheses in
the following example are purely for readability and can be omitted if you wish).
In[3]:= #
2
& 6
Out[3]= 36
We can, if we wish, give a pure function a name and then use that name to call the function
later. This has the same effect as defining the function in the more traditional manner.
In[4]:= squared #
2
&;
102 An Introduction to Programming with Mathematica
In[5]:= squared 6
Out[5]= 36
Pure functions are very commonly used with higher-order functions like Map and
Apply, so, before going further, let us first look at a few simple examples of the use of
pure functions.
Here is a list of numbers.
In[6]:= lis 2, 5, 6.1 ;
Now suppose we wished to square each number and then add 1 to it. The pure function
that does this is: #
2
1 &. So that is what we need to map across this list.
In[7]:= Map #
2
1 &, lis
Out[7]= 5, 26, 38.21
In the next example we will create a set of data and then use the Select function to
filter out outliers.
In[8]:= data 24.39001, 29.669, 9.321, 20.8856,
23.4736, 22.1488, 24.7434, 22.1619, 21.1039,
24.8177, 27.1331, 25.8705, 39.7676, 24.7762
Out[8]= 24.39, 29.669, 9.321, 20.8856, 23.4736, 22.1488, 24.7434,
22.1619, 21.1039, 24.8177, 27.1331, 25.8705, 39.7676, 24.7762
A plot of the data shows there are two outliers.
In[9]:= ListPlot data, PlotStyle PointSize .02 ;
2 4 6 8 10 12 14
10
15
20
25
30
35
40
The Select function takes two arguments the first is the expression from which it will
select elements, and the second argument is a function that must return True or False.
Select[expr,test] will then select those elements from expr that return True when test is
applied to them.
4 Functional programming 103
Suppose we wish to exclude all data points that lie outside of the range 20 to 30.
Then we need a function that returns True if its argument is in that range.
In[10]:= Select data, 20 # 30 &
Out[10]= 24.39, 29.669, 20.8856, 23.4736, 22.1488, 24.7434,
22.1619, 21.1039, 24.8177, 27.1331, 25.8705, 24.7762
A good way to become comfortable with pure functions is to see them in action, so
we will convert some of the functions we defined earlier into pure functions, showing both
the (#)& and the Function forms so that you can decide which you prefer to use.
This function tests whether all the elements of a list are even.
In[11]:= areEltsEven lis_ : Apply And, Map EvenQ, lis
In[12]:= areEltsEven 2, 4, 5, 8
Out[12]= False
Here it is written using pure functions.
In[13]:= Function lis, Apply And, Map EvenQ, lis 2, 4, 5, 8
Out[13]= False
In[14]:= Apply And, Map EvenQ, #1 & 2, 4, 5, 8
Out[14]= False
This function returns each element in the list greater than all previous elements.
In[15]:= maxima[x_] := Union[Rest[FoldList[Max, 0, x]]]
In[16]:= maxima 2, 6, 3, 7, 9, 2
Out[16]= 2, 6, 7, 9
Here it is written using pure functions.
In[17]:= Function x, Union Rest FoldList Max, 0, x 2, 6, 3, 7, 9, 2
Out[17]= 2, 6, 7, 9
In[18]:= Union Rest FoldList Max, 0, # & 2, 6, 3, 7, 9, 2
Out[18]= 2, 6, 7, 9
We can also create nested pure functions. For example, this maps the pure squaring
function over the three-element list {3,2,7}.
In[19]:= Map #
2
&, 3, 2, 7
Out[19]= 9, 4, 49
104 An Introduction to Programming with Mathematica
When dealing with nested pure functions, the shorthand notation can be used for
each of the pure functions but care needs to be taken to avoid confusion as to which #
variable belongs to which pure function. This can be avoided by using Function, in
which case different variable names can be used.
In[20]:= Function y, Map Function x, x
2
, y 3, 2, 7
Out[20]= 9, 4, 49
Exercises
1. Write a function to sum the squares of the elements of a numeric list.
2. Write a function to sum the digits of any integer. You will need the IntegerDig
its function (use ?IntegerDigits, or look up IntegerDigits in the Help
Browser to find out about this function).
3. Using the definition of the distance function from Exercise 3 of Section 4.4, write
a new function diameter[pts] that, given a set of points in the plane, finds the
maximum distance between all pairs of points. Try to incorporate the distance
function into diameter without naming it explicitly; that is, use it as a pure func-
tion. Consider using Distribute to get the set of all pairs of points.
In[1]:= pts p
1
, p
2
, p
3
;
In[2]:= Distribute pts, pts , List
Out[2]= p
1
, p
1
, p
1
, p
2
, p
1
, p
3
, p
2
, p
1
,
p
2
, p
2
, p
2
, p
3
, p
3
, p
1
, p
3
, p
2
, p
3
, p
3
4. Take the removeRand function defined in Section 4.4 and rewrite it as a pure
function.
In[3]:= removeRand[lis_] :=
Delete[lis, Random[Integer, {1, Length[lis]}]]
5. Convert the deal function developed earlier into one that uses pure functions. Use
the pure function version of the removeRand function from the previous exercise in
your new deal function definition.
6. Create a function RepUnit[n] that generates integers of length n consisting
entirely of 1s. For example RepUnit[7] should produce 1111111.
4 Functional programming 105
7. Create a function chooseWithoutReplacement[lis,n] that is a generalization
of the deal function in that it will work with any list.
8. Write a pure function that moves a random walker from one location on a square
lattice to one of the four adjoining locations with equal probability. For example,
starting at {0,0}, the function should return either {0,1}, {0,-1}, {1,0} or
{-1,0} with equal likelihood. Now, use this pure function with NestList to
generate the list of step locations for an n-step random walk starting at {0,0}.
9. Create a function WordsStartingWith[lis,char] that outputs all those words in
lis that begin with the character char. As a sample list, you can use the dictionary.dat
file that comes with Mathematica.
Here is a platform-independent path to the dictionary file.
In[4]:= wordfile ToFileName $InstallationDirectory, "Documentation",
"English", "Demos", "DataFiles" , "dictionary.dat"
Out[4]= C:\Program Files\Wolfram Research\Mathematica\5.1\
Documentation\English\Demos\DataFiles\dictionary.dat
This reads in the file using ReadList, specifying the type of data we are reading in
as a Word.
In[5]:= words ReadList wordfile, Word ;
10. Modify Exercise 9 above so that WordsStartingWith accepts a string of arbitrary
length as its second argument.
11. A naive approach to polynomial arithmetic would require three additions and six
multiplies to carry out the arithmetic in the expression a x
3
b x
2
c x d. Using
Horners method for fast polynomial multiplication, this expression can be repre-
sented as d x c x b a x , where there are now half as many multiplies. In general,
the number of multiplies for an n-degree polynomial is given by:
In[6]:= Binomial n 1, 2
Out[6]=
1
2
n 1 n
This, of course, grows quadratically with n, whereas Horners method grows linearly.
Create a function Horner[lis,var] that implements Horners method for polyno-
mial multiplication. Here is some sample input and the corresponding output that
your function should generate.
106 An Introduction to Programming with Mathematica
In[7]:= Horner a, b, c, d , x
Out[7]= d x c x b a x
In[8]:= Expand %
Out[8]= d c x b x
2
a x
3
4.7 One-liners
In the simplest version of a user-defined function, there are no value declarations or
auxiliary function definitions; the right-hand side is a single nested function call whose
arguments are the names of the arguments on the left-hand side, without the blanks. These
one-liners are fantastically useful and so we will discuss them in the context of three
examples, one from electrical engineering (computing Hamming distance), one from
ancient history (the Josephus problem), and the last a simple and practical problem
(counting change).
Hamming distance
When a code is transmitted over a channel in the presence of noise, errors will often occur.
The task of channel coding is to represent the source information in a manner that mini-
mizes the error probability in decoding. Hamming distance is used in source coding to
represent an information source with the minimum number of symbols. For two lists of
binary symbols, the Hamming distance is defined as the number of nonmatching elements
and so gives a measure of the how well these two lists match up.
Let us first think about how we might determine if two binary symbols are identical.
SameQ[x,y] will return True if x and y are identical.
In[1]:= SameQ 0, 0 , SameQ 1, 0 , SameQ 1, 1
Out[1]= True, False, True
So we need to thread SameQ over the two lists of binary numbers
In[2]:= MapThread SameQ, 1, 0, 0, 1, 1 , 0, 1, 0, 1, 0
Out[2]= False, False, True, True, False
4 Functional programming 107
and then count up the occurrences of False.
In[3]:= Count %, False
Out[3]= 3
So a first definition of HammingDistance could be accomplished by putting these last
two pieces together.
In[4]:= HammingDistance lis1_, lis2_ :
Count MapThread SameQ, lis1, lis2 , False
In[5]:= HammingDistance 1, 0, 0, 1, 1 , 0, 1, 0, 1, 0
Out[5]= 3
We might try to solve this problem by a more direct approach. Since we are dealing
with binary information, we could use some of the logical binary operators built into
Mathematica.
Here is our transposed list again.
In[6]:= lis Transpose 1, 0, 0, 1, 1 , 0, 1, 0, 1, 0
Out[6]= 1, 0 , 0, 1 , 0, 0 , 1, 1 , 1, 0
BitXor[x,y] returns the bitwise XOR of x and y. So if x and y can only be among
the binary integers 0 or 1, BitXor will return 0 whenever they are the same and will
return 1 whenever they are different.
In[7]:= Apply BitXor, 0, 0 , 1, 0 , 1, 1 , 1
Out[7]= 0, 1, 0
Here then is BitXor applied to lis.
In[8]:= Apply BitXor, lis, 1
Out[8]= 1, 1, 0, 0, 1
And here are the number of 1s that occur in that list.
In[9]:= Apply Plus, %
Out[9]= 3
Summing up, our function HammingDistance2 first pairs up the lists (Transpose),
then determines which pairs contain different elements (apply BitXor), and finally counts
up the number of 1s (Apply[Plus,]).
In[10]:= HammingDistance2 lis1_, lis2_ : Apply Plus,
Apply BitXor, Transpose lis1, lis2 , 1
108 An Introduction to Programming with Mathematica
In[11]:= HammingDistance2 1, 0, 0, 1, 1 , 0, 1, 0, 1, 0
Out[11]= 3
Let us compare the running times of these implementations using a large data set, in
this case two lists consisting of one million 0s and 1s.
In[12]:= data1 Table Random Integer , 10
6
;
In[13]:= data2 Table Random Integer , 10
6
;
In[14]:= Timing HammingDistance data1, data2
Out[14]= 1.162 Second, 499801
In[15]:= Timing HammingDistance2 data1, data2
Out[15]= 1.392 Second, 499801
Although these times do not look too bad, they are in fact too slow for any serious
work with signal processing. The exercises ask you to write an implementation of Hamming
Distance that runs about two orders of magnitude faster than those presented here.
As an aside, the above computations are not a bad check on the built-in random
number generator we would expect that about one half of the paired up lists would
contain different elements.
The Josephus problem
Flavius Josephus was a Jewish historian during the RomanJewish war of the first century
AD. Through his writings comes the following story:
The Romans had chased a group of ten Jews into a cave and were about to attack. Rather than
die at the hands of their enemy, the group chose to commit suicide one by one. Legend has it
though, that they decided to go around their circle of ten individuals and eliminate every other
person until only one was left.
Who was the last to survive? Although a bit macabre, this problem has a definite
mathematical interpretation that lends itself well to a functional style of programming. We
will start by changing the problem a bit (the importance of rewording a problem can
hardly be overstated; the key to most problem-solving resides in turning something we can
not work with into something we can work with). We will restate the problem as follows: n
people are lined up. The first person is moved to the end of the line, the second person is
removed from the line, the third person is moved to the end of the line, and so on until
only one person remains in the line.
4 Functional programming 109
The statement of the problem indicates that there is a repetitive action, performed
over and over again. It involves the use of the RotateLeft function (move the person at
the front of the line to the back of the line) followed by the use of the Rest function
(remove the next person from the line).
In[16]:= Rest RotateLeft # & a, b, c, d
Out[16]= c, d, a
At this point it is already pretty clear where this computation is headed. We want to take a
list and, using the Nest function, perform the pure function call (Rest[Rotate
Left[#])& on the list until only one element remains. A list of n elements will need n 1
calls. So we can now write the function, to which we give the apt name survivor.
In[17]:= survivor lis_ :
Nest Rest RotateLeft # &, lis, Length lis 1
Trying out the survivor function on a list of ten, we see that the fifth position will be
the position of the survivor.
In[18]:= survivor Range 10
Out[18]= 5
Tracing the applications of RotateLeft in this example gives a very clear picture of what
is going on. The following form of TracePrint shows only the results of the applica-
tions of RotateLeft that occur during evaluation of the expression survivor[
Range[6]].
In[19]:= TracePrint survivor Range 6 , RotateLeft
RotateLeft
2, 3, 4, 5, 6, 1
RotateLeft
4, 5, 6, 1, 3
RotateLeft
6, 1, 3, 5
RotateLeft
3, 5, 1
RotateLeft
1, 5
Out[19]= 5
110 An Introduction to Programming with Mathematica
Pocket change
As another example, we will write a program to perform an operation most of us do every
day: calculating how much change we have in our pocket. Suppose we have the following
collection of coins.
In[20]:= coins p, p, q, n, d, d, p, q, q, p
Out[20]= p, p, q, n, d, d, p, q, q, p
Assume p, n, d, and q represent pennies, nickels, dimes, and quarters, respectively. Let us
start by using the Count function to determine the number of pennies we have.
In[21]:= Count coins, p
Out[21]= 4
This works. So let us do the same thing for all of the coin types.
In[22]:= Count coins, p , Count coins, n ,
Count coins, d , Count coins, q
Out[22]= 4, 1, 2, 3
Looking at this list, it is apparent that there ought to be a more compact way of
writing the list. If we Map a pure function involving Count and coins on to the list
{p,n,d,q}, it should do the job.
In[23]:= Map Count coins, #1 & , p, n, d, q
Out[23]= 4, 1, 2, 3
Now that we know how many coins of each type we have, we want to calculate how much
change we have. We first do the calculation manually to see what we get for an answer (so
we will know when our program works).
In[24]:= 4 1 1 5 2 10 3 25
Out[24]= 104
From the above computation we see that the lists {4,1,2,3} and {1,5,10,25} are
first multiplied together element-wise and then the elements of the result are added. This
suggests a few possibilities.
In[25]:= Apply Plus, 4, 1, 2, 3 1, 5, 10, 25
Out[25]= 104
In[26]:= 4, 1, 2, 3 . 1, 5, 10, 25
Out[26]= 104
4 Functional programming 111
Either of these operations are suitable for the job (to coin a phrase, theres not a penny,
nickel, quarter, or dimes worth of difference). We will write the one-liner using the first
method.
In[27]:= pocketChange x_ :
Apply Plus, Map Count x, # & , p, n, d, q 1, 5, 10, 25
In[28]:= pocketChange coins
Out[28]= 104
Exercises
1. Write a function to compute the Hamming distance of two binary lists (assumed to
be of equal length), using Select and an appropriate predicate function.
2. All of the implementations of Hamming distance discussed so far are a bit slow for
large datasets. You can get a significant speedup in running times by using functions
that are optimized for working with numbers (a topic we discuss in detail in Chapter
8). Write an implementation of Hamming distance using the Total function and
then compare running times with the other versions discussed in this chapter.
3. One of the best ways to learn how to write programs is to practice reading code. We
list below a number of one-liner function definitions along with a very brief explana-
tion of what these user-defined functions do and a typical input and output. Decon-
struct these programs to see what they do and then reconstruct them as compound
functions without any pure functions.
a. Determine the frequencies with which distinct elements appear in a list.
In[1]:= frequencies lis_ : Map #, Count lis, # &, Union lis
In[2]:= frequencies a, a, b, b, b, a, c, c
Out[2]= a, 3 , b, 3 , c, 2
b. Divide up a list into parts each of whose lengths are given by the second
argument.
In[3]:= split1 lis_, parts_ :
Inner Take lis, #1, #2 &, Drop #1, 1 1,
Rest #1 , List & FoldList Plus, 0, parts
112 An Introduction to Programming with Mathematica
In[4]:= split1 Range 10 , 2, 5, 0, 3
Out[4]= 1, 2 , 3, 4, 5, 6, 7 , , 8, 9, 10
This is the same as the previous program, done in a different way.
In[5]:= split2[lis_, parts_] :=
Map[(Take[lis, # + {1, 0}])&,
Partition[FoldList[Plus, 0, parts], 2, 1]]
c. Another game in the Illinois State Lottery is based on choosing n numbers, each
between 0 and s with no duplicates allowed. Write a user-defined function called
lotto (after the official lottery names of Little Lotto and Big Lotto) to perform
sampling without replacement on an arbitrary list. (Note: The difference between
this function and the function chooseWithoutReplacement is that the order
of selection is needed here.)
In[6]:= lotto1 lis_, n_ : Flatten
Rest MapThread Complement, RotateRight # , # , 1 &
NestList Delete #, Random Integer, 1, Length # &,
lis, n
In[7]:= lotto1 Range 10 , 5
Out[7]= 10, 3, 2, 7, 6
This is the same as the previous program, done in a different way.
In[8]:= lotto2 lis_, n_ : Take Transpose Sort
Transpose Table Random , Length lis , lis 2 , n
As the split and lotto programs illustrate, user-defined functions can be written
in several ways. The choice as to which version of a program to use has to be based
on efficiency. A program whose development time was shorter and which runs faster
is better than a program which took more time to develop and which runs more
slowly. Although concise Mathematica programs tend to run fastest, when execution
speed is a primary concern (when dealing with very large lists) it is a good idea to
take various programming approaches and perform Timing tests to determine the
fastest program.
4. Use the Timing function to determine when (in terms of the relative sizes of the list
and the number of elements being chosen) it is preferable to use the different ver-
sions of the lotto function.
5. Rewrite the pocketChange function in two different ways one, using Dot, and
the other using Inner.
4 Functional programming 113
6. Make change with quarters, dimes, nickels, and pennies using the fewest coins.
In[9]:= makeChange 119
Out[9]= 4, 1, 1, 4
7. Write a one-liner to create a list of the step locations of a two-dimensional random
walk that is not restricted to a lattice. Hint: Each step length must be the same, so the
sum of the squares of the x- and y-components of each step should be equal to 1.
8. Write a one-liner version of convertToDate as described in Exercise 7 from
Section 4.5. Consider the built-in function FromDigits.
114 An Introduction to Programming with Mathematica
5 Procedural programming
Conventional programming languages like C and Fortran embody a style of program-
ming that has roots in the early days of computing when resource constraints forced
programmers to write their code in a step-by-step manner. These procedures, as they
came to be known, typically involved certain basic elements: looping over an array,
conditional statements that controlled the flow of execution, logical constructs to build
up tests, and functions to jump around from one place in a program to another.
Although newer languages have introduced many new programming paradigms,
procedural programming continues to be used and remains an appropriate style for
certain kinds of problems. In this chapter we will look at how procedural program-
ming is used in Mathematica, discuss what types of problems it is most appropriate for,
and compare Mathematicas implementation with other languages.
5.1 Introduction
A procedure is a series of instructions that are evaluated in a definite order. The following
program is a procedure.
In[1]:= mat a, b, c , d, e, f , g, h, k ;
newmat mat;
Do newmat i, j mat j, i ,
i, Length mat , j, Length mat ;
newmat
Out[4]= a, d, g , b, e, h , c, f, k
In[5]:= MatrixForm %
Out[5]//MatrixForm=
a d g
b e h
c f k
We could look at this procedure as a compound expression consisting of a sequence
of four expressions: the first assigns the symbolic 3 3 matrix to the symbol mat; the
second is also an assignment copying the matrix to another symbol, newmat; the third
expression loops through the matrix, interchanging columns and rows of the original and
putting them into the new matrix essentially performing a transpose operation; the final
expression simply outputs the new matrix.
Procedural programs also typically involve some flow control. What this means is that,
depending upon a certain condition, different steps in the procedure will be followed.
Perhaps the simplest example of this is an If statement.
In[6]:= f x_ : If 20 x 30, x,
Print "The number ", x, " is outside the range."
In[7]:= f 23
Out[7]= 23
In[8]:= f 66
The number 66 is outside the range.
The value of the first argument of the If function determines the direction of the
rest of the evaluation. This is a control structure.
These are typical components of procedural programs a series of expressions to
evaluate in some order and functions to control the flow of execution. In this chapter we
will explore these topics in addition to conditional definitions which are another form of
flow control. All of these features will greatly expand what we can do with Mathematica and
we will find many applications of these techniques in later chapters on recursion and
numerics.
5.2 Loops and iteration
Newtons method
One of the most famous of all numerical algorithms is Newtons method for finding the
roots of a function. Even though Mathematica includes a built-in function, FindRoot, that
implements this method, this is a classic use of iteration and so central to numerical
analysis that it is well worth your time learning how to implement it.
Throughout this section we will use the function x
2
50, whose root is, of course,
the square root of 50. Here is the computation using the built-in FindRoot.
In[1]:= FindRoot x
2
50 0, x, 50
Out[1]= x 7.07107
The number 50 in {x,50} is the initial guess of the root.
116 An Introduction to Programming with Mathematica
So why should you learn to program a root-finder yourself? As we stated above, it is a
classical algorithm and the basis of many more advanced root-finding techniques in
numerical analysis. But also, with many numerical problems, the built-in operations do not
always give you optimal results. This is because the built-in functions are designed to work
for the broadest possible set of situations, but might have occasional trouble with certain
exceptional cases. An example is the function f x x
1 3
.
In[2]:= FindRoot x
1 3
0, x, 0.1
FindRoot::lstol :
The line search decreased the step size to within tolerance
specified by AccuracyGoal and PrecisionGoal but was
unable to find a sufficient decrease in the merit
function. You may need more than MachinePrecision digits
of working precision to meet these tolerances. More
Out[2]= x 0.000405502 2.29415 10
15
Although this particular functions root can be better approximated using an option
(DampingFactor) to FindRoot, we will find it very instructive to program our own
root-finding functions that can solve this problem and, in the process, learn about the
structure of iterative programming.
In[3]:= FindRoot x
1 3
0, x, 0.1 , DampingFactor 2
Out[3]= x 8.93553 10
17
Do loops
Suppose we are given a function f and can compute its derivative, f . Then Newtons
algorithm works as follows:
give an initial estimate of the root, say x
0
keep generating better estimates, x
1
, x
2
, , using the following rule until you are
done (we will discuss this later):
x
i 1
x
i
f x
i
f x
i
The method is illustrated in Figure 5.1. Under the favorable circumstances pictured there
the estimates get closer and closer to the root.
5 Procedural programming 117
f x
0
x
0
x
1
Figure 5.1: Illustration of Newtons method
We will discuss in a moment when to stop, but first let us look at an example. For the
function f x x
2
50, the derivative is f x 2 x. This specific case is shown in Figure
5.2, with 50 itself as the initial estimate. Let us see what happens after five iterations of this
procedure.
In[4]:= f x_ : x
2
50
In[5]:= x0 50;
In[6]:= x1 N x0
f x0
f x0
Out[6]= 25.5
In[7]:= x2 N x1
f x1
f x1
Out[7]= 13.7304
In[8]:= x3 N x2
f x2
f x2
Out[8]= 8.68597
In[9]:= x4 N x3
f x3
f x3
Out[9]= 7.22119
In[10]:= x5 N x4
f x4
f x4
Out[10]= 7.07263
118 An Introduction to Programming with Mathematica
f 50
50 25.5 13.7 8.7
Figure 5.2: Newtons method for f x x
2
50
As you can see, these values are getting closer and closer to the real square root of 50,
which is approximately 7.07107.
We need to discuss how to decide when we are satisfied with the answer we have
computed. First, though, note one thing: Wherever we decide to stop, say at the fifth
iteration, all the previous values we computed are of no interest. So we could have avoided
introducing those new names by instead just writing the following:
In[11]:= a 50;
In[12]:= a N a
f a
f a
Out[12]= 25.5
In[13]:= a N a
f a
f a
Out[13]= 13.7304
In[14]:= a N a
f a
f a
Out[14]= 8.68597
In[15]:= a N a
f a
f a
Out[15]= 7.22119
In[16]:= a N a
f a
f a
Out[16]= 7.07263
5 Procedural programming 119
To return to the question of when to terminate the computation, one simple answer
is: repeat it ten times.
In[17]:= Do a N a f a f' a , 10
In general, Do[expr,{n}], evaluates expr n times. So, in this case, we can initialize a and
perform the ten evaluations as follows:
In[18]:= a 50;
Do a N a
f a
f a
, 10
In[20]:= a
Out[20]= 7.07107
Note that the Do loop itself yields no value (or rather, it yields the special value Null,
which is a symbol Mathematica uses when there is no result from an evaluation; nothing is
printed). But the important thing is that the Do loop assigns a value to a that is very close
to the square root of 50.
The arguments of Do are the same as those of Table (see Section 3.2; see also
Exercise 3 at the end of this section).
Do[expr,{i,imin,imax,di}]
This form repeats expr with variable i having values imin, imin di, and so on, as long
as the value of imax is not exceeded. The loop is repeated a total of imax imin di
times. Furthermore, if di is omitted, it is assumed to be 1; and if only i and imax are given,
both imin and di are assumed to be 1. For example, if we wanted to print each approxima-
tion and label it with a number, we could do that by using a compound expression inside
the body of the Do loop, in this case, adding a Print statement.
In[21]:= a 50;
Do a N a f a f' a ;
Print "approximation ", i, ": ", a , i, 1, 6
approximation 1: 25.5
approximation 2: 13.7304
approximation 3: 8.68597
approximation 4: 7.22119
approximation 5: 7.07263
approximation 6: 7.07107
120 An Introduction to Programming with Mathematica
Example: Random permutations
Let us look at another example of a Do loop. We will create a function random
Permutation[lis] that will take a list as an argument and generate a random permuta-
tion of its elements.
To build this function up step by step, we first start with a small list of ten elements.
In[23]:= lis Range 10
Out[23]= 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
The idea will be to choose a position within the list at random and remove the element in
that position and put it into a new list res.
In[24]:= rand : Random Integer, 1, Length lis
In[25]:= x Part lis, rand
Out[25]= 1
In[26]:= res ;
res Append res, x
Out[27]= 1
We then repeat the above process on the remaining elements of the list.
In[28]:= lis Complement lis, x
Out[28]= 2, 3, 4, 5, 6, 7, 8, 9, 10
In[29]:= x lis rand
res Append res, x
lis Complement lis, x
Out[29]= 8
Out[30]= 1, 8
Out[31]= 2, 3, 4, 5, 6, 7, 9, 10
In this example we know explicitly how many iterations to perform in our Do loop: n
times, where n is the length of the list that is being worked on.
First we clear some symbols.
In[32]:= Clear lis, res, x, rand ;
5 Procedural programming 121
Now we just put the pieces of the previous computations together in one input.
In[33]:= lis Range 10 ;
res ;
Do
x Part lis, Random Integer, 1, Length lis ;
res Append res, x ;
lis Complement lis, x ,
i, 1, 10
When we are done, the result is left in the new list res.
In[36]:= res
Out[36]= 7, 1, 2, 5, 8, 10, 4, 3, 9, 6
Here then is our function randomPermutation that takes a list as an argument
and generates a random permutation of that lists elements.
In[37]:= Clear res, rand, x, lis
In[38]:= randomPermutation lis_ : Module res , x, l2 lis ,
Do
x Part l2, Random Integer, 1, Length l2 ;
res Append res, x ;
l2 Complement l2, x ,
i, 1, Length lis ;
res
Here is a permutation of the list consisting of the first 20 integers.
In[39]:= randomPermutation Range 20
Out[39]= 7, 20, 16, 8, 19, 10, 15, 17,
13, 3, 5, 12, 1, 11, 2, 4, 6, 18, 9, 14
And here is a random permutation of the lowercase letters of the English alphabet.
In[40]:= alphabet Map FromCharacterCode , Range 97, 122
Out[40]= a, b, c, d, e, f, g, h, i, j, k,
l, m, n, o, p, q, r, s, t, u, v, w, x, y, z
In[41]:= randomPermutation alphabet
Out[41]= i, l, c, s, t, d, j, q, y, f, e,
k, x, a, h, r, o, g, u, z, v, n, p, w, b, m
122 An Introduction to Programming with Mathematica
While loops
Let us return to Newtons method for finding roots and see how we can use a different
control structure for the iteration. In the previous section on Do loops, we explicitly
stopped the iteration after ten times through the loop. Ten times is okay for f x x
2
50,
but not always. Consider the function x sin x .
In[42]:= g x_ : x Sin x
It has a root at 0.
In[43]:= g 0
Out[43]= 0
However, ten iterations of Newtons algorithm does not get us very close to it.
In[44]:= xi 1.0;
Do xi N xi
g xi
g xi
, 10
In[46]:= xi
Out[46]= 0.0168228
Twenty-five iterations does a bit better.
In[47]:= xi 1.0;
Do xi N xi
g xi
g xi
, 25
In[49]:= xi
Out[49]= 0.0000384172
In truth, no fixed number of iterations is going to do the trick for all functions. We
need to iterate repeatedly until our estimate is close enough to stop. When is that? There
are a number of ways to answer that question, none always best, but here is an easy one:
when f x
i
is very close to zero. So, choose to be a very small number, and iterate until
f x
i
.
But how can we write a loop that will test some condition and stop when the condi-
tion is no longer met? The looping construct Do iterates a number of times that is fixed
when the loop is begun. We need a new kind of iterative function. It is While, and it has
the following form.
While[test,expr]
5 Procedural programming 123
The first argument is the test or condition, the second the body. It works like this: evaluate
the test; if it is true then evaluate the body and then the test again. If it is true again, then
again evaluate the body and the test. Continue this way until the test evaluates to False.
Note that the body may not be evaluated at all (if the test is false the first time), or it may
be evaluated once, or a thousand times.
This is just what we want: if the estimate is not yet close enough, compute a new
estimate and try again.
In[50]:= f x_ : x
2
50
In[51]:= .0001;
xi 50;
While Abs f xi ,
xi N xi
f xi
f xi
In[54]:= xi
Out[54]= 7.07107
To wrap things up, let us put this all into a function.
In[55]:= findRoot fun_, init_, _ : Module xi init ,
While Abs fun xi ,
xi N xi
fun xi
fun xi
;
xi
In[56]:= findRoot f, 50, .0001
Out[56]= 7.07107
Instead of setting a global variable to the final estimate, this function returns that
estimate as its value. (For an explanation of why we introduced the local variable xi, see
the end of this subsection.)
Let us work with this example a little more. Suppose we would like to know how
many iterations were needed to find the answer. One possibility is to insert a Print to
show the value of xi each time through the loop.
In[57]:= findRoot fun_, init_, _ : Module xi init ,
While Abs fun xi ,
Print "x ", xi ;
xi N xi
fun xi
fun xi
;
xi
124 An Introduction to Programming with Mathematica
In[58]:= findRoot f, 50, 0.001
x 50
x 25.5
x 13.7304
x 8.68597
x 7.22119
x 7.07263
Out[58]= 7.07107
Counting the lines shows that the function converged after six iterations (note that we
were seeing the value of xi at the beginning of each execution of the body). A better idea
would be to have the function actually count the number of iterations and return it as part
of its answer.
In[59]:= findRoot fun_, init_, _ :
Module xi init, count 0 , While Abs fun xi ,
count count 1;
xi N xi
fun xi
fun xi
;
xi, count
In[60]:= findRoot f, 50, 0.001
Out[60]= 7.07107, 6
Here is another question: in all these versions of findRoot, f[xi] is computed
two times at each iteration, once in the condition and once in the body. In many circum-
stances, calls to f are very time consuming, and should be minimized. Can we arrange that
f[xi] only be computed once in each iteration?
The solution to this is to create a new local variable, funxi, which always contains
the value of fun[xi] for the current value of xi. We can ensure that it does so by recom-
puting it whenever xi is reassigned.
In[61]:= findRoot fun_, init_, _ :
Module xi init, funxi fun init ,
While Abs funxi ,
xi N xi
funxi
fun xi
;
funxi fun xi ;
xi
5 Procedural programming 125
In all our examples, we used Module to introduce a local variable to which we
assigned values in the body of the While. We did this to avoid a common error in the use
of iteration: attempting to assign to a functions argument. For example, the following version
of findRoot does not work.
In[62]:= findRoot fun_, x_, _ :
While Abs fun x ,
x N x
fun x
fun x
;
x
In[63]:= findRoot Sin, .1, .01
Set::setraw : Cannot assign to raw object 0.1`. More
General::stop : Further output of Set::setraw will
be suppressed during this calculation. More
Out[63]= $Aborted
What happened can be seen from the trace (of which we have only shown some) of the
output.
In[64]:= TracePrint findRoot Sin, .1, .01 , findRoot
findRoot
While Abs Sin 0.1 0.01, 0.1 N 0.1
Sin 0.1
Sin 0.1
; 0.1
Set::setraw : Cannot assign to raw object 0.1`. More
General::stop : Further output of Set::setraw will
be suppressed during this calculation. More
Out[64]= $Aborted
The x in the body of findRoot is replaced by the argument .1, which is perfectly
normal, leaving an expression of the form 0.1 = something, which is not possible. There
is a way around this, using the HoldFirst attribute, but introducing local variables is
much better style. It is very disconcerting, after all, to call a function and find, when it is
done, that your global variables have changed values.
126 An Introduction to Programming with Mathematica
NestWhile and NestWhileList
Let us look again at the last version of the findRoot function we just created.
In[65]:= findRoot fun_, init_, _ :
Module xi init, funxi fun init ,
While Abs funxi ,
xi N xi
funxi
fun' xi
;
funxi fun xi ;
xi
The While loop evaluates the body of this function (the two assignments, one to xi and
the other to funxi) until the test fails. There is another function we could use to simplify
this calculation it is NestWhile.
NestWhile[f,init,test]
This function iterates f with initial value init, while test continues to be true.
Let us rewrite findRoot using NestWhile. The first argument is the function we
are iterating. Here we will use a pure function that represents the Newton iteration. The
second argument is the initial guess, the initial value for the iteration. The third argument
to NestWhile is the test that will be performed each time through the loop until it
returns False. In this case, we are setting an explicit value for of 0.001 and so our test is
f x .001.
In[66]:= f x_ : x
2
50
In[67]:= findRoot fun_, init_ :
NestWhile #
fun #
fun' #
&, N init , Abs fun # .001 &
This computes the square root of 50 with an initial guess of 10.
In[68]:= findRoot f, 10
Out[68]= 7.07108
We can easily write a function findRootList based on NestWhileList that will
output all the intermediate computed values.
In[69]:= findRootList fun_, init_ :
NestWhileList #
fun #
fun' #
&, N init , Abs fun # .001 &
5 Procedural programming 127
In[70]:= findRootList f, 10
Out[70]= 10., 7.5, 7.08333, 7.07108
Note: the functions introduced in this section are rather simplistic implementations of
Newtons algorithm. At this stage, we are only interested in learning about how to use
some of Mathematicas procedural functions to implement the iterations here. In their
current form, they have some serious limitations regarding accuracy and precision that we
will address in Chapter 8, where we will discuss numerical issues in detail. The exercises at
the end of this section also walk the reader through several improvements to these
functions.
Exercises
1. Compute the square roots of 50 and 60 simultaneously, that is, with a single Do loop.
2. Compare the use of a Do loop with using the function Nest (see Section 4.3). In
particular, compute the square root of 50 using Nest.
3. Do is closely related to Table, the main difference being that Do does not return any
value, whereas Table does. Use Table instead of Do in your solution to Exercise 1.
What do you get?
4. Compute Fibonacci numbers iteratively. You will need to have two variables, say
this and prev, giving the two most recent Fibonacci numbers, so that after the ith
iteration, this and prev have the values F
i
and F
i 1
, respectively.
5. One additional improvement can be made to the findRoot program developed in
this section. Notice that the derivative of the function fun is computed each time
through the loop. This is quite inefficient. Rewrite findRoot so that the derivative
is computed only once and that result is used in the body of the loop.
6. Another termination criterion for root-finding is to stop when x
i
x
i 1
; that
is, when two successive estimates are very close. The idea is that if we are not getting
much improvement, we must be very near the root. The difficulty in programming
this is that we need to remember the two most recent estimates computed. (It is
similar to computing Fibonacci numbers iteratively, as in Exercise 4.) Program
findRoot this way.
7. The built-in FindRoot function is set up so that you can monitor intermediate
computations using the option EvaluationMonitor.
128 An Introduction to Programming with Mathematica
In[1]:= xintermed ;
FindRoot x
2
50, x, 50 ,
EvaluationMonitor AppendTo xintermed, x ;
In[3]:= xintermed
Out[3]= 50., 25.5, 13.7304, 8.68597,
7.22119, 7.07263, 7.07107, 7.07107
Modify each of the versions of findRoot presented in the text that use a Do or
While loop to produce a list of all the estimates computed.
f x_ : x
2
50;
findRootList f, 50, 0.001
50, 25.5, 13.7304, 8.68597, 7.22119, 7.07263, 7.07107, 7.07107
8. To guard against starting with a poor choice of initial value, modify findRootList
to take, as an argument, a list of initial values, and simultaneously compute approxima-
tions for each until one converges; then return that one.
9. The bisection method is quite useful for finding roots of functions. If a continuous
function f x is such that f a 0 and f b 0 for two real numbers a and b, then, as a
consequence of the Intermediate Value Theorem of calculus, a root of f must occur
between a and b. If f is now evaluated at the midpoint of a and b, and if
f a b 2 0, then the root must occur between a b 2 and b; if not, then it
occurs between a and a b 2. This bisection can be repeated until a root is found
to any specified tolerance.
Define bisect[f ,{a,b, }] to compute a root of f , within , using the bisection
method. You should give it two initial values a and b and assume that f a f b 0;
that is, one of f a and f b is positive and the other is negative.
10. Using a While loop, write a function gcd[m,n] that computes the greatest com-
mon divisor of m and n. The Euclidean algorithm for computing the gcd of two
numbers m and n, assumed to be positive integers, sets m n, and n mmod n. It
iterates this process until n 0, at which point the gcd of m and n is left in the value
of m.
11. Create a procedural definition for each of the following functions, first by creating a
new list and filling in the elements. For each function, create a definition using a Do
loop and another using Table. For example, the following function first creates an
array of the same dimension as mat, but consisting of 0s. Then inside the Do loop it
assigns the element in position {j,i} in mat to position {i,j} in matA, effectively
5 Procedural programming 129
performing a transpose operation. Finally, it returns matA, since the Do loop itself
does not return a value.
In[4]:= transpose mat_ :
Module matA Table Table 0, n Length mat ,
m Length mat 1 , Do matA i, j mat j, i ,
i, 1, m ,
j, 1, n ;
matA
In[5]:= mat1 a, b, c , d, e, f , h, k, l ;
In[6]:= MatrixForm mat1
Out[6]//MatrixForm=
a b c
d e f
h k l
In[7]:= MatrixForm transpose mat1
Out[7]//MatrixForm=
a d h
b e k
c f l
Note this same computation could be performed with what is referred to as a struc-
tured iteration using Table.
In[8]:= transposeStruc mat_ :
Module
matA Table 0, n Length mat , m Length mat 1 ,
Table matA i, j mat j, i , i, m , j, n
In[9]:= transposeStruc mat1 MatrixForm
Out[9]//MatrixForm=
a d h
b e k
c f l
a. Create the function reverse[vec], which reverses the elements in the list vec.
b. Create a function rotateRight[vec,n], where vec is a vector and n is a
(positive or negative) integer.
130 An Introduction to Programming with Mathematica
c. Create a procedural implementation of rotateRows, which could be defined in
this functional way:
In[10]:= rotateRows[mat_] := Map[(rotateRight[mat[[#]], #-1])&,
Range[1, Length[mat]]]
That is, it rotates the ith row of mat i 1 places to the right.
d. Create a procedural function rotateRowsByS, which could be defined in this
functional way:
In[11]:= rotateRowsByS mat_, S_ ; Length mat Length S :
Map rotateRight mat #1 , S #1 & , Range 1, Length mat
That is, it rotates the ith row of matA by the amount S[[i]].
e. Create a function compress[lisA, lisB], where lisA and lisB are lists of equal
length, and lisB contains only Boolean values (False and True), selects out of
lisA those elements corresponding to True in lisB. For example, the result of
compress[{a,b,c,d,e},{True,True,False,False,True}] should
be {a,b,e}. To know what size list to create, you will first need to count the
occurrences of True in lisB.
5.3 Flow control
Conditional functions
In this section we will look at functions that control the flow of execution of an evaluation.
Perhaps the simplest and easiest to understand of these class of functions is the If state-
ment. Here is a rather simplistic implementation of the absolute value function, using If.
In[1]:= abs x_ : If x 0, x, x
In[2]:= abs 4
Out[2]= 4
The If function takes three arguments: the first is a test; if the test evaluates to
True, then the second argument is evaluated; if the test evaluates to False, then the third
argument of the If is evaluated.
5 Procedural programming 131
If can also be used in conjunction with the higher-order functions discussed in
Chapter 4 to achieve greater flexibility. For example, abs can now be mapped over a list of
numbers.
In[3]:= Map abs, 2, 1, 0, 1, 2
Out[3]= 2, 1, 0, 1, 2
By default, this function will not automatically map across lists.
In[4]:= abs 2, 1, 0, 1, 2
Out[4]= If 2, 1, 0, 1, 2 0, 2, 1, 0, 1, 2 , 2, 1, 0, 1, 2
If you want abs to behave like many of the built-in functions and automatically map
across lists when they are given as the argument to abs, you need to make the function
Listable as described in Sections 2.4 and 4.2.
In[5]:= SetAttributes abs, Listable
In[6]:= abs 2, 1, 0, 1, 2
Out[6]= 2, 1, 0, 1, 2
Here are some additional examples using If. Given a list, the following function
adds 1 to all the numeric quantities occurring in it.
In[7]:= incrementNumbers lis_ : Map If NumericQ #1 , # 1, # &, lis
In[8]:= incrementNumbers 4, f, 6.1 I,
Out[8]= 5, f, 7.1 , 1
Here is a function that divides 100 by every number in a numerical list, except 0s.
In[9]:= divide100By lis_ : Map If # 0, #,
100
#
&, lis
In[10]:= divide100By 5, , 0
Out[10]= 20,
100
, 0
Here is a function to remove consecutive occurrences of the same value.
In[11]:= removeRepetitions lis_ :
Fold If #2 Last #1 , #1, Append #1, #2 &,
First lis , Rest lis
In[12]:= removeRepetitions 0, 1, 1, 2, 2, 2, 1, 1
Out[12]= 0, 1, 2, 1
132 An Introduction to Programming with Mathematica
As a final example of If, the function applyChar takes a list as an argument. This
list must contain, first, a character, which must be one of "+", "-", "*", or "/"; that
character must be followed by all numbers. applyChar applies the function named by the
character to the elements of the rest of the list.
In[13]:= applyChar lis_ : Module op First lis , nums Rest lis ,
If op " ", Apply Plus, nums ,
If op " ", Apply Subtract, nums ,
If op " ", Apply Times, nums ,
If op " ", Apply Divide, nums ,
Print "Bad argument to applyChar"
In[14]:= applyChar " ", 1, 2, 3, 4
Out[14]= 10
(Recall the Module function, which permits us to introduce local variables. In this case, it
saves us from having to write First[lis] and Rest[lis] several times each.)
Even though the argument list in applyChar must contain one of the four operators as its
first element, it is still best to check for it explicitly; otherwise, if the condition is ever
violated, the results may be very mysterious. We have used the Print function, which
prints all of its arguments (of which it can have an arbitrary number) and then skips to a
new line.
In[15]:= applyChar "^", 2, 5, 10
Bad argument to applyChar
Notice that what we have in this code is several nested Ifs, each occurring in the
false part of the previous one. Thus, the structure of the computation is a sequence of tests
of predicates until one is found to be true, at which point a result can be computed. Such a
sequence of cascading If statements can get quite long, and the indentation can become
unmanageable, so it is conventional to violate the usual rule for indenting If expressions
and indent this type of structure as follows:
If cond
1
, result
1
,
If cond
2
, result
2
,
If cond
n
, result
n
,
result
n 1
] ]]
Conditional definitions can be written using another construct in Mathematica, the
Condition operator, /;. For example, the abs function can be entered (using several
definitions) as follows:
5 Procedural programming 133
In[16]:= Clear abs
In[17]:= abs x_ : x ; x 0
In[18]:= abs x_ : x ; x 0
The first definition should be interpreted as abs[x] is equal to x whenever (or under the
condition that) x is greater than or equal to 0 and the second definition as abs[x] is
equal to the opposite of x whenever x is less than 0.
The conditions on the right-hand side of the rules can, in fact, be entered on the
left-hand side of these definitions as follows:
In[19]:= abs x_ ; x 0 : x
In[20]:= abs x_ ; x 0 : x
This last notation has the advantage of preventing the right-hand side of our definitions
from being evaluated whenever the pattern on the left does not match.
In[21]:= abs 4
Out[21]= 4
In[22]:= abs z
Out[22]= abs z
This use of multiple rules associated with the symbol abs is a very useful and powerful
means of associating rules with symbols under user-defined conditions and we turn to it
next.
Multiclause definitions
The abs function defined above is fine for integers and real number arguments, but, since
the complex numbers cannot be ordered, the initial test comparing a complex number
argument with 0 will fail.
In[23]:= abs 3 4 I
GreaterEqual::nord :
Invalid comparison with 3 4 attempted. More
Less::nord : Invalid comparison with 3 4 attempted. More
Out[23]= abs 3 4
134 An Introduction to Programming with Mathematica
We can solve this problem by providing an additional definition for abs.
In[24]:= Clear abs ;
abs x_ : Sqrt Re x
2
Im x
2
; x Complexes;
abs x_ : x ; x 0
abs x_ : x ; x 0
The test as the first argument of If on the right-hand side checks to see if x is an
element of the domain of complex numbers and, if it is, then re x
2
im x
2
is com-
puted. If x is not complex, nothing is done, but then the other definition for abs will be
invoked.
In[28]:= abs 3 4 I
Out[28]= 5
In[29]:= abs 3
Out[29]= 3
The condition itself can appear on the left-hand side of the function definition, as
part of the pattern match. Here is a slight variation on the abs definition.
In[30]:= Clear abs
abs x_ : If x 0, x, x
abs x_ ; x Complexes : Sqrt Re x
2
Im x
2
In[33]:= abs 3 4 I
Out[33]= 5
In[34]:= abs 3
Out[34]= 3
We may want to add an additional rule for symbols.
In[35]:= abs x_ ; Head x Symbol : x
In[36]:= abs z
Out[36]= z
Such a definition is called a multiclause definition. In this case we have associated three
rules with abs; two are rather specific and will only be applied if the argument to abs
passes the conditions specified. If neither of those conditions are met, then the most
general rule (the one with no conditions on x) will be used.
5 Procedural programming 135
Which and Switch
Recall the earlier definition of applyChar defined using cascading Ifs.
In[37]:= applyChar lis_ : Module op First lis , nums Rest lis ,
If op " ", Apply Plus, nums ,
If op " ", Apply Subtract, nums ,
If op " ", Apply Times, nums ,
If op " ", Apply Divide, nums ,
Print "Bad argument to applyChar"
Needless to say, this is a little difficult to read and figure out which clause goes with
which If. Fortunately, cascaded Ifs are so common that Mathematica provides a more
direct way of writing them, using the function Which.
Which cond
1
, result
1
,
cond
2
, result
2
,
cond
n
, result
n
,
True, result
n 1
]
This has exactly the same effect as the cascaded If expression above: it tests each
condition in turn, and, when it finds an i such that cond
i
is true, it returns result
i
as the
result of the Which expression itself. If none of the conditions turns out to be true, then it
will test the final condition, namely the expression True, which always evaluates to true,
and it will then return result
n 1
.
applyChar can now be written more neatly.
In[38]:= applyChar lis_ : Module op First lis , nums Rest lis ,
Which op " ", Apply Plus, nums ,
op " ", Apply Subtract, nums ,
op " ", Apply Times, nums ,
op " ", Apply Divide, nums ,
True, Print "Bad argument to applyChar"
One last form deserves mention. Our use of the Which command is still quite
special, in that it consists of a simple sequence of comparisons between a variable and a
constant. Since this is also a common form, Mathematica again provides a special function
for it, called Switch.
136 An Introduction to Programming with Mathematica
Switch[expr,
pattern
1
, result
1
,
pattern
2
, result
2
,
pattern
n
, result
n
,
_, result
n 1
]
This evaluates expr and then checks each pattern, in order, to see whether expr
matches; as soon as expr matches one, say pattern
i
, it returns the value of result
i
. Of course,
if none of the patterns pattern
1
, , pattern
n
matches, the _ certainly will.
If all the patterns happen to be constants, the Switch expression is equivalent to the
following Which expression.
Which[expr == pattern
1
, result
1
,
expr == pattern
2
, result
2
,
expr == pattern
n
, result
n
,
True, result
n 1
]
Here, then, is our final version of applyChar.
In[39]:= applyChar lis_ : Module op First lis , nums Rest lis ,
Switch op,
" ", Apply Plus, nums ,
" ", Apply Subtract, nums ,
" ", Apply Times, nums ,
" ", Apply Divide, nums ,
_, Print "Bad argument to ApplyChar"
Notice that Switch uses the blank character, _, for the final, or default case, just as
Which uses the always-true expression True. We will have much more to say about
patterns and pattern matching in Chapter 6.
5 Procedural programming 137
Piecewise
Several of the functions we created in previous sections could be caste as piecewise-defined
functions. Although technically not a procedural construct, Piecewise (new in Version
5.1) is designed specifically for such problems. The syntax is
Piecewise e
1
, c
1
, , e
n
, c
n
which outputs e
1
if c
1
is true, e
2
if c
2
is true, , e
n
if c
n
is
true, and 0 otherwise (the default).
So, for example, here is the definition for the absolute value function given as a
piecewise object.
In[40]:= abspw x_ : Piecewise x, x 0 , x, x 0
Piecewise objects display as you would expect in traditional mathematical notation.
In[41]:= abspw x
Out[41]=
x x 0
x x 0
Furthermore, Piecewise is fully integrated with the algebraic, symbolic, and graphical
functions in Mathematica and so is preferable to other approaches.
In[42]:= Integrate abspw x , x, 1, 1
Out[42]= 1
In[43]:= D abspw x , x
Out[43]=
1 x 0
1 x 0
Indeterminate True
In[44]:= Plot abspw x , x, 2, 2 ;
2 1 1 2
0.5
1
1.5
2
Notice that the definition of the absolute value function given in terms of condition-
als is not fully supported by many of the built-in functions.
In[45]:= Clear abs
In[46]:= abs x_ : x ; x 0
138 An Introduction to Programming with Mathematica
In[47]:= abs x_ : x ; x 0
In[48]:= Integrate abs x , x, 1, 1
Out[48]=
1
1
abs x x
In[49]:= D abs x , x
Out[49]= abs x
Argument checking
Often, when we write functions, we know ahead of time that the definitions we give them
are valid only for certain kinds of inputs. For example, the following definition for the
factorial function only makes sense for positive integers.
In[50]:= fact 0 1;
fact n_ : n fact n 1
In[52]:= fact 5
Out[52]= 120
If we were to give fact an argument that was not a positive integer, the recursion could
run away from us.
In[53]:= fact 3.4
$RecursionLimit::reclim :
Recursion depth of 256 exceeded. More
Out[53]= 2.729671867921455 10
494
Hold fact 250.6 1
Conditionals are a convenient way of checking that the arguments to our functions
pass some criteria. For example, there are several ways that we could make the fact
function valid only under the condition that its argument is a positive integer. Here is how
we might approach it using the If construct to test that n passes the appropriate criteria.
In[54]:= Clear fact
In[55]:= fact 0 1;
In[56]:= fact n_ : If IntegerQ n && n 0, n fact n 1
In[57]:= fact 5 , fact 3 , fact 2.4
Out[57]= 120, Null, Null
5 Procedural programming 139
We see that the function works fine for positive integers, but since we did not give an
alternative condition to the If function, nothing is returned (technically Null is returned)
when the test condition fails.
Let us define a message that will be output in the case that the argument to fact
fails the positive integer test.
In[58]:= fact::noint "Argument `1` is not a positive integer.";
We then use Message as the third argument to our If, so that when the condition
fails the message will be triggered. Essentially Message messname, e
1
, e
2
, prints
using StringForm messg, e
1
, e
2
, , where messg is the value of the message name
and the e
i
are substituted in for any expressions of the form `i`. In the above example, the
message name is noint and its value is the string beginning with "Argument...". In
this example, the value of n will be substituted into the string where the `1` occurs.
In[59]:= fact n_ : If IntegerQ n && n 0,
n fact n 1 ,
Message fact::noint, n
In[60]:= fact 3
fact::noint : Argument 3 is not a positive integer.
Of course, there are a variety of ways of using conditionals to do argument checking.
Here are three more implementations, without the messaging.
In[61]:= fact1 0 1;
fact1 n_ : n fact1 n 1 ; IntegerQ n && n 0
In[63]:= fact1 5 , fact1 2.4
Out[63]= 120, fact1 2.4
In[64]:= fact2 0 1;
fact2 n_ ; IntegerQ n && n 0 : n fact2 n 1
In[66]:= fact2 5 , fact2 2.4
Out[66]= 120, fact2 2.4
In[67]:= fact3 0 1;
fact3 n_?IntegerQ ; n 0 : n fact3 n 1
In[69]:= fact3 5 , fact3 2.4
Out[69]= 120, fact3 2.4
140 An Introduction to Programming with Mathematica
Summary
When writing a function whose result must be computed differently, depending upon the
values of its arguments, you have a choice:
1. Use a multiclause definition, where the conditions are optional, and may appear
after the right-hand sides.
f pattern
1
_ /; cond
1
:= rhs
1
f pattern
n
_ /; cond
n
:= rhs
n
2. Use a single-clause definition with a conditional expression.
f[x_] := If cond
1
, rhs
1
,
If cond
n
, rhs
n
,
rhs
n 1
]
In the latter case, if n is greater than two, use the equivalent Which expression; and if all
conditions have the form x == const
i
, for a given variable x and some constants const
i
, use
the Switch function.
The next section contains several applications that use various combinations of the
procedural constructs we have learned in this chapter.
Exercises
1. Write the function signum[x] which, when applied to an integer x, returns 1, 0,
or 1, according as x is less than, equal to, or greater than, 0. Write it in three ways:
using three clauses, using a single clause with If, and using a single clause with
Which.
2. Extend signum from Exercise 1 to apply to both integers and reals; again, write it in
three ways (though you may use more than three clauses for the multiclause version).
3. Write applyChar in multiclause form, using pattern matching on the first element
of its argument.
5 Procedural programming 141
4. Use If in conjunction with Map or Fold to define the following functions:
a. In a list of numbers, double all the positive numbers, but leave the negative
numbers alone.
b. remove3Repetitions is like removeRepetitions except that it only alters
three or more consecutive occurrences, changing them to two occurrences; if
there are only two occurrences to begin with, they are left alone. For example,
remove3Repetitions[{0,1,1,2,2,2,1}] will return {0,1,1,2,2,1}.
c. Add the elements of a list in consecutive order, but never let the sum go below 0.
In[1]:= positiveSum 5, 3, 13, 7, 3, 2
Out[1]= 6
Since the 13 caused the sum to go below 0, it was instead put back to 0 and the
summation continued from there.
5. Using NestWhileList, write a function CollatzSequence[n] that produces
the Collatz sequence for any positive integer n. The Collatz sequence is generated as
follows: starting with a number n, if it is even, then output
n
2
; if n is odd, then output
3 n 1. Iterate this process while n 1.
5.4 Examples
Sieve of Eratosthenes
One of the oldest algorithms in the history of computing is the Sieve of Eratosthenes.
Named after the famous Greek astronomer Eratosthenes (ca. 276 ca. 194 BC), this
method is used to find all prime numbers below a given number n. The great feature of
this algorithm is that it finds prime numbers without doing any divisions an operation
that took considerable skill and concentration before the introduction of the Arabic
numeral system. In fact, its only operations are addition and component assignment.
The algorithm can be summarized as follows: to find all the prime numbers less than
an integer n:
create a list of the integers 1 through n
starting with p 2, cross out all multiples of p
142 An Introduction to Programming with Mathematica
increment p (that is, add 1 to p) and cross out all multiples of p
repeat the previous two steps until p n .
You should convince yourself that the numbers that are left after all the crossings out
are in fact the primes less than n. This algorithm lends itself very well to a procedural
approach, so let us walk through the steps.
We will use a For structure for this problem. The syntax is For[start, test, incr,
body], where start will first be evaluated (initializing values), and then incr and body will be
repeatedly evaluated until test fails.
1. Let lis be a list containing all the integers between 1 and n.
In[1]:= n 20;
lis Range n
Out[2]= 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,
11, 12, 13, 14, 15, 16, 17, 18, 19, 20
2. Let p 2. Repeat the following two steps:
Starting at position 2p, cross out every pth value in lis. We will assign 1 to
lis at positions 2p, 3p, and the 1 will represent a crossed out value.
In[3]:= p 2;
Do lis i 1, i, 2 p, n, p
In[5]:= lis
Out[5]= 1, 2, 3, 1, 5, 1, 7, 1, 9, 1, 11, 1, 13, 1, 15, 1, 17, 1, 19, 1
While p n , increment p by 1, until lis[[p]] is not 1, or until p n 1.
In[6]:= n 20;
lis Range n ;
For p 2,
p 1 && p Floor Sqrt n ,
p ,
Do lis i 1, i, 2 p, n, p
3. The non-1s in lis are all the prime numbers less than or equal to n.
In[9]:= DeleteCases lis, 1
Out[9]= 2, 3, 5, 7, 11, 13, 17, 19
5 Procedural programming 143
Let us put these steps together in our function Sieve.
In[10]:= Clear n, p, lis
In[11]:= Sieve n_Integer : Module lis Range n , p ,
For p 2,
p 1 && p Floor Sqrt n ,
p ,
Do lis i 1, i, 2 p, n, p ;
DeleteCases lis, 1
Here are a few simple tests to check the correctness of our function. First we check that
Sieve produces the correct number of primes less than a large integer.
In[12]:= Length Sieve 10
5
Out[12]= 9592
The built-in PrimePi[x] gives the number of primes x less than or equal to x.
In[13]:= PrimePi 10
5
Out[13]= 9592
Next we do some simple timing tests to check the efficiency of this algorithm against the
built-in functions that are optimized for this task.
In[14]:= Sieve 10
6
; Timing
Out[14]= 13.62 Second, Null
In[15]:= Timing Table Prime i , i, 10
6
;
Out[15]= 5.648 Second, Null
In[16]:= Timing Map Prime, Range 10
6
;
Out[16]= 5.628 Second, Null
For numbers in this range (less than about 10
6
), sieving is fairly efficient its speed is
within an order of magnitude of the built-in algorithms. But, beyond this range, it does
tend to bog down and it would be best to consider specialized algorithms that are asymptoti
cally fast (for large integers, PrimePi uses an algorithm due to Lagarias, Miller, and
Odlyzko that is based on estimates of the density of primes).
144 An Introduction to Programming with Mathematica
Classifying points
Quadrants in the Euclidean plane are conventionally numbered counterclockwise from
quadrant 1 (x and y positive) to quadrant 4 (x positive, y negative). The function point
Loc[{x,y}] will compute the classification of point x, y , according to Table 5.1.
Point Classification
0, 0 0
y 0 on the x axis 1
x 0 on the y axis 2
Quadrant 1 1
Quadrant 2 2
Quadrant 3 3
Quadrant 4 4
Table 5.1: Quadrant classification
We will use this problem to illustrate the features covered in this chapter, by giving a
number of different solutions, using multiclause function definitions with predicates,
single-clause definitions with If and its relatives, and combinations of the two.
Perhaps the first solution that suggests itself is one that uses a clause for each of the
cases above.
In[17]:= pointLoc 0, 0 : 0
pointLoc x_, 0 : 1
pointLoc 0, y_ : 2
pointLoc x_, y_ : 1 ; x 0 && y 0
pointLoc x_, y_ : 2 ; x 0 && y 0
pointLoc x_, y_ : 3 ; x 0 && y 0
pointLoc x_, y_ : 4 ; x 0 && y 0
It is a good idea to include the last condition as a comment, rather than as a condi-
tion in the code, because Mathematica would not realize that the condition has to be true at
that point and would check it anyway.
We will use the following list of points as our test cases.
In[24]:= pts
0, 0 , 4, 0 , 0, 1.3 , 2, 4 , 2, 4 , 2, 4 , 2, 4 ;
In[25]:= Map pointLoc, pts
Out[25]= 0, 1, 2, 1, 2, 3, 4
5 Procedural programming 145
Translated directly to a one-clause definition using If, this becomes:
In[26]:= pointLoc x_, y_ :
If x 0 && y 0, 0,
If y 0, 1,
If x 0, 2,
If x 0 && y 0, 1,
If x 0 && y 0, 2,
If x 0 && y 0, 3, 4
In[27]:= Map pointLoc, pts
Out[27]= 0, 1, 2, 1, 2, 3, 4
Actually, a more likely solution here uses Which.
In[28]:= pointLoc x_, y_ : Which
x 0 && y 0, 0,
y 0, 1,
x 0, 2,
x 0 && y 0, 1,
x 0 && y 0, 2,
x 0 && y 0, 3,
True x 0&&y 0 , 4
In[29]:= Map pointLoc, pts
Out[29]= 0, 1, 2, 1, 2, 3, 4
In[30]:= pointLoc 5, 9
Out[30]= 3
All of our solutions so far suffer from a certain degree of inefficiency, because of
repeated comparisons of a single value with 0. Take the last solution as an example, and
suppose the argument is ( 5, 9). It will require five comparisons of 5 with 0 and three
comparisons of 9 with 0 to obtain this result. Specifically:
1. evaluate x==0; since it is false, the associated y== 0 will not be evaluated, and
we next
2. evaluate y==0 on the following line; since it is false,
3. evaluate x==0 on the third line; since it is false,
4. evaluate x>0 on next line; since it is false, the associated y> 0 will not be evalu-
ated, and we next,
5. evaluate x<0 on the next line; since it is true, we do,
146 An Introduction to Programming with Mathematica
6. the y>0 comparison, which is false, so we next,
7. evaluate x<0 on the next line; since it is true, we then evaluate y< 0, which is
also true, so we return the answer 3.
How can we improve this? By nesting conditional expressions inside other condi-
tional expressions. In particular, as soon as we discover that x is less than, greater than, or
equal to 0, we should make maximum use of that fact without rechecking it. That is what
the following pointLoc function does.
In[31]:= pointLoc x_, y_ :
Which x 0, If y 0, 0, 2 ,
x 0, Which y 0, 1,
y 0, 4,
True y 0 , 1 ,
True, x 0
Which y 0, 3,
y 0, 2,
True y 0 , 1
Let us count up the comparisons for 5, 9 this time: (i) evaluate x ==0; since it is
false, we next, (ii) evaluate x> 0; since it is false, we go to the third branch of the Which,
evaluate True, which is, of course, true; then, (iii) evaluate y<0, which is true, and we
return 3. Thus, we made only three comparisons a substantial improvement.
When pattern matching is used, as in our first, multiclause solution, efficiency
calculations are more difficult. It would be inaccurate to say that Mathematica has to
compare x and y to 0 to tell whether the first clause applies; what actually happens is more
complex. What is true, however, is that it will do the comparisons indicated in the last four
clauses. So, even if we discount the first three clauses with argument 5, 9 , some extra
comparisons are done. Specifically: (i) the comparison x> 0 is done; then, (ii) x <0 and (iii)
y >0; then, (iv) x <0 and (v) y< 0. This can be avoided by using conditional expressions
within clauses.
In[32]:= pointLoc 0, 0 : 0
pointLoc x_, 0 : 1
pointLoc 0, y_ : 2
pointLoc x_, y_ : If x 0, 2, 1 ; y 0
pointLoc x_, y_ : If x 0, 3, 4 ; y 0
Now, no redundant comparisons are done. For 5, 9 , since y >0 fails, the fourth
clause is not used, so the x> 0 comparison in it is not done. Only the single x <0 compari-
son in the final clause is done, for a total of two comparisons.
5 Procedural programming 147
Having done all these versions of pointLoc, we would be remiss if we did not
remind the reader of a basic fact of life in programming: your time is more valuable than
your computers time. You should not be worrying about how slow a function is until there
is a demonstrated need to worry. Far more important is the clarity and simplicity of the
code, since this will determine how much time you (or another programmer) will have to
spend when it comes time to modify it. In the case of pointLoc, we would argue that we
got lucky and found a version (the final one) that wins on both counts (if only program-
ming were always like that!).
Finally, a technical, but potentially important, point: Not all of the versions of
pointLoc work exactly the same. The integer 0, as a pattern, does not match the real
number 0.0, since they have different heads. Thus, using the last version as an example,
pointLoc[{0.0,0.0}] returns 4.
In[37]:= pointLoc 0.0, 0.0
Out[37]= 4
See Section 6.2 for a discussion of alternatives, which allows us to efficiently deal
with these various cases.
Exercises
1. Using an If function, write a function gcd[m,n] that implements the Euclidean
algorithm (see Exercise 10 of Section 5.2) for finding the greatest common divisor of
m and n.
2. Use Piecewise to define the pointLoc function given in this section.
3. Extend pointLoc to three dimensions, following this rule: for point (x, y, z), if
z 0, then give the same classification as (x, y), with the exception that zero is treated
as a positive number (so the only classifications are 1, 2, 3, and 4); if z 0, add 4 to
the classification of (x, y) (with the same exception). For example, (1, 0, 1) is in octant
1, and (0, 3, 3) is in octant 8. pointLoc should work for points in two or three
dimensions.
148 An Introduction to Programming with Mathematica
6 Rule-based programming
The use of rules to transform expressions from one form to another is one of the most
powerful and useful tools available in the Mathematica programming language. The
thousands of rules built in to Mathematica can be expanded limitlessly through the
creation of user-defined rules. Rules can be created to change the form of expressions,
to filter data based on some criteria, and can be set up to apply to broad classes of
expressions or limited to certain narrow domains through the use of appropriate
pattern matching techniques. These rules can perform many of the tasks normally
associated with more traditional programming constructs, such as we have discussed in
the chapters on procedural and functional programming. In this chapter we will
discuss the structure and application of rules to common programming tasks and look
at their application in some concrete examples.
6.1 Introduction
Users of Mathematica typically first encounter rules as the output to many built-in func-
tions. For example, the Solve function returns its solutions as a list of rules.
In[1]:= soln Solve a x
2
b x c 0, x
Out[1]= x
b b
2
4 a c
2 a
, x
b b
2
4 a c
2 a
They are also used to specify options for functions and replacement rules in many kinds of
computations.
In[2]:= FactorInteger 5, GaussianIntegers True
Out[2]= , 1 , 1 2 , 1 , 2 , 1
In[3]:= StringReplace "acgttttccctgagcataaaaacccagcaatacg",
"ca" "CA", "tt" "TT"
Out[3]= acgTTTTccctgagCAtaaaaaccCAgCAatacg
When you define a function via an assignment such as the function f below, you are
defining a rule that says whenever f is given an argument, it should be replaced with that
argument squared. This rule will be applied automatically whenever you evaluate
f[anything].
In[4]:= f x_ : x
2
In[5]:= f bob
Out[5]= bob
2
On the other hand, you can set up rules to be applied on demand by using the
replacement operator ReplaceAll, written in shorthand notation as /. . These rules can
then be used to transform one expression into another. For example, the following rule is
used to extract the real and imaginary parts of a complex number and convert it to an
ordered pair.
In[6]:= 3 4 . Complex a_, b_ a, b
Out[6]= 3, 4
This rule reverses the elements in each ordered pair.
In[7]:= , 1 , , 2 , , 3 . x_, y_ y, x
Out[7]= 1, , 2, , 3,
And here is a rule that turns each of the superscripts in the polynomial below into a
subscript.
In[8]:= poly Factor 1 x
11
Out[8]= 1 x 1 x x
2
x
3
x
4
x
5
x
6
x
7
x
8
x
9
x
10
In[9]:= ToBoxes poly . SuperscriptBox SubscriptBox DisplayForm
Out[9]//DisplayForm=
1 x 1 x x
2
x
3
x
4
x
5
x
6
x
7
x
8
x
9
x
10
Rule-based programming is such a useful construct for manipulating lists and arbi-
trary expressions that no user of Mathematica should be without a working knowledge of
this paradigm. This chapter gives a thorough introduction to pattern matching and then
proceeds to rule-based programs, many of which were introduced earlier as functional or
procedural programs.
150 An Introduction to Programming with Mathematica
6.2 Patterns
Blanks
When you make an assignment to a symbol, like x=4, you are making a rule that should be
applied to the literal expression x. Loosely speaking, the rule says, replace x with the value
4 whenever x is encountered. We have seen that you can also define functions of one or
more arguments that allow you to substitute arbitrary expressions for those arguments.
In[1]:= f x_ : x 1
The left-hand side of the above assignment is a pattern. It contains a blank (underscore)
which can stand for any expression, not just the literal expression x.
In[2]:= f
Out[2]= 1
In[3]:= f bob
Out[3]= 1 bob
While any specific expression can be pattern matched (because any object must
match itself), we usually want to be able to pattern match large classes of expressions (for
example, a sequence of expressions or expressions having Integer as the head). For this
purpose, patterns are defined as expressions that may contain blanks. That is to say, a
pattern may contain one of the following: a single (_) blank, a double (__) blank, or a triple
(___) blank.
We will find it useful to identify the pattern to which an expression is matched (for
example, on the left-hand side of a function definition) so that it can be referred to by
name elsewhere (for example, on the right-hand side of the function definition). A pattern
can be labeled by name_, or name__, or name___ (which can be read as a pattern called
name) and the labeled pattern will be matched by the same expression that matches its
unlabeled counterpart. The matching expression is given the name used in the labeled
pattern.
You can see what class of expressions match a given pattern by using MatchQ. For
example, this tests whether the symbol bob matches any expression because the single
underscore can stand for any Mathematica expression.
In[4]:= MatchQ bob, _
Out[4]= True
6 Rule-based programming 151
This tests whether the number 3.14 matches any expression with head Real.
In[5]:= MatchQ 3.14, _Real
Out[5]= True
Of course 3.14 does not match any expression with head Integer.
In[6]:= MatchQ 3.14, _Integer
Out[6]= False
If you want to look at a list of expressions and see which ones match a particular
pattern, you can use Cases. Cases[expr,patt] outputs those elements of expr that match
the pattern patt. For example, the only two elements of the list below that have head
Integer are 3 and 17. Notice the fourth element is a string.
In[7]:= Cases 3, 3.14, 17, "3", 4 5 I , _Integer
Out[7]= 3, 17
In[8]:= Cases 3, 3.14, 17, "3", 4 5 I , _String
Out[8]= 3
Remember that the OutputForm of strings is to display without the quote characters. If
you want to check the structure of this last output, use FullForm or check its Head.
In[9]:= FullForm %
Out[9]//FullForm=
List "3"
Here are some additional examples of pattern matching. This next example matches
all those expressions with head g.
In[10]:= Cases g x , f x , g h x , g a, 0 , _g
Out[10]= g x , g h x , g a, 0
In the following example, the pattern {p_,q_} matches any list with two elements.
In[11]:= Cases a, b , , 1, 0 , c, d, 3 , p_, q_
Out[11]= a, b , 1, 0
Let us clear symbols we no longer need.
In[12]:= Clear f
152 An Introduction to Programming with Mathematica
Sequence pattern matching
A sequence consists of a number of expressions separated by commas. For example, the
arguments of expressions are written as sequences.
A double blank (BlankSequence) represents a sequence of one or more expres-
sions and __h represents a sequence of one or more expressions, each of which has head h.
An expression that matches a blank will also match a double blank.
A triple blank (BlankNullSequence) represents a sequence of zero or more
expressions and ___h represents a sequence of zero or more expressions, each of which has
head h. An expression that matches a blank will also match a triple blank and a sequence
that matches a double blank pattern will also match a triple blank pattern.
The pattern {p__}, using two _ characters, matches any list containing one or more
elements.
In[13]:= Cases a, b , , 1, 0 , c, d, 3 , p__
Out[13]= a, b , 1, 0 , c, d, 3
The pattern {p___}, using three _ characters, matches any list containing zero or more
elements.
In[14]:= Cases a, b , , 1, 0 , c, d, 3 , p___
Out[14]= a, b , , 1, 0 , c, d, 3
A list {a,b,c} is matched by the pattern _ (using Blank), as well as by List[__]
(using BlankSequence) and List[___] (with BlankNullSequence). However, the
list {a,b,c} is not matched by the pattern List[_] (a list of one expression) because for
the purposes of pattern matching, a sequence is not an expression.
In[15]:= MatchQ a, b, c , _
Out[15]= True
In[16]:= MatchQ a, b, c , _
Out[16]= False
Here are some other examples of successful pattern matches.
In[17]:= MatchQ a, b, c , __
Out[17]= True
In[18]:= MatchQ a, b, c , ___
Out[18]= True
6 Rule-based programming 153
In[19]:= MatchQ a, b, c , x__
Out[19]= True
In[20]:= MatchQ a, b, c , x___
Out[20]= True
In the last two examples above, the labels on the blanks do not affect the success or
failure of the pattern match.
In[21]:= MatchQ a, b, c , __
Out[21]= True
The labels simply serve to identify different parts of the expression. For example, in
MatchQ[{a,b,c},x_], x names the list {a,b,c}, but in Match
Q[{a,b,c},{x___}], x names the sequence a,b,c which is quite different. This is
illustrated further in the section on conditional pattern matching.
Finally, note that the discussion about lists here applies equally to any function. For
example, the following returns True, with x naming the sequence a,b,c.
In[22]:= MatchQ Plus a, b, c , Plus x__
Out[22]= True
Example: Finding subsequences
As an example of sequence pattern matching, consider the problem of finding a particular
subsequence within a sequence of numbers. To simplify this problem, consider both the
sequence and the subsequence to be given as lists of numbers. As a concrete example, we
will find the positions at which the subsequence 3238 occurs in the digits of .
Here are the digits of . Initially, we will look at only 50 digits so we can easily
inspect the progress of our program.
In[23]:= pidigs First RealDigits N , 50 3
Out[23]= 1, 4, 1, 5, 9, 2, 6, 5, 3, 5, 8, 9, 7, 9, 3,
2, 3, 8, 4, 6, 2, 6, 4, 3, 3, 8, 3, 2, 7, 9, 5, 0,
2, 8, 8, 4, 1, 9, 7, 1, 6, 9, 3, 9, 9, 3, 7, 5, 1
Here is our subsequence, given as a list of digits.
In[24]:= subseq 3, 2, 3, 8 ;
154 An Introduction to Programming with Mathematica
One approach to this problem is to partition the list of digits in pidigs into lists of
the same length as the list subseq, with overlapping sublists of offset 1. This means that
we will examine all length 4 sublists from pidigs.
In[25]:= p Partition pidigs, Length subseq , 1
Out[25]= 1, 4, 1, 5 , 4, 1, 5, 9 , 1, 5, 9, 2 ,
5, 9, 2, 6 , 9, 2, 6, 5 , 2, 6, 5, 3 ,
6, 5, 3, 5 , 5, 3, 5, 8 , 3, 5, 8, 9 , 5, 8, 9, 7 ,
8, 9, 7, 9 , 9, 7, 9, 3 , 7, 9, 3, 2 , 9, 3, 2, 3 ,
3, 2, 3, 8 , 2, 3, 8, 4 , 3, 8, 4, 6 , 8, 4, 6, 2 ,
4, 6, 2, 6 , 6, 2, 6, 4 , 2, 6, 4, 3 , 6, 4, 3, 3 ,
4, 3, 3, 8 , 3, 3, 8, 3 , 3, 8, 3, 2 , 8, 3, 2, 7 ,
3, 2, 7, 9 , 2, 7, 9, 5 , 7, 9, 5, 0 , 9, 5, 0, 2 ,
5, 0, 2, 8 , 0, 2, 8, 8 , 2, 8, 8, 4 , 8, 8, 4, 1 ,
8, 4, 1, 9 , 4, 1, 9, 7 , 1, 9, 7, 1 , 9, 7, 1, 6 ,
7, 1, 6, 9 , 1, 6, 9, 3 , 6, 9, 3, 9 , 9, 3, 9, 9 ,
3, 9, 9, 3 , 9, 9, 3, 7 , 9, 3, 7, 5 , 3, 7, 5, 1
Now we are ready for the pattern match. From the list p above, we are looking for
the positions of any sublist that matches {3,2,3,8}. The Position function takes as
its first argument, the expression from which we are trying to match. The second argu-
ment is the pattern to match. We will use BlankNullSequence (___) on either side of
our subsequence because zero or one or two expressions may occur before or after it in p.
In[26]:= Position p, Flatten ___, subseq, ___
Out[26]= 15
So the subsequence 3238 occurs starting at the 15th digit in the sequence given by
pidigs.
Finally, let us turn this into a function and test it on a much larger example. Note
that we use the pattern _List on both arguments so that FindSubsequence will only
match arguments that have head List. (In Exercise 5 at the end of this section, you are
asked to create a version of FindSubsequence that takes numbers instead of lists as its
arguments.)
In[27]:= FindSubsequence lis_List, subseq_List : Module p ,
p Partition lis, Length subseq , 1 ;
Position p, Flatten ___, subseq, ___
We store the first 100,000 digits of in the symbol pidigs.
In[28]:= pidigs First RealDigits N , 10
5
3 ;
6 Rule-based programming 155
We find that the subsequence {3,2,3,8} occurs at the following nine different positions
in the first 100,000 digits of .
In[29]:= FindSubsequence pidigs, 3, 2, 3, 8
Out[29]= 15 , 8990 , 20522 , 20756 ,
28130 , 41865 , 57208 , 86505 , 91936
The subsequence 31415 occurs once in the first 100,000 digits of starting at the
88,008th digit.
In[30]:= FindSubsequence pidigs, 3, 1, 4, 1, 5
Out[30]= 88008
Conditional pattern matching
Attaching a predicate
In addition to specifying the head of an expression, you can also match expressions against
predicate functions. If the blanks of a pattern are followed with ?test, where test is a predi-
cate, then a match is only possible if test returns True when applied to the entire
expression.
So, to match a class of expressions that have head h, you use _h. To match a class of
expressions that evaluate to True when the predicate pred is applied, use _?pred.
In[31]:= MatchQ 1, 2, 3 , _?ListQ
Out[31]= True
In[32]:= MatchQ 1, 2, 3 , _?NumberQ
Out[32]= False
Note that in the above example, even though the list {1,2,3} consists of numbers, it
does not match ?NumberQ because its head (List) does not pass the NumberQ test.
The pattern _?Negative matches any expression that passes the Negative test;
that is, it returns true when Negative is applied to it.
In[33]:= Cases 2, 7, 1.2, 0, 5 2 I , _?Negative
Out[33]= 2, 1.2
The following examples use a pure predicate function. In the first example, we are
asking if {a,b,c} has head List and if the length of {a,b,c} is greater than 2. Since it
passes both of these conditions, MatchQ returns True.
156 An Introduction to Programming with Mathematica
In[34]:= MatchQ a, b, c , _List? Length # 2 &
Out[34]= True
Even though the head of {a,b,c} is List, the condition below fails since the list has
length less than 4.
In[35]:= MatchQ a, b, c , _List? Length # 4 &
Out[35]= False
Note that when using a pure function in ?test, because of the precedence Mathemat-
ica gives to evaluating various quantities, it is necessary to enclose the entire function,
including the &, in parentheses. We have used test to place a constraint on the entire
expression.
Here is a simple application of attaching a predicate. This definition of the Fibonacci
function tests its argument to see that it is an integer (specifically, this tests that the head of
n is Integer).
In[36]:= f 1 f 2 1;
In[37]:= f n_?IntegerQ : f n 1 f n 2
Because of the predicate, f will not evaluate for noninteger arguments.
In[38]:= f 1.2
Out[38]= f 1.2
In[39]:= f 5 , f 10 , f 15
Out[39]= 5, 55, 610
We could also check that the arguments to f are both integral and positive.
In[40]:= Clear f
In[41]:= f 1 f 2 1;
In[42]:= f n_? IntegerQ && Positive : f n 1 f n 2
In[43]:= f 3
Out[43]= f 3
6 Rule-based programming 157
Attaching a condition
If part of a labeled pattern is followed with an expression such as /;condition, where
condition contains labels appearing in the pattern, then a match is possible only if condition
returns True. We use condition to place a constraint on the labeled parts of an expression.
The use of labels in condition is useful for narrowing the scope of a pattern match.
In[44]:= MatchQ[x^2, _^y_ /; EvenQ[y]]
Out[44]= True
In[45]:= MatchQ[x^2, _^y_ /; OddQ[y]]
Out[45]= False
We mentioned above that matching a list like {a,b,c} with the pattern x_ is
different from matching it with x___ because of the various expressions that are associated
with x.
In[46]:= MatchQ 4, 6, 8 , x_ ; Length x 4
Out[46]= False
In[47]:= MatchQ 4, 6, 8 , x___ ; Length x 4
Length::argx : Length called with
3 arguments; 1 argument is expected. More
Out[47]= False
In[48]:= MatchQ 4, 6, 8 , x___ ; Plus x 10
Out[48]= True
In the first example, x was associated with the entire list {4,6,8}; since
Length[{4,6,8}] is not greater than 4, the match failed. In the second example, x
became the sequence 4,6,8 so that the condition was Length[4,6,8]>4; but Length
can only have one argument, hence the error. In the last example, x was again associated
with 4,6,8, but now the condition was Plus[4,6,8]>10, which is perfectly legal, and
true.
In the following example, the pattern matches all those expressions that are between
2 and 5.
In[49]:= Cases 1, 2, 3, 4, 5, 6, 7, 8 , x_ ; 2 x 5
Out[49]= 3, 4
158 An Introduction to Programming with Mathematica
Let us try to recast the Fibonacci function example from the previous section in
terms of a conditional.
In[50]:= Clear f
In[51]:= f 1 f 2 1;
In[52]:= f n_ : f n 1 f n 2 ; IntegerQ n
Because of the predicate, f does not evaluate for noninteger arguments.
In[53]:= f 1.2
Out[53]= f 1.2
In[54]:= f 5 , f 10 , f 15
Out[54]= 5, 55, 610
Similarly, we can check that the arguments to f are both integral and positive.
In[55]:= Clear f
In[56]:= f 1 f 2 1;
In[57]:= f n_ : f n 1 f n 2 ; IntegerQ n && Positive n
In[58]:= f 3 , f 10
Out[58]= f 3 , 55
Note that you can alternatively put the condition inside the left-hand side of your
definition.
In[59]:= Clear f
In[60]:= f 1 f 2 1;
In[61]:= f n_ ; IntegerQ n && Positive n : f n 1 f n 2
In[62]:= f 15 , f 1.4 , f 4
Out[62]= 610, f 1.4 , f 4
Alternatives
A final type of pattern uses alternatives. Alternatives are denoted p
1
p
2
p
n
where the p
i
are independent patterns. This pattern will match an expression whenever any one of those
independent patterns match it.
6 Rule-based programming 159
In the following example, x^2 matches an expression which is either the symbol x
raised to a real number or the symbol x raised to an integer.
In[63]:= MatchQ[x^2, x^_Real | x^_Integer]
Out[63]= True
In this example, x^2 matches x raised to an expression which is either a real num-
ber or an integer.
In[64]:= MatchQ[x^2, x^(_Real | _Integer)]
Out[64]= True
Here the pattern matches any expression that has head Integer or Rational or
Real.
In[65]:= Cases 1, 3.1,
2
3
, x, 3 4 I, "Hello" ,
_Integer _Rational _Real
Out[65]= 1, 3.1,
2
3
As a final example, recall the function pointLoc from Section 5.4.
In[66]:= pointLoc 0, 0 : 0
pointLoc x_, 0 : 1
pointLoc 0, y_ : 2
pointLoc x_, y_ : If x 0, 2, 1 ; y 0
pointLoc x_, y_ : If x 0, 3, 4
The integer 0, as a pattern, does not match the real number 0.0, since they have different
heads.
In[71]:= Head 0 , Head 0.0
Out[71]= Integer, Real
Thus, using the above version of pointLoc, {0.0,0.0} returns 4, which is, of course,
wrong.
In[72]:= pointLoc 0.0, 0.0
Out[72]= 4
On the other hand, the single-clause versions using If and Which returned 0, because
0.0== 0 is true. How can we fix this? There are a number of possibilities. Perhaps the
simplest way is to change the rules involving zeroes by means of alternatives.
In[73]:= Clear pointLoc
160 An Introduction to Programming with Mathematica
In[74]:= pointLoc 0 0.0, 0 0.0 : 0
pointLoc x_, 0 0.0 : 1
pointLoc 0 0.0, y_ : 2
pointLoc x_, y_ : If x 0, 2, 1 ; y 0
pointLoc x_, y_ : If x 0, 3, 4
Now the several cases that led to inconsistencies in the previous versions are dealt with
properly.
In[79]:= pointLoc 0, 0.0
Out[79]= 0
In[80]:= pointLoc 1, 0
Out[80]= 1
String patterns
All of the pattern matching discussed in the previous sections extends to strings in a
very powerful manner. You might find it helpful to think of strings as a sequence of
characters and use the same general principles on these expressions as you do with lists.
Let us look at a few examples to try and make this concrete.
The expression {a,b,c,c,d,e} matches the pattern {__,s_,s_,__} because it
is a list that starts with a sequence of one or more elements, it contains an element
repeated once, and then ends with a sequence of one or more elements.
In[81]:= MatchQ a, b, b, c, d, e , __, s_, s_, __
Out[81]= True
If we now use a string instead of a list and StringMatchQ instead of MatchQ, we
get a similar result using the shorthand notation ~~ for StringExpression, which
essentially concatenates strings.
In[82]:= StringMatchQ "abbcde", __ s_ s_ __
Out[82]= True
In[83]:= "a" "b"
Out[83]= ab
In[84]:= FullForm HoldForm "a" "b"
Out[84]//FullForm=
HoldForm StringExpression "a", "b"
6 Rule-based programming 161
StringExpression is quite similar to StringJoin (both can be used to concatenate
strings) except that with StringExpression, you can concatenate nonstrings.
The next example also shows the similarity between the pattern matching that we
explored earlier and string patterns. Using Cases, we return all those expressions that
match the pattern _Symbol; that is, we pick out all those symbols from the list.
In[85]:= Cases 1, f, g, 6, x, t, 2, 5 , _Symbol
Out[85]= f, g, x, t
With the string "1fg6xt25" we can use StringCases whose second argument is a
pattern that represents a class of characters to match. For example, LetterCharacter
matches a single letter.
In[86]:= StringCases "1fg6xt25", LetterCharacter
Out[86]= f, g, x, t
You can match single digits with DigitCharacter.
In[87]:= StringCases "1fg6xt25", DigitCharacter
Out[87]= 1, 6, 2, 5
Starting in Version 5.1, you can use regular expressions to match string patterns.
Regular expressions in Mathematica follow a syntax very close to that of the Perl program-
ming language. This syntax is quite compact and powerful but it comes at the cost of
readability regular expressions tend to be quite cryptic to humans. As a result, we will
only cover a few examples of their use here and refer the interested reader to the Mathemat-
ica documentation on string patterns.
The regular expression 1.* will be matched by any string starting with 1, followed
by any character repeated zero or more times.
In[88]:= StringMatchQ "1a2b3c4d", RegularExpression "1. "
Out[88]= True
The regular expression \\d represents any digit 0 through 9.
In[89]:= StringCases "1a2b3c4d", RegularExpression "\\d"
Out[89]= 1, 2, 3, 4
In the following example, we use a regular expression to look for the pattern consist-
ing of the character "a" repeated one or more times, followed by the character "c",
followed by any character. The StringReplace function then replaces any expression
matching this pattern with a large, bold formatted expression. The "$0" is used to refer to
the matched pattern.
162 An Introduction to Programming with Mathematica
In[90]:= StringReplace "acgttttccctgagcataaaaacccagcaatacg",
RegularExpression "a..c." "\ \ \ StyleBox
\"$0\",FontSize 14,FontWeight \"Bold\" \ "
Out[90]= acgttttccctgagcataaaaacccagcaatacg
Exercises
1. Find as many patterns as possible that match the expression x^3 + y z.
2. Find as many pattern matches as possible for the following expression.
{5, erina, {}, "give me a break"}
3. Using both forms (predicate and condition), write down five conditional patterns
that match the expression {4,{a,b},"g"}.
4. In Exercise 10 of Section 5.2, we developed a procedural implementation of the
Euclidean algorithm for finding the greatest common divisor of two numbers. The
function given in the solutions does no argument checking and hence can give
erroneous output for arguments that are not integers. Rewrite the gcd function
given there so that it uses pattern matching to check that each of its two arguments
are integers.
5. The function FindSubsequence defined in this section suffers from the limitation
that the arguments lis and subseq must both be lists of numbers. Write another
definition of FindSubsequence that takes integers as its two arguments. So for
example, the following should work:
In[1]:= pi FromDigits RealDigits N Pi, 10
5
3 1 ;
In[2]:= FindSubsequence pi, 1415
Out[2]= 1 , 6955 , 29136 , 45234 , 79687 , 85880 , 88009
6. Write a function Collatz that takes an integer n as an argument and returns 3 n 1
if n is an odd integer and returns
n
2
if n is even. Your function Collatz should
attach a predicate to its argument to check whether it is even or odd.
7. Write the Collatz function from the above exercise, but this time attach a condi-
tion instead of a predicate. In addition, your condition should also check that the
argument to Collatz is positive.
6 Rule-based programming 163
8. Use alternatives to write a function abs[x] that, whenever x is an integer or a
rational, returns x if x 0, and x if x 0. Whenever x is complex, abs[x] should
return re x
2
im x
2
.
9. Create a function swapTwo[lis_List] that returns lis with its first two ele-
ments interchanged; for example, swapTwo[{a,b,c,d,e}] is {b,a,c,d,e}. If
lis has fewer than two elements, swapTwo just returns it. Write swapTwo using
three clauses: one for the empty list, one for one-element lists, and one for all other
lists. Then write it using two clauses: one for lists of length zero or one and another
for all longer lists.
10. Convert this definition to one that has no conditional parts (/;), but instead uses
pattern matching in the argument list:
f x_, y_ : x y ; IntegerQ x
f x_, y_ :
x 1 y ; Head x List && IntegerQ First x && y 1
11. Write a version of the HammingDistance function (described in Section 4.7) that
uses Cases instead of Select.
6.3 Transformation rules
Transformation rules are ubiquitous in Mathematica. They are used to represent solutions
to equations, as a means to specify options for functions, and they form the basis of most of
the algebraic manipulation in Mathematica. In this section we will look at how to use
pattern matching together with replacement rules to transform expressions based on these
rules.
A replacement rule is of the form pattern replacement or pattern replacement. Just
like traditional function definitions, the left-hand side of each of these rules matches an
expression and the right-hand side describes the transformation of that expression.
One of the most common uses for rules is in making substitutions of the form expr /.
rule. Any part of expr that matches the pattern in rule will be rewritten according to that
rule.
In[1]:= x y . y
Out[1]= x
164 An Introduction to Programming with Mathematica
A similar rule but using assignments would look like this:
In[2]:= f x_, y_ x y;
In[3]:= f x,
Out[3]= x
The main difference between the replacement rule and the assignment is that the
assignment will automatically be used whenever there is an appropriate pattern match
during evaluation. The expression f[x, ] matched the rule for f and the substitution
was performed automatically.
If you wish to restrict the use of a rule to a specific expression, you can use the
ReplaceAll function (shorthand notation /.) with the expression as the first argument
and a user-defined Rule or RuleDelayed function as the second argument. In standard
input form, the transformation rule (or local rewrite rule) appears immediately after the
expression, as the second argument to ReplaceAll.
In[4]:= x y . y
Out[4]= x
Here is the standard input form of the above.
In[5]:= ReplaceAll x y, Rule y,
Out[5]= x
When the Rule function is used with an expression, the expression itself is first
evaluated. Then both the left-hand side and right-hand side of the rule are evaluated, except
for those parts of the right-hand side that are held unevaluated by the Hold attribute.
Finally, everywhere that the evaluated left-hand side of the rule appears in the evaluated
expression, it is replaced by the evaluated right-hand side of the rule.
In[6]:= a, a . a Random
Out[6]= 0.474439, 0.474439
Using Trace, we can see the way the transformation rule works. Note in particular, that
the right-hand side of the rule is evaluated first.
In[7]:= Trace a, a . a Random
Out[7]= Random , 0.0883691 , a 0.0883691, a 0.0883691 ,
a, a . a 0.0883691, 0.0883691, 0.0883691
Just as in the case of assignments, there are immediate rules and delayed rules. In an
immediate rule (pattern replacement, with standard input form Rule[pattern, replace-
ment]), the replacement will be evaluated immediately. For delayed rules (pattern replace-
6 Rule-based programming 165
ment, with standard input form RuleDelayed[pattern, replacement]), the replacement is
only evaluated after the substitution is made.
In[8]:= a, a . a Random
Out[8]= 0.672823, 0.703154
Using Trace, we can see the way this transformation rule works.
In[9]:= Trace a, a . a Random
Out[9]= a Random , a Random , a, a . a Random ,
Random , Random , Random , 0.174287 ,
Random , 0.722288 , 0.174287, 0.722288
Transformation rules can be written using symbols.
In[10]:= a, b, c . List Plus
Out[10]= a b c
Transformation rules can also be written using labeled patterns.
In[11]:= 3, 4 , 7, 2 , 1, 5 . x_, y_ y, x
Out[11]= 4, 3 , 2, 7 , 5, 1
We can use multiple rules with an expression by enclosing them in a list.
In[12]:= a, b, c . c b, b a
Out[12]= a, a, b
A transformation rule is applied only once to each part of an expression (in contrast
to a rewrite rule) and multiple transformation rules are used in parallel. Hence, in the
above example, the symbol c is transformed into b but it is not further changed into a. In
order to apply one or more transformation rules repeatedly to an expression until the
expression no longer changes, the ReplaceRepeated function is used.
For example, the product of x and y is replaced by the sum of x and y, but this is
only done for the first such occurrence that matches.
In[13]:= a b c d . x_ y_ x y
Out[13]= a b c d
Using ReplaceRepeated, the rule is applied repeatedly until the expression no longer
changes.
In[14]:= a b c d . x_ y_ x y
Out[14]= a b c d
166 An Introduction to Programming with Mathematica
Let us now look at a few examples of problems that we solved earlier using a func-
tional style of programming but now solve them using a rule-based approach.
Example: Counting coins
Recall the pocket change example from Chapter 4 where a list of coins was given and a
function was constructed to count the value of the set of coins. Let us try to do the same
thing, but with a rule that gives the values of the coins.
In[15]:= coins p, p, q, n, d, d, p, q, q, p
Out[15]= p, p, q, n, d, d, p, q, q, p
Here are the values, given by a list of rules.
In[16]:= values p 1, n 5, d 10, q 25 ;
This replaces each coin by its value.
In[17]:= coins . values
Out[17]= 1, 1, 25, 5, 10, 10, 1, 25, 25, 1
And here is the value of the set of coins.
In[18]:= Apply Plus, coins . values
Out[18]= 104
Finally, here is a function that wraps up all these steps.
In[19]:= CountChange coins_List : Module values ,
values p 1, n 5, d 10, q 25 ;
Apply Plus, coins . values
In[20]:= CountChange p, q, q, n, d, d, p, q, q, d, d
Out[20]= 147
Example: Finding maxima
Our last example employs a sophisticated rewrite rule which demonstrates most of the
things discussed in this section: the repeated use of a transformation rule with delayed
evaluation, sequence patterns, and conditional pattern matching.
Recall the maxima function that we defined in Chapter 4, which returns the ele-
ments in a list of positive numbers that are bigger than all of the preceding numbers in the
list.
6 Rule-based programming 167
In[21]:= maxima x_List : Union Rest FoldList Max, 0, x
In[22]:= maxima 3, 5, 2, 6, 1, 8, 4, 9, 7
Out[22]= 3, 5, 6, 8, 9
We can also write this function using a pattern matching transformation rule.
In[23]:= maximaR x_List :
x . a___, b_, c___, d_, e___ ; d b a, b, c, e
Basically, the transformation rule repeatedly looks through the list for two elements
(b and d here), separated by a sequence of zero or more elements, such that the second
selected element is no greater than the first selected element. It then eliminates the second
element. The process stops when there are no two elements such that the second is less
than or equal to the first.
In[24]:= maximaR 3, 5, 2, 6, 1, 8, 4, 9, 7
Out[24]= 3, 5, 6, 8, 9
Exercises
1. Using Trace on maxima and maximaR, explain why the functional version is much
faster than the pattern matching version of the maxima function.
2. The following compound expression returns a value of 14.
In[1]:= z 11;
a 9;
z 3 . z a
Out[3]= 14
Describe the evaluation sequence that was followed. Use the Trace function to
check your answer.
3. Use the Hold function in the compound expression in the previous exercise to
obtain a value of 12.
4. The function definition f[x_Plus]:= Apply[Times,x] works as follows:
In[4]:= Clear f, a, b, c
In[5]:= f x_Plus : Apply Times, x
168 An Introduction to Programming with Mathematica
In[6]:= f a b c
Out[6]= a b c
The rewrite rule g[x_]:= x /.Plus[z___] Times[z] does not work. Use
Trace to see why and then modify this rule so that it performs the same operation as
the function f above.
5. Create a rewrite rule that uses a repeated replacement to unnest the nested lists
within a list.
In[7]:= unNest a, a, a , a , b, b, b , b, b , a, a
Out[7]= a, a, a , a , b, b, b , b, b , a, a
6. Define a function using pattern matching and repeated replacement to sum the
elements of a list.
7. Using the built-in function ReplaceList, write a function cartesianProduct
that takes two lists as input and returns the Cartesian product of these lists.
In[8]:= cartesianProduct x
1
, x
2
, x
3
, y
1
, y
2
Out[8]= x
1
, y
1
, x
1
, y
2
, x
2
, y
1
, x
2
, y
2
, x
3
, y
1
, x
3
, y
2
8. The function CellularAutomaton[rule, init, t] creates a list of the evolution
of a cellular automaton. For example, this generates five iterations of the cellular
automaton rule number 30 starting with the initial condition of a single 1 surrounded
by 0s.
In[9]:= CellularAutomaton 30, 1 , 0 , 5
Out[9]= 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0 , 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0 ,
0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 0 , 0, 0, 1, 1, 0, 1, 1, 1, 1, 0, 0 ,
0, 1, 1, 0, 0, 1, 0, 0, 0, 1, 0 , 1, 1, 0, 1, 1, 1, 1, 0, 1, 1, 1
Write a function CAGraphics[lis] that takes as argument, a list generated by
CellularAutomaton and produces a Graphics object that can then be displayed
directly with Show. Your function should use RasterArray and also a set of rules
to transform each 0 and 1 into different color directives such as Hue[.2].
In[10]:= ca30 CellularAutomaton 30, 1 , 0 , 500 ;
6 Rule-based programming 169
In[11]:= Show CAGraphics ca30
Out[11]= Graphics
6.4 Examples
This section focuses on two classical problems in computer science: encryption and
sorting. Even though we will only scratch the surface of these two very deep problems,
they are so important and ubiquitous in modern computing that it is well worth while
learning about them. As it turns out, these problems are well suited to a rule-based
approach, at least at an introductory level. We encourage you to investigate further the
theory and implementation of modern cipher and sorting algorithms. See, for example,
(Sedgewick, 1988) and (Wagon, 1999) for details.
Encoding text
In this example, we will develop functions for coding and decoding strings of text. The
particular coding that we will do is quite simplistic compared with contemporary commer-
cial-grade ciphers, but it will give us a chance to see how to combine string manipulation,
the use of functional programming constructs, and rule-based programming all in a very
practical example that should be accessible to anyone.
The problem in encryption is to develop an algorithm that can be used to encode a
string of text and then a dual algorithm that can be used to decode the encrypted message.
At first we will just limit ourselves to the 26 lowercase letters of the alphabet.
In[1]:= alphabet Map FromCharacterCode , Range 97, 122
Out[1]= a, b, c, d, e, f, g, h, i, j, k,
l, m, n, o, p, q, r, s, t, u, v, w, x, y, z
One of the simplest encryption schemes is attributed to Julius Caesar who is said to
have used this cipher to encode communications with his generals. The scheme is simply
170 An Introduction to Programming with Mathematica
to shift each letter of the alphabet some fixed number of places to the left. Using Thread,
we can set up rules that implement this shift.
In[2]:= CaesarCodeRules Thread alphabet RotateLeft alphabet
Out[2]= a b, b c, c d, d e, e f, f g, g h, h i,
i j, j k, k l, l m, m n, n o, o p, p q, q r,
r s, s t, t u, u v, v w, w x, x y, y z, z a
The decoding rules are simply to shift back in the other direction.
In[3]:= CaesarDecodeRules Thread alphabet RotateRight alphabet
Out[3]= a z, b a, c b, d c, e d, f e, g f, h g,
i h, j i, k j, l k, m l, n m, o n, p o, q p,
r q, s r, t s, u t, v u, w v, x w, y x, z y
To code a string, we will decompose the string into individual characters, apply the
code rules, and then join up the resulting characters in a word.
In[4]:= Characters "hello"
Out[4]= h, e, l, l, o
In[5]:= % . CaesarCodeRules
Out[5]= i, f, m, m, p
In[6]:= Apply StringJoin, %
Out[6]= ifmmp
Here is the function to accomplish this.
In[7]:= encode str_String, coderules_ :
Apply StringJoin, Characters str . coderules
Similarly, here is the decoding function.
In[8]:= decode str_String, decoderules_ :
Apply StringJoin, Characters str . decoderules
Let us try it out on a phrase.
In[9]:= encode "squeamish ossifrage", CaesarCodeRules
Out[9]= trvfbnjti pttjgsbhf
In[10]:= decode %, CaesarDecodeRules
Out[10]= squeamish ossifrage
6 Rule-based programming 171
Other ciphers can be created using permutations on the letters of the alphabet. We
will need the randomPermutation function we created in Section 5.2 of the chapter on
procedural programming.
In[11]:= randomPermutation lis_ : Module x, res , l2 lis ,
Do
x Part l2, Random Integer, 1, Length l2 ;
res Append res, x ;
l2 Complement l2, x ,
i, 1, Length lis ;
res
First we create a random permutation of the letters of the alphabet.
In[12]:= p randomPermutation alphabet
Out[12]= y, g, r, e, j, h, f, b, t, p, a,
k, i, m, o, c, u, z, w, d, v, q, s, n, x, l
Then, using Thread, we create a rule for each letter paired up with the corresponding
letter from the permutation p.
In[13]:= PermutationCodeRules Thread alphabet p
Out[13]= a y, b g, c r, d e, e j, f h, g f, h b,
i t, j p, k a, l k, m i, n m, o o, p c, q u,
r z, s w, t d, u v, v q, w s, x n, y x, z l
The decoding rules are obtained by simply reversing the above rules.
In[14]:= PermutationDecodeRules Thread p alphabet
Out[14]= y a, g b, r c, e d, j e, h f, f g, b h,
t i, p j, a k, k l, i m, m n, o o, c p, u q,
z r, w s, d t, v u, q v, s w, n x, x y, l z
In[15]:= encode "squeamish ossifrage", PermutationCodeRules
Out[15]= wuvjyitwb owwthzyfj
In[16]:= decode %, PermutationDecodeRules
Out[16]= squeamish ossifrage
Sorting a list
This next example also incorporates several of the concepts discussed in this chapter. It
uses a delayed rule, contains a conditional, and has several types of pattern matching.
172 An Introduction to Programming with Mathematica
We will create a rule named listsort that, upon repeated application, will put a
list of numbers into numerical order. To account for the first and last elements in the list,
we use BlankNullSequence (___).
In[17]:= listsort
x___, a_?NumericQ, b_?NumericQ, y___ x, b, a, y ; b a
Out[17]= x___, a_?NumericQ, b_?NumericQ, y___ x, b, a, y ; b a
The pattern that has to match {x___,a_,b_,y___} is a list of at least two ele-
ments since x___ and y___ will match zero or more elements. The condition on the
right-hand side of the rule says that whenever b is less than a, switch the order of a and b
in the original list to output {x,b,a,y}.
Here is a list of ten real numbers between 0 and 1.
In[18]:= nums Table Random , 10
Out[18]= 0.237736, 0.182151, 0.822792, 0.264693, 0.968603,
0.599673, 0.602053, 0.101958, 0.219543, 0.539043
In[19]:= nums //. listsort
Out[19]= 0.101958, 0.182151, 0.219543, 0.237736, 0.264693,
0.539043, 0.599673, 0.602053, 0.822792, 0.968603
Notice that because we used ?NumericQ as part of the pattern match, listsort
will work on expressions that may not be explicit numbers, but are numerical in nature;
that is, expressions that return explicit numbers when N is applied to them.
In[20]:= , , EulerGamma, GoldenRatio . listsort
Out[20]= EulerGamma, GoldenRatio, ,
This algorithm is far less efficient than many classical sorting algorithms, especially
those that employ a divide-and-conquer strategy.
In[21]:= nums Table Random , 100 ;
In[22]:= Timing nums . listsort ;
Out[22]= 0.942 Second, Null
The built-in Sort function uses a classical algorithm called merge sort (discussed
in Section 7.5), which starts by dividing the list into two parts of approximately equal size.
It then sorts each part recursively and finally merges the two sorted sublists.
In[23]:= Timing Sort nums ;
Out[23]= 0. Second, Null
6 Rule-based programming 173
The above implementation of listsort only works for numerical arguments. We
can overload listsort to work on characters of strings by making only two small
changes. First, we pattern match a and b with head String instead of ?NumericQ.
Second, instead of comparing a<b, we need to compare their character codes.
In[24]:= ToCharacterCode "z" 1
Out[24]= 122
Here then is the definition of listsort that operates on lists of string characters.
In[25]:= listsort x___, a_String, b_String, y___ x, b, a, y ;
Part ToCharacterCode b , 1 Part ToCharacterCode a , 1 ;
Out[25]= x___, a_String, b_String, y___
x, b, a, y ; ToCharacterCode b 1 ToCharacterCode a 1
Here are ten random characters.
In[26]:= chars
Table FromCharacterCode Random Integer, 97, 122 , 10
Out[26]= c, x, z, e, c, i, d, c, a, l
Here they are sorted.
In[27]:= chars . listsort
Out[27]= a, c, c, c, d, e, i, l, x, z
Exercises
1. Modify the Caesar cipher so that it encodes by shifting five places to the right.
2. Modify the alphabet permutation cipher so that instead of being based on single
letters, it is instead based on adjacent pairs of letters. Whereas the single letter cipher
will have 26 403291461126605635584000000 permutations, the adjacent pairs
cipher will have 26
2
permutations a very large number.
In[1]:= N 26
2
Out[1]= 1.883707684133810 10
1621
3. You can quickly create a graphics function to plot binary data (0s and 1s) using
Raster. For example:
174 An Introduction to Programming with Mathematica
In[2]:= data Table Random Integer , 5 , 5
Out[2]= 1, 1, 0, 0, 1 , 1, 0, 1, 1, 0 ,
0, 1, 1, 1, 1 , 0, 1, 0, 1, 1 , 0, 1, 1, 1, 0
In[3]:= Show Graphics Raster Reverse data , AspectRatio Automatic
Out[3]= Graphics
If you wanted to color the squares with color directives such as RGBColor or Hue or
GrayLevel, then you need to use RasterArray instead. Create a function
matrixPlot[mat,rules] that takes a matrix mat as its first argument and a list of
rules as the second argument. The list of rules should specify what color directive
each of the values in mat should be mapped to. Finally, compare your function with
ArrayPlot (new in Version 5.1).
In[4]:= ArrayPlot data, ColorRules 0 Black, 1 White ;
4. Plot the function sin x over the interval [ 2 , 2 ] and then reverse the x- and
y-coordinates of each point by means of a transformation rule.
5. Plot the function sin x y with x and y taking on values from 0 to 3 2. Then use a
transformation rule to perform a shear by shifting the graphic in the x-direction by a
factor of four.
6 Rule-based programming 175
6. Create a function rotatePlot[gr, ] that takes a plot gr and rotates it about the
origin by an angle . For example, to rotate a plot of the sine function, first create the
plot:
In[5]:= plot1 Plot Sin x , x, 0, 2 ;
1 2 3 4 5 6
1
0.5
0.5
1
Then perform the rotation of radians.
In[6]:= rotatePlot plot1, ;
6 5 4 3 2 1
1
0.5
0.5
1
7. Create a function rotatePlot3D[gr, , , ] that will rotate a Graphics3D
object gr about the origin by the angles , , and in the x, y, and z directions,
respectively.
176 An Introduction to Programming with Mathematica
7 Recursion
Some very important and classical problems in mathematics and computer science are
defined, or have solutions in terms of recursive definitions. A function is defined using
recursion if in its definition, it makes calls to itself. This programming paradigm is
easily implemented in Mathematica in a manner that is both natural and quite efficient.
In fact, many of the built-in operations of Mathematica could be written in Mathemat-
ica itself using recursion. In this chapter, we will present many examples of recursion
and explain how recursive functions are written.
7.1 Fibonacci numbers
Recursive definitions of mathematical quantities were used by mathematicians for centu-
ries before computers even existed. One famous example is the definition of a special
sequence of numbers first studied by the thirteenth-century Italian mathematician
Leonardo Fibonacci. The Fibonacci numbers have since been studied extensively, finding
application in many areas; see (Knuth 1997) for a detailed discussion.
The Fibonacci numbers are obtained as follows: write down two 1s, then continue
writing numbers computed by adding the last two numbers you have written down.
1 1 2 3 5 8 13 21
F
1
F
2
F
3
F
4
F
5
F
6
F
7
F
8

The simplest way to define these numbers is with recursion.

F
1
1
F
2
1
F
n
F
n 2
F
n 1
, for n 2
If we think of this sequence as a function, we would just change this to a functional
definition.
F 1 1
F 2 1
F n F n 2 F n 1 , for n 2
In this form, we can translate the definition directly into Mathematica.
In[1]:= F 1 1;
F 2 1;
F n_ : F n 2 F n 1 ; n 2
As it turns out, the condition /;n> 2 is unnecessary because Mathematica looks up
specific rules such as F[1]=1 before more general rules like that for F[n].
Here is a table of the first ten Fibonacci numbers.
In[4]:= Table F i , i, 1, 10
Out[4]= 1, 1, 2, 3, 5, 8, 13, 21, 34, 55
It is somewhat amazing that this works, but note that whenever we want to compute
F[n] for some n 2, we only apply F to numbers smaller than n. Tracing the evaluation of
F[4] makes the point well.
In[5]:= TracePrint F 4 , F _Integer F _ F _
F 4
F 4 2 F 4 1
F 2
F 3
F 3 2 F 3 1
F 1
F 2
Out[5]= 3
The first two lines indicate that F[4] is rewritten to F[4-2]+F[4-1], and the
lines that are indented one space show the calls of F[2] and F[3]. The lines showing calls
to F[1] and F[2] do not have any indented lines under them, since those values are
computed directly by a single rewrite rule, without making any recursive calls (for a fuller
explanation of this use of TracePrint, see Appendix B.
178 An Introduction to Programming with Mathematica
The key thing to understand about recursion is this: you can always apply a function
within its own definition, so long as you apply it only to smaller values. We will see this
principle used repeatedly in this chapter.
There is one other key point as well: we can apply the function to smaller and smaller
values, but we must eventually reach a value that can be computed without recursion. In the
case of the Fibonacci numbers, the numbers that can be computed without recursion the
base cases are F 1 and F 2 .
We will return to the Fibonacci numbers later in this chapter, in Section 7.6, where
we will see what can be done about a serious inefficiency in our implementation of F (also,
see Exercise 2 below).
Exercises
Before doing the exercises in this chapter, you may want to take a look at Appendix
B, which discusses some common programming errors, and how to debug recursive
functions.
1. For each of the following sequences of numbers, see if you can deduce the pattern
and write a Mathematica function to compute the ith value.
a.
2, 3, 6, 18, 108, 1944, 209952,
A
1
A
2
A
3
A
4
A
5
A
6
A
7

b.
0, 1, 1, 2, 3, 5, 8, 13, 21,
B
1
B
2
B
3
B
4
B
5
B
6
B
7
B
8
B
9

c.
0, 1, 2, 3, 6, 11, 20, 37, 68,
C
1
C
2
C
3
C
4
C
5
C
6
C
7
C
8
C
9

2. The numbers FA
n
represent the number of additions that are done in the course of
evaluating F[n].
0 0 1 2 4 7 12 20 33
FA
1
FA
2
FA
3
FA
4
FA
5
FA
6
FA
7
FA
8
FA
9

Write a function FA such that FA[n] = FA

n
.
7 Recursion 179
7.2 List functions
In Chapter 4, we looked at functional implementations of some list-oriented functions in
Mathematica. Although some of these functions have more efficient implementations in
terms of functional constructs, they provide a nice vehicle for discussing recursion, and so
in this section we will use them to introduce some of the basic concepts of recursive
programming.
We noted in our discussion of Fibonacci numbers that recursion works if the argu-
ments of recursive calls are smaller than the original argument. The same principle applies
to functions on lists. One common case is when the argument in the recursive call is the
tail (that is, Rest) of the original argument. An example is length, our recursively
defined version of the built-in function Length. The idea is that the length of a list is
always one greater than the length of its tail.
In[1]:= length lis_ : length Rest lis 1
Applying length to a list, however, leads to trouble.
In[2]:= length a, b, c
Rest::norest :
Cannot take Rest of expression with length zero. More
Rest::argx :
Rest called with 0 arguments; 1 argument is expected. More
General::stop : Further output of Rest::norest will
be suppressed during this calculation. More
General::stop :
Further output of $RecursionLimit::reclim will
be suppressed during this calculation. More
Out[2]= 255 Hold Hold length Rest Rest Rest
Well, perhaps it is already obvious, but what we are experiencing is one of the most
common errors in defining functions recursively we forgot the base cases. For length,
there is just one base case, the empty list.
In[3]:= length : 0
Now length works as we had intended it to.
In[4]:= length a, b, c
Out[4]= 3
180 An Introduction to Programming with Mathematica
Here is another simple example (for which we again have better solutions using
built-in operations): adding the elements of a list. We know several ways to do this, using
functional constructs for example.
In[5]:= sumElements lis_ : Apply Plus, lis
In[6]:= sumElements lis_ : Fold Plus, 0, lis
But for now we are just trying to get some practice with recursion. Here is the most
obvious recursive solution.
In[7]:= sumElements : 0
sumElements x_, r___ : x sumElements r
In[9]:= sumElements a, b, c
Out[9]= a b c
A trace of this computation shows the evaluation procedure in detail.
In[10]:= Trace sumElements a, b, c
Out[10]= sumElements a, b, c , a sumElements b, c ,
sumElements b, c , b sumElements c ,
sumElements c , c sumElements ,
sumElements , 0 , c 0, c , b c , a b c , a b c
We can use recursion for functions with multiple arguments as well. add
Pairs[lis
1
,lis
2
] is given two lists of numbers of equal length and returns a list containing
the pairwise sums.
Here, the idea is to apply addPairs recursively to the tails of both lists.
In[11]:= addPairs , :
addPairs x1_, r1___ , x2_, r2___ :
Join x1 x2 , addPairs r1 , r2
In[13]:= addPairs 1, 2, 3 , 4, 5, 6
Out[13]= 5, 7, 9
In[14]:= addPairs x
1
, y
1
, z
1
, x
2
, y
2
, z
2
Out[14]= x
1
x
2
, y
1
y
2
, z
1
z
2
The recursive calls do not always have to be on the tail of the original argument. Any
smaller list will do. The function multPairwise multiplies together every pair of
elements in a list. The trick is to make the recursive call on the tail of the tail.
In[15]:= multPairwise :
multPairwise x_, y_, r___ : Join x y , multPairwise r
7 Recursion 181
In[17]:= multPairwise 3, 9, 17, 2, 6, 60
Out[17]= 27, 34, 360
As a last simple example, consider the function deal defined in Chapter 4. The
function deal[n] produces a list of n playing cards randomly chosen from a 52-card deck
(stored as the value of cardDeck, a 52-element list). Here is how we might write this
function recursively.
First, dealing zero cards is easy.
In[18]:= deal 0 :
Now, suppose we have dealt n 1 cards; how do we deal n? Just randomly deal a card
from the remaining 52 n 1 53 n. To do this, randomly choose an integer r
between 1 and 53 n, remove the rth card, and add it to the list of cards already dealt.
In[19]:= deal n_ : Module dealt deal n 1 , Append dealt,
Complement cardDeck, dealt Random Integer, 1, 53 n
Here again is the cardDeck function defined earlier in Chapter 4.
In[20]:= cardDeck Flatten Outer List,
, , , , Join Range 2, 10 , J, Q, K, A , 1 ;
And here is the recursive deal.
In[21]:= deal 5
Out[21]= , 7 , , 5 , , J , , 2 , , A
Exercises
1. Write a recursive function sumOddElements[lis] that adds up only the elements
of the list lis that are odd integers. lis may contain even integers and nonintegers.
(Use IntegerQ to determine if a given element is an integer.)
2. Write a recursive function sumEveryOtherElement[lis] that adds up
lis[[1]], lis[[3]], lis[[5]], etc. Each of these elements is a number. lis may
have any number of elements.
3. Write a function addTriples[lis
1
,lis
2
,lis
3
] that is like addPairs in that it adds
up the corresponding elements of the three equal-length lists of numbers.
182 An Introduction to Programming with Mathematica
4. Write a function multAllPairs[lis] that multiplies every consecutive pair of
integers in the numerical list lis.
In[1]:= multAllPairs 3, 9, 17, 2, 6, 60
Out[1]= 27, 153, 34, 12, 360
5. Write the function maxPairs[lis
1
,lis
2
] which, for numerical lists of equal length,
returns a list of the greater value in each corresponding pair.
6. The function interleave[lis
1
,lis
2
], which merges two lists of equal length, can
be defined as follows:
In[2]:= interleave lis1_, lis2_ : Flatten Transpose lis1, lis2
In[3]:= interleave a, b, c , x, y, z
Out[3]= a, x, b, y, c, z
Rewrite interleave using recursion.
7.3 Thinking recursively: examples
The manner in which expressions are rewritten during Mathematicas evaluation process
completely explains how recursion works, and it can be seen using Trace or Trace
Print, as we did above. But that knowledge is of only limited usefulness in writing
recursive functions.
Indeed, the real trick is to forget the evaluation process and simply assume that the
function you are defining will return the correct answer when applied to smaller values.
Suspend disbelief you will begin to see how simple recursion really is.
Finding maxima
Recall the function maxima (from Section 4.4 of the Functional Programming chapter),
which, given a list of numbers, produces a list of those numbers greater than all those that
precede them.
In[1]:= maxima 9, 2, 10, 3, 14, 9
Out[1]= 9, 10, 14
7 Recursion 183
We again start by assuming that we can easily compute maxima[Rest[lis]] for any
list lis, and then ask ourselves: how can we compute maxima[lis] starting from maxima[
Rest[lis]]?
In[2]:= maxima Rest 9, 2, 10, 3, 14, 9
Out[2]= 2, 10, 14
The answer is to remove any values not greater than First[lis], then put First[lis] at
the beginning of the result.
In[3]:= Select %, # 9 &
Out[3]= 10, 14
In[4]:= Join 9 , %
Out[4]= 9, 10, 14
Again, the base case needs to be accounted for, and we end up with the following:
In[5]:= maxima :
In[6]:= maxima x_, r___ : Join x , Select maxima r , # x &
In[7]:= maxima 3, 6, 2, 1, 8, 7, 12
Out[7]= 3, 6, 8, 12
The lesson of this section (and it is an important one) is not to worry about how the
recursive cases are computed assume that they work, and just think about how to compute
the value you want from the result of the recursive call.
Subsets
The second problem we will tackle is to generate a list of k-element subsets of any given
set. Our sets will be represented by lists in Mathematica, so this amounts to a recursion on
the elements of this list. The syntax of our function will be subsets[lis, k], so if k 2,
for example, this would generate all 2-element subsets of lis.
Let us apply the basic principle we have just learned. Given lis, we assume that
subsets[Rest[lis],k-1] will give the correct result, and call that result res. How can
we then compute subsets[lis]?
A possible approach to defining subsets is to take subsets[Rest[lis],k]
together with the result of joining {First[lis]} to all the elements in subsets[
Rest[lis],k 1].
184 An Introduction to Programming with Mathematica
We need to first define the base cases: subsets of length 0 and the subsets of an
empty set.
In[8]:= subsets lis_, 0 :
subsets , k_ :
Here then is the recursion.
In[10]:= subsets lis_, k_ :
Module res subsets Rest lis , k 1 , Join
Map Join First lis , # & , res , subsets Rest lis , k
Here are a few tests.
In[11]:= subsets Range 5 , 1
Out[11]= 1 , 2 , 3 , 4 , 5
In[12]:= subsets Range 5 , 2
Out[12]= 1, 2 , 1, 3 , 1, 4 , 1, 5 ,
2, 3 , 2, 4 , 2, 5 , 3, 4 , 3, 5 , 4, 5
In[13]:= subsets Range 5 , 4
Out[13]= 1, 2, 3, 4 , 1, 2, 3, 5 ,
1, 2, 4, 5 , 1, 3, 4, 5 , 2, 3, 4, 5
The recursion is fairly deep, so this function is not terribly efficient for large values.
This computes all of the 5-element subsets from the 30-element set given by Range[30].
In[14]:= Timing xx subsets Range 30 , 5 ;
Out[14]= 6.189 Second, Null
There are
30
5
142,506 of them.
In[15]:= Length xx
Out[15]= 142506
We should check that our function at least produced the correct number of subsets.
In[16]:= Binomial 30, 5
Out[16]= 142506
7 Recursion 185
Comparing our subsets with the built-in Subsets, we can see that the built-in function
is clearly superior in terms of speed, being more than two orders of magnitude faster.
In[17]:= Timing yy Subsets Range 30 , 5 ;
Out[17]= 0.03 Second, Null
In[18]:= Length yy
Out[18]= 142506
The exercises at the end of this section include a problem to modify the subsets
function so that it behaves more like the built-in Subsets, which allows you to also
generate subsets of all lengths up to a given size.
Run-length encoding
We now turn to another, somewhat more involved example programming run-length
encoding. runEncode implements a method commonly used to compress large amounts
of data in those cases where the data are likely to contain long sequences (runs) of the
same value. A good example is the representation of video images in a computer as collec-
tions of color values for the individual dots, or pixels, in the image. Since video pictures
often contain large areas of a single color, this representation may lead to lists of hundreds,
or even thousands of occurrences of the identical color value, one after another. Such a
sequence can be represented very compactly using just two numbers, the color value and
the length of the run.
runEncode compresses a list by dividing it into runs of occurrences of a single
element, and returns a list of the runs, each represented as a pair containing the element
and the length of its run. So the following list,
9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5
should produce the following runs once encoded.
9, 5 , 4, 1 , 3, 4 , 5, 6
Given list lis, we just assume that runEncode[Rest[lis]] gives the compressed
form of the tail of lis (call it res), and ask ourselves: given the list lis and the list res, how can
we compute runEncode[lis]? Let x be lis[[1]], and consider the cases:
1. First we define what runEncode should do in the two base cases: when the list is
empty and when the list consists of only one element.
In[19]:= runEncode[{}] := {}
runEncode[{x_}] := {{x, 1}}
186 An Introduction to Programming with Mathematica
2. res might be {}, if lis has one element. In this case, lis ={x} and run
Encode[lis]} = {x,1}.
3. If the length of lis is greater than 1, res has the form{{y,k},}, and there
are two cases:
y x: runEncode[lis] ={{y,k 1},}
y x: runEncode[lis]= {{x,1},{y,k},}
In[21]:= runEncode x_, res___ : Module R runEncode res , p ,
p First R ;
If x First p ,
Join x, p 2 1 , Rest R ,
Join x, 1 , R
In[22]:= runEncode 9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5
Out[22]= 9, 5 , 4, 1 , 3, 4 , 5, 6
This can be made a lot clearer by replacing the last clause above with a transformation rule.
In[23]:= runEncodeT x_, res__ : runEncodeT res .
y_, k_ , s___ If x y, x, k 1 , s , x, 1 , y, k , s
In[24]:= runEncodeT[{}] := {}
runEncodeT[{x_}] := {{x, 1}}
In[26]:= runEncodeT 9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5
Out[26]= 9, 5 , 4, 1 , 3, 4 , 5, 6
Incidentally, a program for this problem, due to Frank Zizza of Willamette College,
won an honorable mention in the programming contest at the 1990 Mathematica Confer-
ence. It uses no recursion, just repeated substitution.
In[27]:= runEncodeZ lis_ : Map #, 1 &, lis .
x___, y_, i_ , y_, j_ , z___ x, y, i j , z
In[28]:= runEncodeZ 9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5
Out[28]= 9, 5 , 4, 1 , 3, 4 , 5, 6
Impressively clever, and quite similar conceptually to the listsort function we
created in Section 6.4, but our recursive version is much more efficient on most examples.
Mathematica contains a function Split which effectively does run length encoding,
although it represents the output slightly different from our runEncode functions.
7 Recursion 187
In[29]:= Split 9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5
Out[29]= 9, 9, 9, 9, 9 , 4 , 3, 3, 3, 3 , 5, 5, 5, 5, 5, 5
You could easily convert the output of Split to that produced by our runEncode
functions by mapping the appropriate pure function.
In[30]:= Map First # , Length # &, %
Out[30]= 9, 5 , 4, 1 , 3, 4 , 5, 6
We leave it as an exercise to go in the other direction that is, convert the output of
our runEncode function to that produced by Split.
Finally, we should mention some efficiency issues. Each of the run-length encoding
implementations presented in this section are reasonably fast for relatively small inputs,
vectors of length less than a few hundred. But for larger vectors and for certain cases, they
get quite bogged down, mostly due to the deep recursion needed in these cases. This can
be seen quite plainly as follows:
In[31]:= data Range 300 ;
In[32]:= runEncode data
$RecursionLimit::reclim :
Recursion depth of 256 exceeded. More
General::stop :
Further output of $RecursionLimit::reclim will
be suppressed during this calculation. More
Join::heads : Heads List and If at positions
1 and 2 are expected to be the same. More
General::stop : Further output of Join::heads will
be suppressed during this calculation. More
Out[32]= If 1 2, 1 , Join 1, p$178481 2 1 , Rest R$178481 ,
Join 1, 1 , R$178481
A possible solution would be to recognize that there is quite a deep recursion here
and hence to increase the built in recursion limit; and this seems to work.
In[33]:= Block $RecursionLimit ,
Timing runEncode data ;
Out[33]= 0.01 Second, Null
188 An Introduction to Programming with Mathematica
But trying a larger example shows that the underlying algorithm, although mostly
linear in the size of the input, is quite slow for input as small as about 10,000 in length.
In[34]:= Block $RecursionLimit ,
Table Timing runEncode Range 2
k
10
3
; 1 , k, 0, 3
Out[34]= 0.07 Second, 0.18 Second, 0.671 Second, 2.413 Second
In such cases it is best to rethink your algorithm and either try to refine it or find a
different and better implementation. In the case of run-length encoding, a more direct,
functional approach proves to be much more efficient. Although the following code does
not use recursion, we present it here anyway so the reader can compare it with the recur-
sive functions and perform some efficiency tests on the various implementations.
Here is an example list we will use to develop the prototype code.
In[35]:= vec 9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5 ;
We first take overlapping pairs fromvec.
In[36]:= Partition vec, 2, 1
Out[36]= 9, 9 , 9, 9 , 9, 9 , 9, 9 , 9, 4 , 4, 3 , 3, 3 , 3, 3 ,
3, 3 , 3, 5 , 5, 5 , 5, 5 , 5, 5 , 5, 5 , 5, 5
Each run ends at the position at which a pair from the above partition contains different
elements.
In[37]:= end Flatten Position %, a_, b_ ; a b
Out[37]= 5, 6, 10
We have to add the positions at the beginning and end of the list.
In[38]:= end Join 0 , end, Length vec
Out[38]= 0, 5, 6, 10, 16
Creating pairs again shows the ending position paired up with the next ending position for
each run.
In[39]:= Partition end, 2, 1
Out[39]= 0, 5 , 5, 6 , 6, 10 , 10, 16
To indicate where the run starts, not where the previous run ended, we add 1 to each first
coordinate.
In[40]:= runs Map Plus #, 1, 0 &, %
Out[40]= 1, 5 , 6, 6 , 7, 10 , 11, 16
7 Recursion 189
Now each pair from runs consists of the starting position and the run length. We can use
these pairs as the second argument to Take as in the following example.
In[41]:= Take a, b, c, d, e , 3, 5
Out[41]= c, d, e
So, finally, here is the list of runs.
In[42]:= Map Take vec, # &, runs
Out[42]= 9, 9, 9, 9, 9 , 4 , 3, 3, 3, 3 , 5, 5, 5, 5, 5, 5
Here then is the function split that produces output identical to the built-in
Split.
In[43]:= split lis_ : Module end, t, runs ,
end
Flatten Position Partition lis, 2, 1 , a_, b_ ; a b ;
t Partition Join 0 , end, Length lis , 2, 1 ;
runs Map Plus #, 1, 0 &, t ;
Map Take lis, # &, runs
In[44]:= split vec
Out[44]= 9, 9, 9, 9, 9 , 4 , 3, 3, 3, 3 , 5, 5, 5, 5, 5, 5
This implementation is extremely efficient.
In[45]:= data Range 10
5
;
In[46]:= Timing split data ; 1
Out[46]= 0.641 Second
In[47]:= data Table Random Integer , 10
5
;
In[48]:= Timing split data ; 1
Out[48]= 0.591 Second
By comparison, we see that our split is only about one order of magnitude slower
than the built-in function, which is optimized for such tasks.
In[49]:= Timing Split data ; 1
Out[49]= 0.04 Second
190 An Introduction to Programming with Mathematica
Exercises
1. Write the function prefixMatch[lis
1
,lis
2
] that finds the starting segments of lis
1
and lis
2
that match.
In[1]:= prefixMatch 1, 2, 3, 4 , 1, 2, 5
Out[1]= 1, 2
2. Modify runEncode so that it leaves single elements as they are.
In[2]:= runEncode2 9, 9, 9, 4, 3, 3, 5
Out[2]= 9, 3 , 4, 3, 2 , 5
For this version, you need to assume that the argument is a list of atoms, otherwise
the output would be ambiguous.
3. Modify one of the runEncode functions so that it produces output in the same
form as the built-in Split function.
In[3]:= Split 9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5
Out[3]= 9, 9, 9, 9, 9 , 4 , 3, 3, 3, 3 , 5, 5, 5, 5, 5, 5
4. A slightly more efficient version of runEncode uses a three-argument auxiliary
function.
runEncode[{}] := {}
runEncode[{x_, r___}] := runEncode[x, 1, {r}]
runEncode[x,k,{r}] computes the compressed version of x, x, x, , x, r ,
where the xs are given k times. Define this three-argument function. (Note that it is
legal to have a function be defined for different numbers of arguments; rules in which
runEncode appears on the left-hand side with two arguments will only be applied
when runEncode is called with two arguments, and likewise for the three-argument
version.) Using the Timing function, compare the efficiency of this version with our
earlier version; be sure to try a variety of examples, including lists that have many
short runs and ones that have fewer, but longer runs. You will need to use Table to
generate lists long enough to see any difference in speed.
5. maxima can also be computed more efficiently with an auxiliary function.
maxima :
maxima x_, r___ : maxima x, r
7 Recursion 191
The two-argument version has this meaning: maxima[x,lis] gives the maxima of
the list Join[{x},lis]. Define it. (Hint: the key point about this is that
maxima[x,lis] is equal to maxima[x,Rest[lis]] if x First[lis].) Compare its
efficiency with the version in the text.
6. Write the function runDecode, which takes an encoded list produced by runEn
code and returns its unencoded form.
In[4]:= runDecode 9, 5 , 4, 1 , 3, 4 , 5, 6
Out[4]= 9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5
7. The code we developed to compute the k-element subsets of any given list differs
from the built-in Subsets function in that the latter has a mechanism for generat-
ing all subsets of length less than or equal to k.
In[5]:= A a, b, c, d ;
Subsets A, 2
Out[6]= , a , b , c , d , a, b ,
a, c , a, d , b, c , b, d , c, d
If you want to get only two-element subsets you use a slightly different form.
In[7]:= Subsets A, 2
Out[7]= a, b , a, c , a, d , b, c , b, d , c, d
Modify the function subsets developed in this section to take either form:
subsets[lis,k] or subsets[lis,{k}] so that it mimics the behavior of the
built-in Subsets.
7.4 Recursion and symbolic computations
Chapters 2 and 6 emphasized the idea that expressions and data are really the same things
in Mathematica. All that distinguishes an expression like 2 3 from one like x y is that
Mathematica has rules for rewriting 2 3 but not for x y.
Symbolic computations are those that transform expressions into other expressions.
Programming symbolic computations is no different from any other type of computation:
you write rewrite rules, and use local transformations, built-in operations, and recursion.
We will illustrate symbolic computation with what may be the most famous recursive
definition of them all: the differential calculus. Every elementary calculus book includes
192 An Introduction to Programming with Mathematica
rules for finding derivatives of functions. Generally, they assume that there are expressions
u containing the variable x and they show how to find the derivative of u with respect to x,
u
x
, by giving rules like the following.
c
x
0, for c a constant
x
n
x
n x
n 1
u v
x
u
x
v
x
If we think of u x as a function
x
being applied to an expression u, then these
rules would be written in the following notation.
x
c 0, for c a constant
x
x
n
n x
n 1
x
u v
x
u
x
v
In this form, it is clear that
x
is just a recursively defined function from expression
to expression, and we can render this function in Mathematica directly.
In[1]:= ddx c_ : 0
ddx x
n_
: n x
n 1
ddx u_ v_ : ddx u ddx v
In[4]:= ddx x
2
x
3
Out[4]= 2 x 3 x
2
So far, so good, but there are two problems with this, one big and the other bigger.
The bigger one is that this function gives completely wrong answers for many expressions.
In[5]:= ddx 5 x
3
Out[5]= 0
We have not been careful enough about our base cases. Specifically, the first rule
handles all expressions not specifically treated elsewhere, instead of just those for which it
was intended: constants. This is easily remedied, by replacing that rule with one that makes
sure its argument is a number.
First we remove the original definition we gave above for the derivative of a constant.
In[6]:= ddx c_ .
In[7]:= ddx c_?NumericQ : 0
7 Recursion 193
Now, ddx always gives an answer that is correct, but it still misses a lot of cases.
In[8]:= ddx 5 x
3
Out[8]= ddx 5 x
3
At this point, we need to take a close look at the cases we want to cover; that is, the
precise set of expressions we want ddx to differentiate. We can define this set using
recursion.
An expression (that ddx can differentiate) is one of the following:
a number
the variable x
a sumu v, where u and v are expressions
a difference u v of two expressions
a product u v of two expressions
a quotient u v of two expressions
a power u
n
of an expression and a number
Now, let us start from scratch, dealing systematically with all the cases.
In[9]:= Clear ddx
ddx c_?NumericQ : 0
ddx x : 1
ddx u_ v_ : ddx u ddx v
ddx u_ v_ : ddx u ddx v
ddx u_ v_ : u ddx v v ddx u
ddx
u_
v_
:
v ddx u u ddx v
v
2
ddx u_
c_?NumericQ
: c u
c 1
ddx u
In[17]:= ddx 5 x
3
Out[17]= 15 x
2
Note the use of NumericQ (as opposed to NumberQ). NumberQ returns a value of
True only if its argument is explicitly a number. It returns False for symbols that are
numeric though.
194 An Introduction to Programming with Mathematica
In[18]:= NumberQ
Out[18]= False
NumericQ, on the other hand, returns true for any expression that is numeric, including
symbols such as , , and .
In[19]:= ddx
Out[19]= 0
One interesting point to note here is that one of the cases from our first definition,
x
n
, does not appear here in that form. Still, this case is handled correctly, as we have just
seen. A Trace makes it clear why.
In[20]:= Trace ddx x
3
, ddx
Out[20]= ddx x
3
, 3 x
3 1
ddx x , ddx x , 1
In other words, it is handled as part of a more general case, namely u
n
for arbitrary u.
Our new rule works in additional cases.
In[21]:= ddx x 2 x
2
4
Out[21]= 4 1 4 x x 2 x
2
3
It is very common to make the mistake of covering cases in more ways than one. For
example, many calculus books include both the case c x
n
and, separately, the cases for c, x,
u
n
, and u v, which together can handle expressions of the form c x
n
. It is harmless, but a
more systematic treatment of the cases avoids giving extra rules, while also ensuring that all
cases are covered.
Finally, we might want to make use of simple algebraic identities to simplify this
code. For example, the rule for quotients is already covered by the rules for products and
powers, since
u
v
u v
1
. Similarly, u v u 1 v.
In[22]:= ddx u_ v_ : ddx u v
ddx
u_
v_
: ddx
u
v
Trying these new definitions out on an example still fails.
In[24]:= ddx
x
2
x 1
$IterationLimit::itlim :
Iteration limit of 4096 exceeded. More
Out[24]= Hold ddx
x
2
1 x
7 Recursion 195
In other words, this computation was going on forever. Alas, here Mathematicas own
simplification rules defeated us, as we can see by looking at the rules for ddx.
In[25]:= ?ddx
Global`ddx
ddx x : 1
ddx c_?NumericQ : 0
ddx u_ v_ : ddx u v
ddx u_ v_ : ddx u ddx v
ddx
u_
v_
: ddx
u
v
ddx u_ v_ : u ddx v v ddx u
ddx u_
c_?NumericQ
: c u
c 1
ddx u
When we entered the new rules, Mathematica rewrote the right-hand sides, so that
the rules just say, in effect, rewrite ddx[u-v] to ddx[u-v] and rewrite ddx[u/v] to
ddx[u/v]. This fails to satisfy our rule that recursive calls can only be made to smaller
values.
On the other hand, let us try just deleting those rules entirely and see what happens.
In[26]:= ddx u_ v_ .
ddx
u_
v_
.
In[28]:= Simplify ddx
x
x 1
Out[28]=
1
1 x
2
Again, we need to take into account what Mathematica is doing with the expressions
we enter. It turns out that it actually reads expressions of the form u v as u v
1
and
expressions of the formu v as u 1 v.
In[29]:= FullForm
u
v
Out[29]//FullForm=
Times u, Power v, 1
196 An Introduction to Programming with Mathematica
In[30]:= FullForm u v
Out[30]//FullForm=
Plus u, Times 1, v
When we entered ddx[x/(x-1)], Mathematica read it as ddx x x 1
1
. In
this form, the existing rules apply.
In[31]:= FullForm Hold ddx
x
x 1
Out[31]//FullForm=
Hold ddx Times x, Power Plus x, 1 , 1
Exercises
1. Add rules to ddx for the trigonometric functions sine, cosine, and tangent.
2. When variables other than x are present in an expression, the rules for differentiation
with respect to x actually do not change. That is, expressions that have no occur-
rences of x are treated like constants. So there should be a rule that says ddx[u]=0,
if x does not occur anywhere in u. Define the function nox[e] to return True if x
does not occur within e, then add the new rule for those expressions. You will need to
use the comparison function =!=, called UnsameQ, which tests whether two symbols
are unequal; the usual Unequal comparison (!=) cannot be used to compare
symbols.
3. Define a two-argument version of ddx whose second argument is the variable with
respect to which the derivative of the expression is to be computed. Thus, ddx[u,x]
will be the same as our current ddx[u]. You will need to determine when an expres-
sion has no occurrences of a variable; you can use the built-in function FreeQ.
7 Recursion 197
7.5 Classical examples
Merge sort
Sorting the elements of a list is one of the most important tasks in computer science.
There are quite a few well-studied algorithms that have been developed for performing
various types of sorting. These include selection sort, insertion sort, bubble sort, quick
sort, heap sort, merge sort, and many others. We have already looked at a rather primitive
list sorting algorithm in Section 6.4. In this section, we will develop an algorithm for
merge sort, which is a classical divide-and-conquer algorithm.
The procedure for merge sort consists of three basic steps:
first, split the original list into two parts of roughly equal size
sort each part recursively
finally, merge the two sorted sublists
We will start with the last step first creating a function merge that takes two lists,
each assumed to be sorted, and produces a single merged, sorted list. Using pattern match-
ing we can set this up as a recursion. First we deal with the cases of when either of the two
lists is empty.
In[1]:= merge lis_List, : lis
merge , lis_List : lis
The recursion then is on the tail of the sublists. We use the triple-blank to pattern
match ra and rb here so that they can represent zero, one, or more arguments.
In[3]:= merge a_, ra___ , b_, rb___ :
If a b,
Join a , merge b , ra, rb ,
Join b , merge a, ra , rb
Here are several test cases.
In[4]:= merge 1, 4, 7 , 2, 6, 9, 14
Out[4]= 1, 2, 4, 6, 7, 9, 14
In[5]:= merge 14 , 2, 5, 7, 8
Out[5]= 2, 5, 7, 8, 14
198 An Introduction to Programming with Mathematica
Now we turn to the sorting function. This too will be defined recursively by first
dividing the list into two sublists, performing the sort on each sublist and then merging
these two sorted sublists using the above merge function.
Here are the two base cases; the empty list and a list with a single element in it.
In[6]:= MergeSort : ;
MergeSort x_ : x ;
Here is the recursion.
In[8]:= MergeSort lis_List : Module div Floor
Length lis
2
,
merge
MergeSort Take lis, div , MergeSort Drop lis, div
Let us look at a few test cases to get a sense of the efficiency of our program.
In[9]:= dataInt Table Random Integer, 1, 100 , 20
Out[9]= 84, 83, 58, 8, 30, 99, 72, 29, 77,
95, 63, 67, 47, 40, 95, 71, 14, 57, 57, 24
In[10]:= MergeSort dataInt
Out[10]= 8, 14, 29, 30, 58, 72, 77, 83, 84,
95, 24, 40, 47, 57, 63, 67, 57, 71, 95, 99
In[11]:= dataReal Table Random , 1000 ;
In[12]:= Timing
Block $RecursionLimit , MergeSort dataReal ;
Out[12]= 0.16 Second, Null
Notice the need to increase the built-in recursion limit for larger computations. This
limitation in our current definitions is due to the fact that both merge and MergeSort
use recursion and that MergeSort has a double recursive call in it.
In comparison, the built-in Sort function, which uses a modified merge sort, is
optimized for dealing with large arrays of numbers and is much, much faster.
In[13]:= Timing Sort dataReal ;
Out[13]= 0. Second, Null
Here we see that Sort can perform this computation in about the same time it took
our MergeSort to sort a dataset that was two orders of magnitude smaller.
In[14]:= Timing Sort Table Random , 10
5
;
Out[14]= 0.08 Second, Null
7 Recursion 199
The exercises will give you a chance to refine the MergeSort and improve its
efficiency.
Gaussian elimination
An extremely common problem in mathematical computation is to solve a linear system S
of the following form.
E
1
: a
11
x
1
a
12
x
2
a
1 n
x
n
b
1
E
2
: a
21
x
1
a
22
x
2
a
2 n
x
n
b
2
E
n
: a
n1
x
1
a
n2
x
2
a
nn
x
n
b
n
The values of the variables x
1
, , x
n
are called the unknowns, and the a
ij
and b
i
are
constants.
Mathematica has a built-in function LinearSolve that will usually give the correct
answer. For example, here is a simple 2 2 system, two equations in two unknowns.
x
1
2 x
2
3
4 x
1
5 x
2
6
With a little work, you can see that the solution is x
1
1, x
2
2. Here is how to
solve this system using the built-in LinearSolve.
In[15]:= m
1 2
4 5
;
b 3, 6 ;
In[17]:= LinearSolve m, b
Out[17]= 1, 2
So why learn to program it yourself? Because LinearSolve, like any algorithm,
may run into trouble on certain kinds of input, and when confronted with a system for
which it fails, your only recourse will be to write your own program.
The Hilbert matrices, containing elements h
ij
1 i j 1 , cause problems for
LinearSolve.
In[18]:= HilbertMatrix n_ : Table
1
i j 1
, i, n , j, n
Here is a random1 10 vector.
In[19]:= b Table Random , 10 ;
200 An Introduction to Programming with Mathematica
In[20]:= xsoln LinearSolve HilbertMatrix 10 , b
LinearSolve::luc :
Result for LinearSolve of badly conditioned
matrix 1., 0.5, 6 , 0.111111, 0.1 ,
8 , 0.1, 8 , 20 may
contain significant numerical errors. More
Out[20]= 1.56074 10
6
, 1.28819 10
8
, 2.64644 10
9
,
2.33592 10
10
, 1.08702 10
11
, 2.92604 10
11
,
4.71438 10
11
, 4.48426 10
11
, 2.32154 10
11
, 5.04243 10
10
Using this last result in the original system should give all 0s, but it does not.
In[21]:= HilbertMatrix 10 .xsoln b
Out[21]= 0.0000109427, 3.52143 10
6
, 0.0000175983,
8.07756 10
6
, 9.30083 10
6
, 6.04337 10
6
, 1.22291 10
6
,
1.89195 10
6
, 8.85718 10
6
, 7.31392 10
6
In this section, we will show how to program a simple and classic method, called
Gaussian elimination, to solve linear systems. Our method, unfortunately, will also fail on
the Hilbert matrix, but we will revisit the problem in Chapter 8 and show how a variant of
this method can solve it.
For now, consider that we have the system shown above. By the principle of recur-
sion, we can assume the ability to solve any smaller system in particular, any system of
n 1 equations in n 1 unknowns, and ask our usual question: How can the ability to solve
smaller systems be used to solve this system?
The idea behind Gaussian elimination is to eliminate all occurrences of x
1
from the
equations E
2
, , E
n
. For example, here is how to eliminate x
1
from E
2
:
1. Multiply the first equation E
1
by a
21
a
11
.
a
21
a
11
a
11
x
1
a
21
x
2
a
1 n
x
n
a
21
a
11
b
1
This simplifies to:
a
21
x
1
a
21
a
11
a
12
x
2
a
21
a
11
a
1 n
x
n
a
21
a
11
b
1
2. Subtract this modified equation from E
2
.
a
21
x
1
a
22
x
2
a
2 n
x
n
b
2
a
21
x
1
a
21
a
11
a
12
x
2
a
21
a
11
a
1 n
x
n
a
21
a
11
b
1
a
22
a
21
a
11
a
12
x
2
a
2 n
a
21
a
11
a
1 n
x
n
b
2
a
21
a
11
b
1
7 Recursion 201
We have obtained an equation having only n 1 variables. Now do this for every
equation: Transform E
i
, for all 2 i n, to E
i
'
E
i
a
i1
a
11
E
1
. Call this new system of
equations S
'
.
We are almost there. We can (recursively) find the solution to the system S
'
, obtain-
ing the values of x
2
, , x
n
. Then x
1
is found by computing the following.
x
1
b
1
a
12
x
2
a
1 n
x
n
a
11
In programming this procedure, the system will be represented by the n n matrix of
coefficients, together with the vector of the b
i
. In fact, it is somewhat more convenient to
represent the entire system as an n 1 n matrix (called the augmented matrix), with the b
i
included as the last column. We will define solve[s], where s is such an n 1 n matrix,
to return a list of the values of the n unknowns x
1
, , x
n
. Once we understand the algo-
rithm, the programming is simply a lot of list manipulation.
In[22]:= solve s_ : Module E1 First s , x2toxn solve elimx1 s ,
Join
Last E1 Drop Rest E1 , 1 .x2toxn First E1 , x2toxn
We need to define elimx1[s], which produces the smaller system. But first, let us
not forget the base case, n 1 (that is, a
11
x
1
b
1
), which is trivial to solve.
In[23]:= solve a11_, b1_ :
b1
a11
Again the elimination phase takes each row a
i1
, a
i2
, , a
in
, b
i
and transforms it to the
following.
a
i2
a
i1
a
11
a
12
, , a
in
a
i1
a
11
a
1n
, b
i
a
i1
a
11
b
1
Here then is the code that implements these steps.
In[24]:= elimx1 s_ : Map subtractE1 s 1 , # &, Rest s
In[25]:= subtractE1 E1_, Ei_ : Rest Ei
Ei 1
E1 1
Rest E1
Finally, we will overload this version of solve so that it works like the built-in
LinearSolve; that is, it accepts a matrix of coefficients and a column vector as argu-
ments. This will avoid having to compute the transposition manually.
In[26]:= solve A_, b_ : solve Transpose Join Transpose A , b
In[27]:= solve 1, 2 , 4, 5 , 3, 6
Out[27]= 1, 2
202 An Introduction to Programming with Mathematica
Trees
Mathematica expressions can be visualized as upside-down trees; for example, f[x,y+1]
could be expressed using TreeForm.
In[28]:= TreeForm f x, y 1
Out[28]//TreeForm=
f x,
Plus 1, y
A common visualization of such an expression is by means of a picture like this.
f
x
y 1
Such structures are called trees, drawn upside-down, and they have many uses in
programming. In this section, we will discuss a way of representing trees in Mathematica,
develop some basic functions on trees, and, in the next section, give a well-known applica-
tion, Huffman encoding.
First, some terminology: Trees consist of nodes, which have labels (the symbols f, x,
+, y, and 1 in the example above) and some number of children, which are themselves
nodes (that is, the nodes labeled x and + are the children of the node labeled f). If a node
has no children, it is called a leaf; otherwise, it is an interior node. The node at the top of the
tree is called the root of the tree. In the example above, the interior nodes are the ones
labeled f and +, and the root is the node labeled f.
More specifically, we will be discussing binary trees trees in which every interior
node has two children, called the left child and right child.
fig
date
kumquat papaya
mango
We will be interested in trees whose labels are data values like numbers and strings.
The simplest way to represent them in Mathematica is to use lists: an interior node is
7 Recursion 203
represented by a three-element list containing the nodes label and its two children; a leaf
node by a one-element list containing the label. For example, the tree above is represented
as follows:
{"fig", {"date", {"kumquat"}, {"papaya"}}, {"mango"}}
Many in fact, most algorithms that operate on trees are recursive. It is natural,
because simply visiting every node in a tree is a recursive process. For example, suppose
we have a tree of strings like the fruit tree above, and we want to find the alphabetically
smallest string in the tree; that is, the first string in the lexicographic ordering of strings.
In[29]:= fruittree
"fig", "date", "kumquat" , "papaya" , "mango" ;
As usual, we should try not to think about exactly how the function works, but just
ask this question: given the minimum strings in the children of a node, how can we find the
minimum for the entire tree? Just pick the minimum among the label of this node and the
minima (recursively computed) of its children. The easiest way to find the minimum of a
collection of strings is to sort them and take the first element.
In[30]:= minInTree lab_ : lab
minInTree lab_, lc_, rc_ :
Sort lab, minInTree lc , minInTree rc 1
In[32]:= minInTree fruittree
Out[32]= date
It will be useful to have a function that determines the height of a tree, given by the
distance from the root to the farthest leaf node.
In[33]:= height lab_ : 0
height lab_, lc_, rc_ : 1 Max height lc , height rc
It would be nice to have a better way to display trees than as lists. In Chapter 9, we
will discuss the graphical display of trees, but for now we can at least print them in a nicely
indented style. To do so, we need an auxiliary function: printTree[t,k] prints t in
indented form, with the entire tree moved over k units. To put it another way, it prints t,
assuming it occurs k levels down. We have chosen, arbitrarily, to indent three spaces for
each level in the tree.
In[35]:= printTree t_ : printTree t, 0
204 An Introduction to Programming with Mathematica
In[36]:= printTree lab_ , k_ : printIndented lab, 3 k
printTree lab_, lc_, rc_ , k_ :
printIndented lab, 3 k ;
Map printTree #, k 1 &, lc, rc ;
printIndented x_, spaces_ :
Print Apply StringJoin, Table " ", spaces , x
In[39]:= printTree fruittree
fig
date
kumquat
papaya
mango
Huffman encoding
Computers represent textual information such as lists of characters, as bit strings, which are
sequences of 0s and 1s. Especially in the transmission of large amounts of data, it is impor-
tant to minimize the number of bits used to encode the text.
ASCII Codes
Character Decimal 8 Bit Binary
A 65 01000001
B 66 01000010
E 69 01000101
H 72 01001000
N 78 01001110
O 79 01001111
S 83 01010011
T 84 01010100
space 32 00100000
For simplicity, most of the time strings are represented using fixed-length codes,
those in which each character is represented by a bit string of the same length. The most
common such code, as discussed in Section 3.5, is ASCII. Each character has a number
that can be represented in 8 bits, as given in the ASCII codes table.
7 Recursion 205
For example, the string HONEST ABE is represented as the following binary
code:
0100100001001111010011100100010101010011
0101010000100000010000010100001001000101
However, this representation is far from being optimally compact. Better codes are
variable-length codes, using shorter bit strings for more common characters (just as Morse
code uses the shortest code a single dot for the most common letter in English, e.
Given a list of characters and their relative frequencies, the most compact encoding of
strings that respect those frequencies is called the Huffman encoding. David Huffman
showed how to construct this code and represent it using a tree (see Knuth 1997 or
Sedgewick 1988 for more information). We will define what Huffman encoding trees are
and show how to use them to encode and decode strings, and then show how to construct
them.
Simply put, a Huffman encoding tree is a binary tree with characters labeling the leaf
nodes. An example is shown in Figure 7.1. Note that the space (B) appears in the tree as an
ordinary character, just as it does in the ASCII code.
b
T A E
O N H B
S
Figure 7.1: A Huffman encoding tree
To use the tree to find the code for a character, look for the character in the tree and
record the sequence of branches going from the root to the character. For example, for H,
the trip is: right branch, then right, then left, and left again. Recording a 1 for a right
branch and a 0 for a left, this gives the code for H: 1100. Here, then, are the codes for all
the characters given in this tree.
206 An Introduction to Programming with Mathematica
Character Code Character Code Character Code
b 00 E 100 O 1010
A 011 H 1100 S 111
B 1101 N 1011 T 010
Note how the most common characters have shorter codes; for example, the space,
which occurs very often, uses only two bits. Of course, if we included the entire alphabet,
our tree would be much bigger, and many letters would have longer codes.
With this code, the string HONEST ABE is represented by:
110010101011100111010000111101100
We need to put some more information in our tree. To allow for efficiently finding
where a character occurs in the tree, we need to label every interior node with the set of
characters labeling leaves below it, as shown in Figure 7.2. Now we can give two programs:
one to encode character strings, and one to decode bit strings. The programs we write will
assume that Htree contains the tree in Figure 7.2.
In[40]:= Htree = {" ABEHONST", {" AT", {" "}, {"AT", {"T"}, {"A"}}},
{"BEHONS", {"EON", {"E"}, {"ON", {"O"}, {"N"}}},
{"BHS", {"BH", {"H"}, {"B"}}, {"S"}}}}
Out[40]= ABEHONST, AT, , AT, T , A ,
BEHONS, EON, E , ON, O , N , BHS, BH, H , B , S
bABEHONST
bAT
b AT
T A
BEHONS
EON
E ON
O N
BHS
BH
H B
S
Figure 7.2: A Huffman encoding tree, with interior labels
7 Recursion 207
We consider encoding character strings first. What we really need is the function to
give the bit-string encoding of a single character. Given that function call it encode
Char we can easily encode an entire string.
In[41]:= encodeString str_ : Flatten Map encodeChar, Characters str
In[42]:= encodeString "HONEST ABE"
Out[42]= encodeChar H , encodeChar O , encodeChar N ,
encodeChar E , encodeChar S , encodeChar T ,
encodeChar , encodeChar A , encodeChar B , encodeChar E
So how do we encode a single character? The method is essentially recursive: find
whether the character occurs in the left or the right subtree, recursively find its code in
that subtree, and then prepend a 0 if it was in the left or a 1 if it was in the right. For
example, consider H again: we can tell from Htree that H occurs in the right subtree;
within that subtree, its code is 100 (right, left, left); since it was in the right subtree, we
prepend a 1, to get 1100. To do this we give encodeChar two arguments, the character
and the Huffman tree.
In[43]:= encodeChar c_, _, lc_, rc_ : If stringMemberQ First lc , c ,
Join 0 , encodeChar c, lc , Join 1 , encodeChar c, rc
Here is the auxiliary function stringMemberQ.
In[44]:= stringMemberQ str_, char_ :
Length StringPosition str, char 1
In[45]:= stringMemberQ "ABE", "B"
Out[45]= True
The base case is when we reach a leaf; of course, if the character is in the tree at all
which we are assuming then it must be the label on this leaf, so we do not have to check.
In[46]:= encodeChar _, _ :
Finally, we can give a one-argument version of encodeChar that uses Htree.
In[47]:= encodeChar c_ : encodeChar c, Htree
Decoding of messages works similarly. We use the list of bits to guide our path down
the tree, and when we get to a leaf we emit that character and start over at the root.
Again, we will use a function with two arguments: the list of bits, and the tree. There are
two cases: when we are at a leaf, we have reached the end of the encoding of a character;
otherwise, we choose the left or right subtree, depending upon the next bit in the code.
208 An Introduction to Programming with Mathematica
In[48]:= decode code_, ch_ : StringJoin ch, decode code, Htree
decode 0, r___ , _, lc_, _ : decode r , lc
decode 1, r___ , _, _, rc_ : decode r , rc
decode , _ : ""
As usual, we can then give the desired one-argument form.
In[52]:= decode code_ : decode code, Htree
There is an important point to notice here: in Huffman codes, we always know when
a characters code ends. But how? The decode function breaks up the code into charac-
ters in some way, but how do we know it is the only possible way?
In fact, a bit of thought will convince you that it must be, because Huffman codes
have an interesting property: no characters code can be extended to be the code of another
character. For example, no characters code begins with 00, which is the code for space,
except space itself; and none begins with 100 except letter Es. This property implies that
our decoding algorithm finds the unique decoding of a string of bits.
Finally, we discuss how Huffman trees are constructed. This is actually very simple
and not really recursive so we will describe the method and leave the programming as an
exercise.
Keep in mind that the code for a character should be based on a set of frequencies of
the characters, given at the outset. For example, these might be the frequencies of the
characters in our example, based on their occurrences in a large body of English writing
(not just our sample phrase).
Characters Frequency
space 6
E 5
S, T, A 3
H, O, N 2
B 1
So now suppose we are given the list of characters along with their frequencies. For
purposes of the algorithm, it is better for us to think that what we have obtained is a list of
trees, each of which has only a single node, which is labeled by a letter and its frequency.
{ {{{b}, 6}}, {{{A}, 3}}, {{{B}, 1}}, {{{E}, 5}}, ...}
7 Recursion 209
Still thinking of this as a list of trees, the frequency of each character is called the
weight of the node containing that character. What we want to do is to combine these
single-node trees into larger trees, and keep doing it until they have all been joined into
one big tree. So repeatedly perform the following operation on the list of trees:
Suppose t1={{cl1,w1},} and t2 = {{cl2,w2},} are the trees in the list
with the lowest weights (that is, w1 and w2 are as small as possible)
Remove them from the list, and replace them by the single tree t={{
Join[cl1,cl2],w1+w2},t1,t2}
This operation always reduces the number of trees in the list by one. When there is
only one tree in the list, that is the Huffman encoding tree for these characters. Or rather
it is a Huffman encoding tree. The algorithm does not specify how to choose when there
are more than two trees of minimal weight, nor in which order to place those two trees
once they are chosen, so there are actually many trees that might result. Huffman proved
that they all give equally compact representations of bit strings.
Let us see how this works for our example. To make it easier to read, we will draw
the trees instead of writing them in Mathematica list notation:
1. Start with
b,6 A,3 B,1 E,5 H,2 N,2 O,2 S,3 T,3
2. Pick H and B (we could have picked N or O instead of H, but we picked H).
b,6 A,3 BH,3
H B
E,5 N,2 O,2 S,3 T,3
We have dropped the weights from the H and B nodes, since they will not contribute any
more to the algorithm.
3. Now we have to choose N and O (although we can put them in either order).
b,6 A,3 BH,3
H B
E,5 NO,4
O N
S,3 T,3
4. We have four trees of weight 3. We (arbitrarily) choose T and A.
b,6 BH,3
H B
E,5 NO,4
O N
S,3 AT,6
T A
210 An Introduction to Programming with Mathematica
5. Now we join the BH tree with the S tree.
b,6 BHS,6
BH
H B
S
E,5 NO,4
O N
AT,6
T A
6. Join E with NO.
b,6 BHS,6
BH
H B
S
ENO,9
E NO
O N
AT,6
T A
7. And b with AT.
bAT,12
b AT
T A
BHS,6
BH
H B
S
ENO,9
E NO
O N
8. BHS with ENO.
bAT,12
b AT
T A
BEHNOS,15
ENO
E NO
O N
BHS
BH
H B
S
Finally, we join the last two trees, yielding the tree shown in Figure 7.2.
Exercises
1. The Gaussian elimination procedure can fail for a variety of reasons. We have
already mentioned that it will not give good results for the Hilbert matrix, but the
reason for this is quite subtle and we will postpone our explanation to Section 8.5.
Another reason it can fail is that there may be no unique solution at all; consider, for
example, the following system.
x
1
x
2
0
2 x
1
2 x
2
0
7 Recursion 211
Here, the two equations are essentially the same, so we do not have enough informa-
tion to determine x
1
and x
2
uniquely. This problem is inherent in this system and
cannot be solved, no matter how sophisticated an algorithm we devise.
There is, however, another kind of problem that we should be able to overcome. It is
illustrated by the following system.
x
1
x
2
x
3
1
x
1
x
2
2 x
3
2
x
1
2 x
2
2 x
3
1
Our elimination procedure will produce the smaller system corresponding to the call
solve[{{0,1,1},{1,1,0}}].
x
3
1
x
2
x
3
0
This system obviously does have a solution, but solve will fail because, in attempt-
ing to eliminate x
2
, it will compute the new coefficient of x
3
as 1
1
0
, which involves
a division by 0.
The solution to this problem is easily found by observing that in any system of
equations, changing the order of the equations does not change the solution. Thus,
the above system is equivalent to:
x
2
x
3
0
x
3
1
solve has no difficulty with this system at all.
Modify solve such that it reorders the rows of its argument to ensure that a
11
is
non-0. (If every row has 0 as its first element, the system cannot be solved.) This
process of reordering the equations is called pivoting.
2. In Exercise 1 above, suppose A is known to be upper triangular, meaning it has 0s
below the diagonal (formally, a
ij
0 for all i j). Define solveUpper, having the
same arguments as solve, but under the assumption that A is upper triangular.
(This is much simpler than solve, since it requires no elimination.) Then define
solveLower, with the same arguments, but for the case where A is lower triangular
(has 0s above the diagonal). solveLower should work by manipulating A so as to
make it upper triangular, and then calling solveUpper.
212 An Introduction to Programming with Mathematica
3. Suppose we could find lower triangular and upper triangular matrices, L and U, such
that A L U. Then for any vector B, we could easily compute solve[A,B] by
computing solveUpper[U,solveLower[L,B]]. (Note that a vector X is a
solution to the original system when AX B. But this implies that L U X B, which
implies that there is a vector Y such that L Y B and U X Y; solveLower[L,B]
is Y, and solveUpper[U,Y] is X.)
So, given a square matrix A, if we can find such a decomposition of A, then we can
efficiently solve AX B for any given B. In fact, finding this so-called LU-decomposi-
tion of A is very similar to doing Gaussian elimination. Specifically, suppose that A is
the smaller matrix produced by the elimination process (that is, the coefficients in the
system S ), and suppose further that A L U , where L is lower triangular and U is
upper triangular (so L and U can be computed recursively). Then consider the
following two matrices U and L.
U is U with the first row of coefficients of A added as the top row, and 0s
added as the left column:
U
a
11
a
12
a
1 n
0
U
0
U is, of course, upper triangular.
L is L with the following changes: add the row 1, 0, 0, , 0 as the top row.
For the left column, add the multipliers computed in the elimination process;
that is, the quotients a
i1
a
11
:
L
1 0 0
a
21
a
11
L
a
n1
It can be shown that, when this construction works, as it does in the same situations
in which solve works, L U A.
Program two versions of LU-decomposition:
a. LUdecomp1[A] returns two matrices L and U, as just described. That is, it
returns a list containing these two matrices.
7 Recursion 213
b. LUdecomp2[A] returns one matrix which contains both L and U, specifically,
the matrix L I U, where I is the identity matrix. In other words, forget the
diagonal elements of L (which are all 1s) and just place the elements of L below
the diagonal and the elements of U at or above the diagonal in a single matrix.
4. Suppose you have a tree all of whose labels are numbers. Write a function to sum all
the labels.
In[1]:= numbertree 4, 5 , 6, 7 , 9, 10 , 11 ;
In[2]:= sumNodes numbertree
Out[2]= 52
5. Assume now that your trees labels are all strings. Write a function to concatenate
the strings in depth-first order. This is the order you get by following the leftmost
children of any node as far as possible before visiting their siblings on the right.
In[3]:= fruittree
"fig", "date", "kumquat" , "papaya" , "mango" ;
In[4]:= catNodes fruittree
Out[4]= figdatekumquatpapayamango
6. A tree is said to be balanced if, for every node, the heights of its children differ from
one another by no more than 1; that is, the difference in height between the taller
child and the shorter is 0 or 1. (fruittree is balanced, but numbertree from
Exercise 4 is not.) Note that the condition must hold at all nodes, not just the root.
Here is a function to test whether a tree is balanced.
In[5]:= balanced _ : True
balanced _, lc_, rc_ :
balanced lc && balanced rc && Abs height lc height rc 1
This is very expensive due to the computing of heights of subtrees. For example, it
first checks the height of the two children of the root (which involves visiting every
node in the tree except the root itself), and then it calls balanced on those two
children, which then computes the height of their children for the second time.
To avoid this extra cost, define a function balancedHeight[t] that returns a list
of two elements: the first is the height of t, and the second is a Boolean value saying
whether t is balanced. Then you can define balanced by
In[7]:= balanced t_ : balancedHeight t 2
214 An Introduction to Programming with Mathematica
7. Write a function listLevel[tr,n] which gives a list of all the labels in tree tr at
level n, where the root is at level 0, its children are at level 1, its grandchildren at level
2, and so on.
In[8]:= listLevel numbertree, 2
Out[8]= 7, 9
8. In trees of arbitrary degree, one node can have any (finite) number of children.
Represent such a tree by a list containing the label of the root and its children. For
example, consider the following tree:
World
Asia Europe America Africa
North South
It can be represented as a list:
{World, {Asia}, {Europe}, {America, {North}, {South}}, {Africa}}
Write functions minInTree, height, and printTree for trees of any degree.
9. Program the function that constructs a Huffman encoding tree, as shown in the last
part of this section.
10. Write a more efficient version of encodeString that creates a table of all the
encodings of all the characters in a given tree, then applies the table to the list of
characters. This table can be represented as a list of rewrite rules, like "a"
{0,1,1}, which can then be applied to the list of characters using ReplaceAll
(/.).
11. The MergeSort function defined in this section becomes quite slow for moderately
sized lists. Perform some experiments to determine if the bottleneck is caused mostly
by the auxiliary merge function or the double recursion inside MergeSort itself.
Once you have identified the cause of the problem, try to rewrite MergeSort to
overcome the bottleneck issues.
7 Recursion 215
7.6 Dynamic programming
Term rewriting can be used to dynamically create rewrite rules during evaluation. In a
process known as dynamic programming, a SetDelayed function whose right-hand side is
a Set function of the same name is defined.
f[x_] := f[x] = right-hand side
When an expression is pattern matched to this rewrite rule, term rewriting creates a
Set function with the specific argument value which, upon evaluation of the right-hand
side, becomes a rewrite rule. Since the global rule base is always consulted during evalua-
tion, storing results as rewrite rules can cut down on computation time, especially in
recursive computations.
In this way, dynamic programming can be described as a method in which rewrite
rules are added to the global rule base dynamically; that is, during the running of a program.
A well-known application of this is to speed up the computation of Fibonacci numbers.
The function F defined in Section 7.1 is simple, but quite expensive to execute. For
example, here is a table giving the number of additions needed to compute F[n] for
various values of n (these are the values FA
n
from Exercise 2 in Section 7.1).
n 5 10 15 20 25
F n 5 55 610 6765 75025
number of additions 7 88 986 10945 121392
Here is some code to count the number of additions in the computation of the
Fibonacci function. First, define the Fibonacci function.
In[1]:= fib 1 fib 2 1;
fib n_ : fib n 1 fib n 2
The following code initializes a counter (PlusCount), and then traces the computa-
tion of fib[10] incrementing the counter whenever the pattern Plus[__fib] is
encountered; in other words, the counter is incremented for each computation of the form
fib[x]+fib[y].
In[3]:= Module PlusCount 0 ,
TraceScan PlusCount &, fib 10 , Plus __fib ; PlusCount
Out[3]= 54
216 An Introduction to Programming with Mathematica
The reason for this excessive cost is easy to see in the course of computing F[n],
there are numbers m n for which F[m] is computed many times. For instance, F[n 2]
is computed twice (it is called from F[n] and also from F[n 1], F[n 3] three times,
and F[n 4] five times. This continual recalculation can be eliminated easily by memoriz-
ing these values as they are computed that is, by dynamic programming.
The following definition of function FF is just like the definition of F, but it adds a
rule FF[n]= F
n
to the global rule base the first time the value is computed. Since Mathe-
matica always chooses the most specific rule to apply when rewriting, whenever a future
request for FF[n] is made, the new rule will be used instead of the more general rule in
the program. Thus, for every n, FF[n] will be computed just once; after that, its value will
be found in the rule base.
In[4]:= FF 1 : 1; FF 2 : 1
FF n_ : FF n FF n 2 FF n 1
We can see the change in the trace of FF[4] as compared with that in Section 7.1
Specifically, there is only one evaluation of FF[2] now, since the second evaluation of it is
just a use of a global rewrite rule.
In[7]:= TracePrint FF 4 , FF _Integer FF _ FF _ FF _
FF 4
FF 4 FF 4 2 FF 4 1
FF 2
FF 3
FF 3 FF 3 2 FF 3 1
FF 1
FF 2
FF 3
FF 4
Out[7]= 3
7 Recursion 217
Another way to understand what is going on is to look at the global rule base after
evaluating FF[4].
In[8]:= ?FF
Global`FF
FF 1 : 1
FF 2 : 1
FF 3 2
FF 4 3
FF n_ : FF n FF n 2 FF n 1
The cost of executing this version of F is dramatically lower.
n 5 10 15 20 25
number of additions of FF n 4 9 14 19 24
Furthermore, these costs are only for the first time FF[n] is computed; in the
future, we can find FF[n] for free, or rather, for the cost of looking it up in the global rule
base.
Dynamic programming can be a useful technique, but needs to be used with care. It
will entail some increased cost in memory, as the global rule base is expanded to include
the new rules.
Exercises
1. Using dynamic programming is one way to speed up the computation of the
Fibonacci numbers, but another is to use a different algorithm. A much more effi-
cient algorithm than F can be designed, based on the following identities.
F
2 n
2 F
n 1
F
n
F
n
2
, for n 1
F
2 n 1
F
n 1
2
F
n
2
, for n 1
ProgramF using these identities.
2. You can still speed up the code for generating Fibonacci numbers by using dynamic
programming. Do so, and construct tables, like those in this section, giving the
218 An Introduction to Programming with Mathematica
number of additions performed for various n by the two programs you have just
written.
3. Calculation of the Collatz numbers, as described in Exercise 5 from Section 5.3, can
be implemented using recursion and sped up by using dynamic programming. Using
recursion and dynamic programming, create the function collatz[n,i], which
computes the ith iterate of the Collatz sequence starting with integer n. Compare its
speed with that of your original solution.
7.7 Higher-order functions and recursion
As a final wrap-up on recursion, we note that many of the built-in functions discussed in
Chapter 4 could be written as user-defined functions using recursion. Although they may
not be as efficient as the built-in functions, creating them will give you good practice with
recursion and should also give you some insight into how these functions operate.
Our first example of programming some built-in functions in a recursive style is Map.
We will call our version map. map[f ,lis] applies f to each element of the list lis. This is a
simple recursion on the tail of lis: if we assume that map[f, Rest[lis]] works,
map[f ,lis] is easily obtained from it by joining f[First[lis]] to the beginning.
In[1]:= map f_, :
map f_, x_, y___ : Join f x , map f, y
We can quickly check that our map does what it was intended to.
In[3]:= map f, 1, 2, 3
Out[3]= f 1 , f 2 , f 3
Like many of the functions in Chapter 4, this function has a function as an argument.
This is the first time we have seen user-defined higher-order functions.
We will give one more example of a built-in function that can be defined using
recursion, and leave the rest as exercises.
Nest[f ,x,n] applies f to x, n times. The recursion is, obviously, on n.
In[4]:= nest f_, x_, 0 : x
nest f_, x_, n_ : f nest f, x, n 1
Here is an example of the use of this function.
In[6]:= nest Sin, , 4
Out[6]= Sin Sin Sin Sin
7 Recursion 219
Before leaving this topic, we note that, beyond a basic exercise in recursion, it is
sometimes quite useful to write your own higher-order functions. Given a function f
whose argument must be an integer in the range 1, , 1000, and whose result is also in
that range, answer the following question: on average, for a number n
1
, how many times
can f be applied before it repeats itself? That is, on average, if we form the sequence
n
1
, n
2
f n
1
, n
3
f n
2
, , what is the smallest i such that n
i
n
j
for some j i? Assume
f is so expensive to compute that we prefer to approximate this average by just checking
ten randomly chosen numbers. This technique, known as random sampling, is used in many
areas where statistical analysis of data is required.
If we had a function repeatCount[n] to answer this question for a particular n,
then we might answer the question in this way:
Sum repeatCount Random Integer, 1,1000 , 10
10
So how do we write repeatCount? We will define our own higher-order function.
In[7]:= repeat f_, lis_, pred_ : lis ; pred Drop lis, 1 , Last lis
In[8]:= repeat f_, lis_, pred_ :
repeat f, Append lis, f Last lis , pred
repeat takes an argument list lis, and repeatedly applies f to its last element, and
adds that new value to the end, until the predicate pred returns True. repeatCount
becomes:
In[9]:= repeatCount f_, n_ : repeat f, n , MemberQ
In[10]:= plus4mod20 x_ : Mod x 4, 20
In[11]:= repeatCount plus4mod20, 0
Out[11]= 0, 4, 8, 12, 16, 0
Exercises
1. Write recursive definitions for Fold, FoldList, and NestList.
2. Recall the notion of a random walk on a two-dimensional lattice from Chapter 3.
Use repeat to define a special kind of random walk, one which continues until it
steps on to a location it had previously visited. That is, define landMineWalk as a
function of no arguments which produces the list of the locations visited in such a
random walk, starting from location (0, 0).
220 An Introduction to Programming with Mathematica
8 Numerics
Of the many data types that are available in Mathematica numbers, strings, symbols,
lists numbers are perhaps the most familiar. You can work with all kinds of numbers
in Mathematica, but, most importantly, what distinguishes it from traditional program-
ming languages and other computational systems is that with it you can operate on
numbers of any size and to any degree of precision. In this chapter we will explore
some of the issues related to working with numerical quantities and show how you can
incorporate some of these ideas into any programs that involve numerical
computations.
8.1 Introduction
One of the first things that users of Mathematica notice when they begin to use it is how
different is its treatment of numbers from other systems including calculators, traditional
programming languages, and other technical computing systems. In most traditional
programming languages, you must declare the type of number your functions can take as
an argument. Although Mathematica automatically handles such details for you, an under-
standing of the different number types and how they invoke different algorithms will be
helpful for taking full advantage of Mathematicas capabilities and writing efficient
programs.
Although you can work with both exact and approximate numbers, Mathematica
operates differently depending upon the type of input you give it.
In[1]:= Sin
4
Out[1]=
1
2
In[2]:= Sin
4.0
Out[2]= 0.707107
It is important to understand that not only are different kinds of output returned in
such cases, but Mathematica uses entirely different algorithms for these two computations.
In the first case, Mathematica looks up identities involving the sin function and multiples of
4 and applies the appropriate transformation rule to give an algebraic result. In the
second example, because a floating point number is involved in the input, a numerical
routine (a series expansion for sin) is used and the computation is carried out to insure a
result with the same precision as the input.
Another feature that is important to understand involves computations with high-pre-
cision numbers. By default, Mathematica operates on approximate numbers using a fixed
precision that is determined by the machine on which you are working.
In[3]:= N Precision
Out[3]= MachinePrecision
The number of decimal digits of precision for machine numbers is approximately 16
(we will discuss precision in detail in Section 8.3).
In[4]:= $MachinePrecision
Out[4]= 15.9546
When you need to, you can raise the number of digits of precision of the numbers
you are working with. For example, this computes to 200 digit precision.
In[5]:= N , 200
Out[5]= 3.14159265358979323846264338327950288419716939937510582097494
459230781640628620899862803482534211706798214808651328230664
709384460955058223172535940812848111745028410270193852110555
96446229489549303820
You can extend such arbitrary-precision computations to Mathematicas built-in
functions. As a simple, but illuminating example, consider the numerical solution of the
van der Pol equation x t
1
5
1 x
2
t x t x t 0 with the given initial conditions.
In[6]:= soln NDSolve x t
1
5
1 x t
2
x t x t 0,
x 0 1, x 0 0 , x, t, 0, 30
Out[6]= x InterpolatingFunction 0., 30. ,
The solution is represented as an interpolating function, one that passes through the
solution over the range for t from 0 to 30. Here is a plot of the original function evaluated
at this numerical solution, essentially giving a visual picture of the error in the numerical
solution.
222 An Introduction to Programming with Mathematica
In[7]:= Plot Evaluate x t
1
5
1 x t
2
x t x t . soln ,
t, 0, 30 , PlotRange 10
5
, 10
5
;
5 10 15 20 25 30
0.00001
5 10
6
5 10
6
0.00001
By increasing the precision of the internal algorithms used to solve this differential
equation, we can get a more precise solution.
In[8]:= soln24 NDSolve
x t
1
5
1 x t
2
x t x t 0, x 0 1, x 0 0 ,
x, t, 0, 30 , WorkingPrecision 27, PrecisionGoal 24
Out[8]= x InterpolatingFunction
0, 30.0000000000000000000000000 ,
The plot of the original function evaluated at this higher precision solution clearly
shows the higher degree of precision obtained with soln24.
In[9]:= Plot Evaluate x t
1
5
1 x t
2
x t x t . soln24 ,
t, 0, 30 , PlotRange 10
7
, 10
7
;
5 10 15 20 25 30
1 10
7
7.5 10
8
5 10
8
2.5 10
8
2.5 10
8
5 10
8
7.5 10
8
1 10
7
Working with numbers and understanding issues of precision and accuracy and the
interplay between your machines hardware and the software are essential to working with
any computational system or programming language. In this chapter we will discuss all
these issues and look at how to make your numeric computations as efficient as possible.
8 Numerics 223
8.2 Numbers
Types of numbers
There are four kinds of numbers represented in Mathematica integer, rational, real, and
complex. In addition, mathematical constants like and are treated as a special type of
number. Integers are considered to be exact and are represented without a decimal point;
rational numbers are quotients of integers and are also considered to be exact.
As mentioned in Chapter 2, numbers are atomic expressions, meaning they cannot be
broken down into smaller parts. Use the Head function to identify the type of number you
are working with.
In[1]:= Map Head, 3,
3
9
, 0.33333, 4 3.1 I,
Out[1]= Integer, Rational, Real, Complex, Symbol
Using FullForm we can see how Mathematica represents these objects internally.
In[2]:= Map FullForm, 3,
3
9
, 0.33333, 4 3.1 I,
Out[2]= 3, Rational 1, 3 , 0.33333`, Complex 4, 3.1` , Pi
As can be seen in the above example, Mathematica simplifies rational numbers to
lowest terms and leaves them as exact numbers. (We will have more to say about the
seemingly strange internal form of real numbers when we discuss their representation in
Section 8.3.)
This representation of rational (and complex) numbers as a pair of integers has one
more consequence. If you need to pattern match with rational numbers, you should be
aware of their internal representation. For example, trying to pattern match with x_/y_
will not work.
In[3]:=
3
4
.
x_
y_
x, y
Out[3]=
3
4
But pattern matching instead with Rational works fine.
In[4]:=
3
4
. Rational x_, y_ x, y
Out[4]= 3, 4
Any number containing a decimal point is classified as a real number in Mathematica.
These numbers are not considered exact and are hence often referred to as approximate
224 An Introduction to Programming with Mathematica
numbers. This often leads to confusion for new users of Mathematica. You may know that
the number 6.0 is identical to the number 6, from a mathematical perspective, but from the
perspective of the floating point unit (FPU) of your computer and as we saw in the exam-
ple above, they are quite different both in terms of their representation and in terms of the
algorithms that are used to do arithmetic with them. We will have much more to say about
this in Section 8.3.
Complex numbers are of the form a bi, where a and b are any numbers integer,
rational, or real. Mathematica represents 1 by the symbol I or .
Mathematica views complex numbers as a distinct data type, different from integers or
real numbers.
In[5]:= z 3 4
Out[5]= 3 4
In[6]:= Head z
Out[6]= Complex
You can add and subtract complex numbers.
In[7]:= z 2
Out[7]= 5 3
You can find the real and imaginary parts of any complex number.
In[8]:= Re z , Im z
Out[8]= 3, 4
The conjugate and absolute value can also be computed. The absolute value of any
number is its distance to the origin in the complex plane. The conjugate can be thought of
as the reflection of the complex number in the real axis of the complex plane.
In[9]:= Conjugate z , Abs z
Out[9]= 3 4 , 5
The phase angle is given by the argument.
In[10]:= Arg 4
Out[10]=
2
8 Numerics 225
Each of these properties of complex numbers can be visualized geometrically, as shown in
Figure 8.1.
a
Re
b
Im
z a b
Abs z
Conjugate z a b
Arg z
Figure 8.1: Geometric representation of complex numbers in the plane
For purposes of pattern matching, complex numbers are quite similar to rational
numbers. x+ Iy will not match with complex numbers. A complex number z a b is
treated as a single object for many operations, and is stored as Complex[a,b]. So to
match a complex number z, use Complex[x_,y_] (or z_Complex and Re[z] and
Im[z]) on the right-hand side of any rule you define.
Built-in constants such as , , , and Degree are not treated as real numbers by
Mathematica.
In[11]:= Head , NumberQ
Out[11]= Symbol, False
Although Mathematica does not consider constants like and like real numbers, it
does recognize that they are numerical in nature and thus you can use them more like
ordinary numbers.
In[12]:= Random Real, ,
Out[12]= 2.77033
In[13]:= Rationalize , .0001
Out[13]=
333
106
Mathematical constants have an attribute, NumericQ, that essentially alerts Mathe-
matica to the fact that they are numeric in nature.
In[14]:= Map NumericQ, , , EulerGamma, ,
Out[14]= True, True, True, False, True
226 An Introduction to Programming with Mathematica
All mathematical constants and any expressions which are explicit numbers are
considered numeric and will return a value of True when NumericQ is applied to them
(note in particular that the symbol is not numeric). When Mathematica recognizes that a
quantity has this attribute, it converts the symbol to a real number, using what it perceives
to be necessary precision.
In[15]:=
Out[15]= True
In[16]:= NumericQ
Out[16]= True
If you have to distinguish between explicit numbers and symbols that represent
numbers, then use NumberQ.
In[17]:= Map NumberQ, 3.14,
Out[17]= True, False
Digits and number bases
A list of the digits of a number can be obtained with the functions IntegerDigits or
RealDigits.
In[18]:= IntegerDigits[1293]
Out[18]= 1, 2, 9, 3
In[19]:= RealDigits N EulerGamma
Out[19]= 5, 7, 7, 2, 1, 5, 6, 6, 4, 9, 0, 1, 5, 3, 2, 9 , 0
Numbers in base 10 can be displayed in other bases by means of the BaseForm
function. For example, the following displays 18 in base 2.
In[20]:= BaseForm[18, 2]
Out[20]//BaseForm=
10010
2
The operator b^^n takes the number n in base b and converts it to base 10.
In[21]:= 2^^10010
Out[21]= 18
8 Numerics 227
The letters of the alphabet are used for numbers in bases larger than 10. For exam-
ple, here are the numbers 1 through 20 in base 16.
In[22]:= Table BaseForm j, 16 , j, 1, 20
Out[22]= 1
16
, 2
16
, 3
16
, 4
16
, 5
16
, 6
16
, 7
16
, 8
16
, 9
16
, a
16
,
b
16
, c
16
, d
16
, e
16
, f
16
, 10
16
, 11
16
, 12
16
, 13
16
, 14
16
Numbers other than integers can be represented in bases different from 10. Here are
the first few digits of in base 2.
In[23]:= BaseForm N , 5 , 2
Out[23]//BaseForm=
11.00100100010000
2
Recall that Mathematica is only displaying six significant decimal digits while storing
quite a few more. In the exercises you are asked to convert the base 2 representation back
to base 10. You will need the digits from the base 2 representation, which are obtained
with the RealDigits function.
In[24]:= RealDigits N , 2
Out[24]= 1, 1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1,
1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0,
0, 0, 0, 1, 0, 1, 1, 0, 1, 0, 0, 0, 1, 1, 0, 0, 0 , 2
The 2 in this last result indicates where the binary point is placed and can be stripped
off this list by wrapping the First function around the expression RealDigits[
N[ ],2].
Here are the first 16 decimal digits of given in base 2.
In[25]:= BaseForm N , 2
Out[25]//BaseForm=
11.00100100001111111
2
You are not restricted to integral bases such as in the previous examples. The base
can be any real number greater than 1. For example:
In[26]:= RealDigits N , N GoldenRatio
Out[26]= 1, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0, 1,
0, 1, 0, 1, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0,
0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1,
0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0 , 3
228 An Introduction to Programming with Mathematica
Random numbers
Statistical work and numerical experimentation often require random numbers to test
hypotheses. You use the Random function to generate random numbers in various ranges,
domains, and distributions.
Using Random without any arguments will generate a uniformly distributed random
real number between 0 and 1.
In[27]:= Random
Out[27]= 0.0691989
Random takes two optional arguments. The first indicates the type of number to
generate and the second argument specifies the range. For example, this generates a
random integer in the range 0 to 100.
In[28]:= Random Integer, 0, 100
Out[28]= 69
A good random number generator will distribute random numbers evenly over many
trials. For example, this generates a list of 1,000 integers between 0 and 9.
In[29]:= numbers Table Random Integer, 0, 9 , 1000 ;
Here is a plot of the frequency with which each of the digits 0 through 9 occur. We
first load the packages containing the definitions for Frequencies and BarChart.
In[30]:= Needs["Statistics`DataManipulation`"]
In[31]:= Needs["Graphics`Graphics`"]
In[32]:= BarChart Frequencies numbers ;
0 1 2 3 4 5 6 7 8 9
20
40
60
80
100
We see each of the numbers 0 through 9 occur roughly 1/10 of the time. You would
not want these numbers to occur exactly 1/10 of the time, as there would be no randomness
in this. In fact, for a uniform distribution of the numbers 0 through 9, any sequence of
1,000 digits is equally as likely to occur as any other sequence of 1,000 digits. A sequence
of 1,000 numbers that contains exactly 100 occurrences of the digit 0 followed by 100
8 Numerics 229
occurrences of the digit 1, followed by 100 occurrences of the digit 2, etc., is no more
likely than the sequence that contains 1000 7s, for example.
In addition to working with uniformly distributed random numbers (the default for
Random), you can also work with any of the built-in distributions that are defined in the
Add-ons packages, or even your own user-defined distribution. For example, suppose you
wished to work with the chi-square distribution (normal random variable with unit vari-
ance and mean about 0).
First we load the package in which this distribution is defined.
In[33]:= <<Statistics`ContinuousDistributions`
Here is a chi-square distribution with four degrees of freedom.
In[34]:= ChiSquareDistribution 4
Out[34]= ChiSquareDistribution 4
This generates an array of ten random numbers distributed according to this distribution.
In[35]:= RandomArray ChiSquareDistribution 4 , 10
Out[35]= 9.12669, 3.97231, 3.59231, 8.40731, 0.901804,
7.79067, 9.37819, 8.14669, 4.44091, 3.01975
Suppose instead of using one of the built-in distributions, you wish to generate
random numbers for a continuous distribution based on a small sample of that distribution.
For example, suppose this is the sample from a distribution from which you would
like to generate a quantile function.
In[36]:= sample 192, 155, 154, 152, 107, 149, 128, 111, 139,
108, 127, 130, 189, 119, 200, 178, 116, 180, 108, 129 ;
To construct a quantile function consistent with this sample, we need to generate some
probability points, one for each sample point and then pair them up with the sample data
points.
To generate a discrete quantile function that is consistent with the above sample, we
will first generate some probability values interpolated between the sample points, and
then pair them up with the sample points.
In[37]:= probvals N Range 0, 1,
1
Length sample 1
Out[37]= 0., 0.0526316, 0.105263, 0.157895, 0.210526,
0.263158, 0.315789, 0.368421, 0.421053, 0.473684,
0.526316, 0.578947, 0.631579, 0.684211, 0.736842,
0.789474, 0.842105, 0.894737, 0.947368, 1.
230 An Introduction to Programming with Mathematica
In[38]:= quantilepts Transpose probvals, Sort sample
Out[38]= 0., 107 , 0.0526316, 108 , 0.105263, 108 , 0.157895, 111 ,
0.210526, 116 , 0.263158, 119 , 0.315789, 127 ,
0.368421, 128 , 0.421053, 129 , 0.473684, 130 ,
0.526316, 139 , 0.578947, 149 , 0.631579, 152 ,
0.684211, 154 , 0.736842, 155 , 0.789474, 178 ,
0.842105, 180 , 0.894737, 189 , 0.947368, 192 , 1., 200
To generate a continuous quantile function, we need to interpolate through these points.
In[39]:= continuousQuantile Interpolation quantilepts
Out[39]= InterpolatingFunction 0., 1. ,
Here is a plot of this continuous quantile function.
In[40]:= Plot continuousQuantile x , x, 0, 1 ;
0.2 0.4 0.6 0.8 1
140
160
180
200
Finally, we generate 20 random numbers that are consistent with the sample.
In[41]:= Table continuousQuantile Random , 20
Out[41]= 117.556, 118.491, 112.166, 127.481, 120.256, 107.959, 138.437,
107.926, 120.28, 117.019, 137.763, 118.467, 109.176,
152.131, 198.151, 199.112, 178., 133.012, 149.839, 179.577
Exercises
1. Define a function complexToPolar that converts complex numbers to their polar
representations. Then, convert the numbers 3 3 and
3
to polar form.
2. Using the built-in Fold function, write a function convert[lis,b] that accepts a
list of digits in any base b (less than 20) and converts it to a base 10 number. For
example, 1101
2
is 13 in base 10, so your function should handle this as follows:
In[1]:= convert 1, 1, 0, 1 , 2
Out[1]= 13
8 Numerics 231
3. Write a function sumsOfCubes[n] that takes a positive integer argument n and
computes the sums of cubes of the digits of n. This exercise and the next three
exercises are excerpted from an article in The Mathematica Journal, Sums of cubes of
digits, driven to abstraction (Hayes 1992).
4. Use NestList to iterate this process of summing cubes of digits; that is, generate a
list starting with an initial integer, say 4, of the successive sums of cubes of digits. For
example, starting with 4, the list should look like: {4,64,280,520,133,}. Note,
64 4
3
, 280 6
3
4
3
, etc. Extend the list for at least 15 values and make an observa-
tion about any patterns you notice. Experiment with other starting values.
5. Prove the following statements:
a. If n has more than four digits, then sumsOfCubes[n] has fewer digits than n.
b. If n has four digits or less, then sumsOfCubes[n] has four digits or less.
c. If n has four digits or less, then sumsOfCubes[n] 4 9
3
.
d. If n is less than 2,916, then sumsOfCubes[n] is less than 2,916.
6. Write a function sumsOfPowers[n,p] that computes the sums of pth powers of n.
7. Binary shifts arise in the study of computer algorithms because they often allow you
to speed up calculations by operating in base 2 or in bases that are powers of 2. Try
to discover what a binary shift does by performing the following shift on 24 (base 10).
First get the integer digits of 24 in base 2.
In[2]:= IntegerDigits 24, 2
Out[2]= 1, 1, 0, 0, 0
Then, do a binary shift, one place to the right.
In[3]:= RotateRight %
Out[3]= 0, 1, 1, 0, 0
Finally, convert back to base 10.
In[4]:= 2^^01100
Out[4]= 12
Experiment with other numbers (including both odd and even integers) and make
some conjectures.
232 An Introduction to Programming with Mathematica
8. The survivor[n] function from Chapter 4 can be programmed using binary
shifts. This can be done by rotating the base 2 digits of the number n by one unit to
the left and then converting this rotated list back to base 10. For example, if n 10,
the base 2 representation is 1010
2
; the binary shift gives 0101
2
; converting this
number back to base 10 gives 5, which is the output to survivor[5]. Program a
new survivor function using the binary shift.
9. Simulate the throwing of two dice by defining a function rollEm that, when evalu-
ated, displays two integers between 1 and 6.
10. Experiment with creating random two-dimensional images using ListDensity
Plot.
11. A surprisingly simple pseudorandom number algorithm is the linear congruential
method. It is quite easy to implement and has been studied extensively. Sequences of
random numbers are generated by a formula such as the following:
x
n 1
x
n
b 1 mod m
The starting value x
0
is the seed, b is the multiplier, and m is the modulus. Recall that
7 mod 5 is the remainder upon dividing 7 by 5. This is represented in Mathematica as
In[5]:= Mod 7, 5
Out[5]= 2
Implement the linear congruential method and test it with a variety of numbers m
and b. If you find that the generator gets in a loop easily, try a large value for the
modulus m. (See Knuth 1997 for a full treatment of random number generating
algorithms.)
12. Write a function quadCong a, b, c, m, x
0
that implements a quadratic congruential
method, where a, b, and c are the parameters, m is the modulus, and x
0
is the starting
value. The iteration is given by:
x
n 1
a x
n
2
b x
n
c mod m
13. Numerous tests are available for determining the effective randomness of a
sequence. One of the more fundamental tests is known as the
2
(chi-square) test. It
tests to see how evenly spread out the numbers appear in the sequence and uses their
frequency of occurrence. If n is the upper bound of a sequence of m positive num-
8 Numerics 233
bers, then, in a well-distributed random sequence, we would expect about m n copies
of each number. To take into account the actual frequency with which each number
occurs, the
2
test is implemented by the formula below where the function f
i
is the
number of copies of i in the sequence. If the
2
statistic is close to n, then the
numbers are reasonably random. In particular, we will consider the sequence suffi-
ciently random if the statistic is within 2 n of n.
2 1 i n
f
i
m n
2
m n
Write a function chiSquare[lis] that takes a list of numbers and returns the
2
statistic. You will find the built-in Count function helpful for calculating the
frequencies.
14. Determine the
2
statistic for a sequence of 1000 integers generated with the linear
congruential method with m 381, b 15, and a starting value of 0.
15. John von Neumann, considered by many to be the father of computer science,
suggested a random number generator known as the middle-square method. Starting
with a ten-digit integer, square the initial integer and then extract its middle ten
digits to get the next number in the sequence. For example, starting with
1234567890, squaring it produces 1524157875019052100. The middle digits are
1578750190, so the sequence starts out 1234567890, 1578750190, 4521624250, .
Implement a middle square random number generator and then test it on a 1,000-
number sequence using the
2
test. Was the father of computer science a good
random number generator?
8.3 Working with numbers
Precision and accuracy
When you work with real numbers in any programming language, you are working with
inexact, or approximate quantities. In Mathematica, any number which contains a decimal
point is considered to be an approximate number. You can specify an approximate number
explicitly, such as 1.57, or you can get approximations to exact quantities using N.
In[1]:= e N
Out[1]= 2.71828
234 An Introduction to Programming with Mathematica
Precision of an approximate number is defined as the number of significant decimal
digits in that number. You should think of precision as giving a measure of the relative size
of the uncertainty in the value of a number. Accuracy is defined as the number of these
digits to the right of the decimal point. Accuracy can be thought of as a measure of the
absolute size of the uncertainty in the value of a number.
In[2]:= Precision e , Accuracy e
Out[2]= MachinePrecision, 15.5203
The symbol MachinePrecision (new in Version 5) is used to indicate a machine-
precision number. To see the effective precision of any machine number on your com-
puter, evaluate $MachinePrecision.
In[3]:= $MachinePrecision
Out[3]= 15.9546
The numbers that can be operated with on the hardware (on the floating point unit,
or FPU) of your computer are called machine numbers. Typically, 64 binary digits (IEEE
double floats) are needed to specify a machine number: 1 for the sign, 11 for the exponent,
and 52 for the mantissa (actually 53, since the leading one is implicitly taken as 0). A typical
value of $MachinePrecision is 64 11 log
10
2, giving machine numbers about 16
decimal digits.
In[4]:= 53 Log 10, 2 N
Out[4]= 15.9546
To say that a real number x has some uncertainty associated with its value, can be
formalized by saying that the value of x lies somewhere inside of an interval x
2
to x
2
for some uncertainty . A number with precision p is then defined to have uncertainty
x 10
p
.
In[5]:= p . Solve Abs x 10
p
, p
Solve::ifun :
Inverse functions are being used by Solve, so some
solutions may not be found; use Reduce
for complete solution information. More
Out[5]=
Log
Abs x
Log 10
In other words, the precision of a real number x is given by log
10 x
for some
uncertainty . So we could manually compute the precision of e above using an uncer-
8 Numerics 235
tainty of 10
15
, which is approximately what Mathematica assumes for machine-precision
numbers.
In[6]:= Log 10,
10
15
Abs e
Out[6]= 15.4343
On the other hand, a number with accuracy a will have uncertainty 10
a
and
hence accuracy can be expressed as log
10
.
Before discussing accuracy and precision of non-machine numbers, let us first look at
a few examples.
InputForm can be used to see how you would have to input the full number as it is
represented internally in Mathematica. The Accuracy indicates there are approximately
16 decimal digits to the right of the decimal point. The Precision indicates that this is a
machine-precision number.
In[7]:= x N EulerGamma ;
InputForm x , Accuracy x , Precision x
Out[8]= 0.5772156649015329, 16.1933, MachinePrecision
The number 123.456 is a machine-precision number, but its accuracy is reduced
because it has three digits to the left of the decimal point.
In[9]:= x 123.456;
InputForm x , Accuracy x , Precision x
Out[10]= 123.456, 13.8631, MachinePrecision
You can see more clearly how Mathematica computes Accuracy by looking at the
following example.
In[11]:= Accuracy 1.23 , Accuracy 12.3 , Accuracy 123.
Out[11]= 15.8647, 14.8647, 13.8647
Each addition of a digit to the left of the decimal point has the effect of reducing the
number of significant digits to the right of the decimal point by 1.
Representation of approximate numbers
Usually, when Mathematica displays numbers, it does so in a form that is as close to tradi-
tional mathematics as possible, printing six digits for example.
In[12]:= pi N
Out[12]= 3.14159
236 An Introduction to Programming with Mathematica
Do not assume that typing in what is displayed will result in the same value.
In[13]:= pi
Out[13]= 3.14159
In[14]:= pi 3.14159
Out[14]= 2.65359 10
6
This seemingly strange behavior the fact that pi does not appear to be equal to
3.14159 can be explained by looking at the internal representation of pi.
In[15]:= FullForm pi
Out[15]//FullForm=
3.141592653589793`
The command N[ ] causes Mathematica to first convert to a machine-precision
number, and then to display only six digits. Any computations with this number occur
using the machine precision.
Note that a number mark ` was printed at the end of the above number. This is a
machine-independent mark used to indicate that this is a machine-precision number.
When you work with numbers that are not at machine precision, this will be indicated by a
number following the number mark. For example, here is a high-precision number.
In[16]:= N , 35
Out[16]= 3.1415926535897932384626433832795029
The following shows the full internal representation of this number with the preci-
sion indicated by the 35 following the number mark.
In[17]:= FullForm %
Out[17]//FullForm=
3.141592653589793238462643383279502884197169399375`35.
Finally, note that Mathematica, in a sense, treats all machine real numbers as having
the same precision.
In[18]:= Precision 1.23
Out[18]= MachinePrecision
Although this last result may seem odd at first, it is a consequence of how Mathemat-
ica represents real numbers internally. A Precision of 16 (on a computer with $Ma
chinePrecision of 16) indicates that the number 1.23 is viewed as a machine-precision
real number which will allow Mathematica to perform arithmetic with it using the efficient
machine-precision arithmetic routines. Mathematica views the number 1.23 as a machine-
8 Numerics 237
precision real by effectively padding with 0s out to 16 significant digits. If you are uncer-
tain about the precision of the numbers you are working with, it is best to check with
Precision.
Exact vs. approximate numbers
As stated earlier, all integers and rational numbers are considered exact. You can see this by
examining the Precision of any integer or rational number.
In[19]:= Precision 7 , Precision
1
9
Out[19]= ,
Mathematica represents complex numbers similarly to rational numbers. If both the
real and imaginary parts are exact, then the complex number is treated as exact.
In[20]:= Precision 3 4 I
Out[20]=
Exact numbers have more precision than any approximate number. Representing a
number with infinite precision is another way of saying that it is exact.
In[21]:= Map Precision, 4,
1
9
, 3 4 I
Out[21]= , ,
As we saw in the example at the beginning of this chapter, this allows Mathematica to
operate on such a number differently than if the number were only approximate.
In[22]:= Cos
4
, Cos
4.0
Out[22]=
1
2
, 0.707107
But, in fact, more is true. As far as Mathematica is concerned, all integers are not
created equal.
In stark contrast to programming languages, such as C or Pascal that typically
restrict computations with integers to 16 or 32 bits (this restricts integers to a magnitude
of 2
16
in the case of 16-bit integers, or to a magnitude of 2
32
in the case of 32-bit integers),
Mathematica allows you to compute with integers and rational numbers of arbitrary size.
If two numbers are to be added, 3 6 for example, Mathematica checks to see if the
numbers can be added as machine integers. A machine integer is an integer whose magni-
tude is small enough to fit into your machines natural word size, and to be operated on by
the machines instructions, generally on its floating point processor. Word size means the
238 An Introduction to Programming with Mathematica
number of bits used to represent integers. On many computer systems, the most common
word size is 64 bits.
Arithmetic operations on integers within this range can be performed using the
machines own instructions (that is, on the that machines floating point unit), whereas
operations on integers out of that range must be done by programs, which can be less
efficient.
If the two numbers to be added are machine integers and Mathematica can determine
that their sum is a machine integer, then the addition is performed at this low level.
If, on the other hand, the two integers to be added are large and either the integers
themselves or their sum is larger than the size of a machine integer, then Mathematica
performs the arithmetic using special algorithms. Integers in this range are referred to as
extended-precision integers. For example, the following computation, although impossible to
execute on most machine floating point units, is handled by Mathematicas arithmetic
algorithms for operating on extended-precision integers.
In[23]:= 2
256
2
1024
Out[23]= 1797693134862315907729305190789024733617976978942306572734300
811577326758055009631327084773224075360211201138798713933576
587897688144166224928474306394741243777678934248654852763022
196012460941194530829520850057688381506823424628815897051997
781434325869214956932742060932172306041202803442929403375373
53777152
Rational numbers are treated somewhat similarly to integers in Mathematica since the
rational number a b can be thought of as a pair of integers, and, in fact, as we saw earlier,
it is represented as Rational[a,b]. In this way, algorithms for exact rational arithmetic
will use integer arithmetic (either machine or extended) to perform many of the necessary
computations.
High precision vs. machine precision
Real numbers (often referred to as floating point numbers) contain decimal points, and,
as mentioned above, although they can contain any number of digits, they are not consid-
ered exact.
In[24]:= Head 1.61803 , Precision 1.61803
Out[24]= Real, MachinePrecision
8 Numerics 239
In[25]:= Head 1.4987349873487454511 ,
Precision 1.4987349873487454511
Out[25]= Real, 19.1757
In a manner similar to how integers are treated Mathematica uses different internal
algorithms to do arithmetic on real numbers, depending upon whether you are using very
high precision reals or not. Whenever possible arithmetic operations on real numbers are
performed using machine-precision (fixed) reals. Real numbers that can be computed at
the hardware level of the machine are referred to as fixed precision reals, and, as stated
above, the number of digits that each machine uses for fixed-precision real numbers is
given by the system variable $MachinePrecision.
In[26]:= $MachinePrecision
Out[26]= 15.9546
One fact to keep in mind when working with machine-precision numbers is that any
computations of expressions containing machine-precision numbers will be done at the
machine precision level.
In[27]:= 2.0
100
Out[27]= 1.26765 10
30
In[28]:= Precision %
Out[28]= MachinePrecision
So, if a machine-precision number is added to a high-precision number, Mathematica
will perform the computation at the lower, machine precision.
In[29]:= Precision 2.1 3.1111111111111111111111
Out[29]= MachinePrecision
Here are the limits on the size of machine numbers that you can work with.
In[30]:= $MaxMachineNumber, $MinMachineNumber
Out[30]= 1.79769 10
308
, 2.22507 10
308
To get a sense of the limit given by $MaxMachineNumber, note that this limit is
essentially given by 2
1023
1.111111 (53 total binary digits), a number just smaller than
2
1024
. The number 53 comes from the number of binary digits that are used to specify the
mantissa for any floating point number.
In[31]:= N 2
1024
Out[31]= 1.797693134862316 10
308
240 An Introduction to Programming with Mathematica
In[32]:= 1.1111111111111111111111111111111111111111111111111111 2^1023
Out[32]= 9.987184082568421709607251059939026297877649883012814 10
307
In[33]:= $MaxMachineNumber
Out[33]= 1.79769 10
308
Although there is a limit to the magnitude of the machine-precision numbers on any
given computer, you can still compute with numbers outside of this range. Real numbers
larger than machine-precision reals are referred to as multiple precision reals and arithmetic
on such numbers is called multiple precision arithmetic or variable precision floating point
arithmetic. So, for example, on a machine whose $MachinePrecision is 16 decimal
digits, computations involving real numbers with greater than 16 significant digits will be
performed using multiple-precision algorithms.
When doing exact arithmetic multiplying two integers, for example Mathematica
first checks that both numbers are in fact integers (actually, machine integers). If they are
small and do not overflow the machines arithmetic registers, then it goes ahead and
multiplies them at the hardware level. If they are large (on most machines, integers are 32
bits long), then Mathematica goes to its extended-precision algorithms and multiplies the
integers there. In either case, all work done is exact.
When doing computations on inexact numbers, Mathematica uses two different types
of arithmetic, depending upon the precision of the numbers involved. Fixed precision
floating point arithmetic is used whenever the numbers can be handled in the machines
hardware routines. Sometimes, this arithmetic is referred to as machine precision arithmetic.
In the previous section, we gave the following example.
In[34]:= Precision 1.23 , Accuracy 1.23
Out[34]= MachinePrecision, 15.8647
Mathematica has converted 1.23 to a machine floating point number and will use
machine arithmetic on it whenever possible. The Accuracy of 16 in this example indi-
cates that there are implicit trailing 0s in this number. In the following example, n has
smaller accuracy due to the fact that there are an explicit number of numbers to the right
of the decimal point and roughly speaking, for machine-precision numbers, the number of
digits to the right of the decimal plus the number of digits to the left of the decimal should
add up to the number of decimal digits given by $MachinePrecision.
In[35]:= n 12345.6789101112
Out[35]= 12345.7
8 Numerics 241
In[36]:= Precision n , Accuracy n
Out[36]= MachinePrecision, 11.8631
You can adjust the precision of numbers with SetPrecision, although you should
note that this function will not make an inexact number more exact. Consider the follow-
ing example.
In[37]:= a SetPrecision
1
3
, 30
Out[37]= 0.333333333333333333333333333333
When SetPrecision is used with exact numbers, such as integers and rational
numbers, it creates a few more bits than were asked for, 30 in this case. You can see this by
trying to increase the precision.
In[38]:= b SetPrecision a, 50
Out[38]= 0.33333333333333333333333333333333333333333333333311
When the number a was first created, the extended-precision number was repre-
sented as a finite number of binary bits, followed by infinitely many (implicit) trailing 0s.
Increasing the precision of this number uncovered the decimal digits which are not 0s. We
can see this by converting 1/3 to a binary representation and then taking a finite number of
the binary digits to convert back to base 10.
In[39]:= RealDigits N
1
3
, 2
Out[39]= 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1,
0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0,
1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 , 1
In[40]:= 2^^.01010101 FullForm
Out[40]//FullForm=
0.33203125`
Let us clear unneeded symbols.
In[41]:= Clear a, b, n, x
Roundoff errors
Precision and accuracy are affected by performing computations with inexact numbers in
ways that can be quite surprising. One such situation concerns a magnification of error due
to roundoff. This can be seen with a simple example.
242 An Introduction to Programming with Mathematica
Here is a machine-precision approximation to 2 raised to a large power.
In[42]:= N 2
200
Out[42]= 1.26765 10
30
Working with approximations necessarily introduces some error. Comparing the
machine-precision result with the exact result gives a measure of how the error is
magnified.
In[43]:= % 2
200
Out[43]= 1.74514 10
16
That is an error of over 17 thousand trillion! This loss of accuracy is typically
referred to as roundoff error. You can see how this loss gets progressively worse by repeat-
ing the above example for larger and larger exponents.
In[44]:= Table N 2
j
2
j
, j, 100, 1000, 100
Out[44]= 7.75, 1.74514 10
16
, 2.9156 10
31
,
4.38879 10
46
, 6.18671 10
61
, 8.36779 10
76
, 1.1 10
92
,
1.4105 10
107
, 1.78821 10
122
, 2.23866 10
137
In[45]:= Map Accuracy, %
Out[45]= 15.0653, 0.287242, 15.5101, 30.6878, 45.8369,
60.968, 76.0868, 91.1948, 106.298, 121.395
Recall that Accuracy[x] gives the number of significant digits to the right of the
decimal point in x. The negative values indicate that the significant digits are to the left of
the decimal point.
Of course, if you need to work with such numbers, you can increase the precision
with either N or SetPrecision. Since almost all of the digits in this particular number
are to the right of the decimal point, this effectively increases its accuracy.
In[46]:= N 2 , 100
200
Out[46]= 1.26765060022822940149670320537600000000000000000000000000000
00000000000000000000000000000000000000 10
30
Now the result has much greater accuracy.
In[47]:= 2
200
%
Out[47]= 0. 10
68
8 Numerics 243
In[48]:= Accuracy %
Out[48]= 67.596
Computing with different number types
When doing computations with numbers, Mathematica tries to work with the most general
type of number in the expression at hand. For example, when adding two rational num-
bers, the sum is a rational number, unless of course it can be reduced to an integer.
In[49]:=
34
21
2
11
Out[49]=
416
231
In[50]:=
3
4
9
4
Out[50]= 3
But, if one of the terms is a real number, then all computations are done using
real-number arithmetic Mathematica works at the lowest precision of the numbers in the
expression.
In[51]:= Precision 10
100
1.3
Out[51]= MachinePrecision
One point to keep in mind is that when a symbol is present in the expression to be
computed, Mathematica does not convert the symbol to a machine number. This ability to
perform symbolic computations is an extremely important feature that separates Mathemat-
ica from most other computer languages.
In[52]:= Simplify Sin n , n Integers
Out[52]= 0
In[53]:= Simplify Sin n N , n Integers
Out[53]= Sin 3.14159 n
When two extended-precision approximate numbers are multiplied, the precision of
the result will be the minimum of the precision of the two factors.
In[54]:= Precision N 2 , 50 N 3 , 80
Out[54]= 50.
244 An Introduction to Programming with Mathematica
In fact, whenever two numbers are multiplied, the precision of the product will be
the minimum of the precision of the factors, even if one factor is a machine precision real
number and the other factor is a high precision real number.
In[55]:= a N 2 ;
In[56]:= b N 2
99
, 30 ;
In[57]:= Precision a , Precision b , Precision a b
Out[57]= MachinePrecision, 30., MachinePrecision
For addition of real numbers, it is their accuracy that counts most. Recall, Accuracy
gives the number of significant digits to the right of the decimal point. In essence,
Accuracy[x] measures the absolute error in the number x.
In[58]:= Accuracy 1.23 , Accuracy 12.5
Out[58]= 15.8647, 14.8577
For machine-precision numbers, adding a digit to the left of the decimal point
essentially removes one digit from the right of the decimal point. These numbers have a
fixed number of digits. This is not the case though for extended-precision numbers, where
all the digits to the right of the decimal can be considered significant.
In[59]:= Accuracy 123.4444444444444444444444444444
Out[59]= 28.
In[60]:= Accuracy 12321.4444444444444444444444444444
Out[60]= 28.
In an analogous manner to the use of Precision with multiplication, the Accu
racy of an addition will be the minimum of the accuracies of the summands.
In[61]:= Accuracy 1.23 12.3
Out[61]= 14.8233
In[62]:= Accuracy 12.3
Out[62]= 14.8647
This last point can lead to some unexpected results if you are not careful.
In[63]:= 1.0 10
25
Out[63]= 1.
8 Numerics 245
In[64]:= Accuracy %
Out[64]= 15.9546
The number 1.0 is a machine number, so this computation was performed using
machine accuracy, hence the 1 in the 25th decimal place to the right in the number 10
25
was lost when this computation was performed in machine arithmetic. You can avoid
machine arithmetic and get the intended result by extending the precision of 1.0 to 25
digits.
In[65]:= 1.0`25 10
25
FullForm
Out[65]//FullForm=
1.0000000000000000000000001`25.
In[66]:= Accuracy %
Out[66]= 25.
Exercises
1. Explain why Mathematica is unable to produce a number with 100 digits of precision
in the following example.
In[1]:= N 1.23, 100
Out[1]= 1.23
In[2]:= Precision %
Out[2]= MachinePrecision
2. Determine what level of precision is necessary when computing N 2 , prec
200
to
produce accuracy in the output of at least 100 digits.
3. Explain why the following computation produces an unexpected result (that is, why
the value 0.000000000001 is not returned).
In[3]:= 1.0 0.999999999999
Out[3]= 9.99978 10
13
4. How close is the number
163
to an integer? Use N, but be careful about the
precision of your computations.
246 An Introduction to Programming with Mathematica
8.4 Working with arrays of numbers
Scientists, engineers, and anyone who works with numbers typically do so in the context of
arrays of data. In many applications these arrays can become quite large and hence pose
special problems when computing with them. Mathematica uses two special data types to
make computations with arrays faster and more efficient sparse arrays and packed arrays.
In this section we will introduce each of these data types and see how a working knowledge
of them can help you work with very large sets of data.
Sparse arrays
In many applications, particularly solving ordinary and partial differential equations,
optimization problems, and solving large systems of equations, it is not uncommon to
work with very large matrices that have mostly 0s as elements. Such matrices or arrays are
referred to as sparse and many optimized algorithms have been developed for working with
such objects. These algorithms allow you to work with arrays that are often several orders
of magnitude larger than dense arrays and generally at speeds that are several orders of
magnitude faster.
Sparse arrays are created with the SparseArray function. The first argument to
SparseArray specifies the rules to be used to create the non-0 elements and the second
argument specifies the dimensions of the array.
For example, this creates a 5 5 sparse array object with elements on the diagonal
equal to 1.
In[1]:= spmat SparseArray i_, i_ 1, 5, 5
Out[1]= SparseArray 5 , 5, 5
Wrapping Normal around a sparse array object converts it into a list of lists, which
can then be displayed in a traditional form with MatrixForm.
In[2]:= Normal spmat MatrixForm
Out[2]//MatrixForm=
1 0 0 0 0
0 1 0 0 0
0 0 1 0 0
0 0 0 1 0
0 0 0 0 1
Here are the rules associated with this sparse array object. Notice that in addition to
the explicit rules we specified, Mathematica uses the rule {_,_} 0 for the default cases;
that is, any element not explicitly specified by a rule should be set to 0.
8 Numerics 247
In[3]:= ArrayRules spmat
Out[3]= 1, 1 1, 2, 2 1, 3, 3 1,
4, 4 1, 5, 5 1, _, _ 0
Using a third argument to SparseArray, you can specify that the implicit elements
are other than 0.
In[4]:= spmat2 SparseArray i_, i_ 1, 5, 5 , 13
Out[4]= SparseArray 5 , 5, 5 , 13
In[5]:= Normal spmat2 MatrixForm
Out[5]//MatrixForm=
1 13 13 13 13
13 1 13 13 13
13 13 1 13 13
13 13 13 1 13
13 13 13 13 1
Here is a slightly more complicated specification for the rules associated with a
sparse array. In this example, the diagonal elements are 1, and the elements whose vertical
and horizontal positions differ by 1 will be 2.
In[6]:= spmat3 SparseArray
i_, i_ 1, i_, j_ ; Abs i j 1 2 , 5, 5
Out[6]= SparseArray 13 , 5, 5
In[7]:= MatrixForm Normal spmat3
Out[7]//MatrixForm=
1 2 0 0 0
2 1 2 0 0
0 2 1 2 0
0 0 2 1 2
0 0 0 2 1
248 An Introduction to Programming with Mathematica
Here is a simple pictorial representation of a sparse array using ArrayPlot.
In[8]:= ArrayPlot spmat3 ;
Using a larger array, you can clearly see the nature of the sparseness of values.
In[9]:= ArrayPlot SparseArray i_, j_ ; Abs i j 2 1, 100, 100 ;
Let us take a look at some computations with sparse arrays to see how speed and
memory issues are affected. First we create a 100000 100000 sparse array with random
numbers on and just off the diagonal, and 0s everywhere else.
In[10]:= mat SparseArray
i_, j_ ; Abs i j 2 Random , 10
5
, 10
5
Out[10]= SparseArray 499994 , 100000, 100000
Here is a vector consisting of 100,000 random numbers.
In[11]:= b Table Random , 10
5
;
First, note the difference in size of this sparse array compared with a dense array.
The sparse array takes up approximately six megabytes.
In[12]:= sparseMemory N ByteCount mat Byte
Out[12]= 6.4003 10
6
Byte
8 Numerics 249
The corresponding dense array would require 80 gigabytes to store.
In[13]:= N 10
5
10
5
8 Byte
Out[13]= 8. 10
10
Byte
Computations involving this sparse linear system are extremely fast.
In[14]:= Timing LinearSolve mat, b ;
Out[14]= 0.802 Second, Null
In[15]:= Timing mat.mat;
Out[15]= 0.09 Second, Null
Packed arrays
One of the great advantages of the Mathematica programming language is that it seamlessly
handles the administrative tasks of dealing with a wide variety of data types. So for exam-
ple, when you perform computations with floating point numbers, Mathematica determines
the type of numbers you are working with and then chooses to perform the computation
either on your machines floating point processor (if working with numbers that fit there)
or does the computation using extended-precision software routines. Similarly computa-
tions involving integers will be done in hardware or using special software routines depend-
ing upon the size of the integers relative to your machines hardware constraints.
But all this comes at a cost, and the cost involves the administrative overhead neces-
sary to determine the appropriate routine and whether to perform the computation in
hardware or software. For small computations, this overhead is not noticeable, but for
large computations involving tens of thousands of rows and columns of a matrix, say, this
overhead could start to slow things down.
Fortunately, there is a way to bypass some of this overhead and get significant speed
improvements together with a smaller memory footprint. The technology that does this is
referred to as packed arrays and they are fairly simple to understand. Whenever possible
Mathematica will represent a list of a single type of machine numbers (integer, real, or
complex) as an array, in fact, a packed array object. So a matrix consisting of all machine
real numbers will be represented internally as a packed array. This internal representation
is transparent to the user.
Here is a 1000 1000 array consisting of random real numbers.
In[16]:= mat Table Random , 1000 , 1000 ;
250 An Introduction to Programming with Mathematica
Mathematica recognizes that this array consists entirely of machine numbers and so it
packs the array automatically.
In[17]:= Developer`PackedArrayQ mat
Out[17]= True
Let us also create an array that is not packed. We can do this by replacing one of the
elements in mat with a number that is not a machine floating point number. Here we
replace the element in the first row, second column of mat with a 1.
In[18]:= mat2 ReplacePart mat, 1, 1, 2 ;
In[19]:= Developer`PackedArrayQ mat2
Out[19]= False
The first thing to notice is the memory savings obtained by using packed arrays.
In[20]:= Map ByteCount, mat, mat2
Out[20]= 8000060, 20036032
In this example, it takes 60% less memory to store the packed array over the similar
unpacked array.
In[21]:=
20036032 8000060
20036032
N
Out[21]= 0.600716
The time to compute the minimum value is roughly an order of magnitude faster for the
packed array.
In[22]:= Map Timing Min # ; &, mat, mat2
Out[22]= 0.01 Second, Null , 0.15 Second, Null
Simple arithmetic on such objects is also significantly sped up with packed arrays.
In[23]:= Timing Do mat mat, 100 ;
Out[23]= 3.816 Second, Null
In[24]:= Timing Do mat2 mat2, 100 ;
Out[24]= 59.685 Second, Null
When packed arrays are used in Mathematica, the compiler is invoked, thus generally
improving the time it takes for the computation to take place. Many of the built-in func-
tions are designed to take advantage of the packed array technology. But they do not
invoke the compiler whenever the time it takes to compile is close to the running time of
8 Numerics 251
the computation itself. There are length limits on many common Mathematica functions
that determine whether the compiler will be used or not. For example, the length limit for
Table is 250.
In[25]:= m1 Table Random , 249 ;
Developer`PackedArrayQ m1
Out[26]= False
In[27]:= m2 Table Random , 250 ;
Developer`PackedArrayQ m2
Out[28]= True
For NestList, it is 100 (remember that NestList[f,init,n] produces a list of
n 1 elements because it prepends the initial value to the list of iterates).
In[29]:= n1 NestList Sin, .5, 98 ;
Developer`PackedArrayQ n1
Out[30]= False
In[31]:= n2 NestList Sin, .5, 99 ;
Developer`PackedArrayQ n2
Out[32]= True
These length limits are all system parameters that can be set with SystemOptions.
In[33]:= Developer`SystemOptions "CompileOptions"
Out[33]= CompileOptions ApplyCompileLength ,
ArrayCompileLength 250, AutoCompileAllowCoercion False,
AutoCompileProtectValues False,
AutomaticCompile False, CompileAllowCoercion True,
CompileConfirmInitializedVariables True,
CompiledFunctionArgumentCoercionTolerance 2.10721,
CompileEvaluateConstants True,
CompileReportCoercion False,
CompileReportExternal False, CompileReportFailure False,
CompileValuesLast True, FoldCompileLength 100,
InternalCompileMessages False, MapCompileLength 100,
NestCompileLength 100, NumericalAllowExternal True,
SystemCompileOptimizations All, TableCompileLength 250
So how do you best take advantage of packed arrays when you write your code? First,
it is important that you insure that your lists and arrays consist of machine numbers all of
the same type integer, real, or complex. In addition, whenever possible, try to operate on
lists and arrays all at once instead of looping through your arrays. Listable operations with
252 An Introduction to Programming with Mathematica
packed array input will use the compiler and will produce packed array output. Fortu-
nately, many of the commonly used functions have this attribute.
In[34]:= names Select Names "System` " ,
MemberQ Attributes # , Listable & ;
Here we display a representative sample of the symbols that have this attribute.
In[35]:= Take names, 1, Length names , 10
Out[35]= Abs, ArcCsc, Attributes, BitNot, Conjugate, Csch, EllipticPi,
ExpIntegralE, Fibonacci, Hypergeometric0F1, IntegerDigits,
LegendreQ, MathieuCharacteristicExponent, NonNegative,
PolyLog, Quotient, Sign, StringLength, ToUpperCase
Exercises
1. Create a function RandomSparseArray[n] that generates an n n sparse array
with random numbers along the diagonal.
2. Create a function tridiagonalMatrix[n,p,q] that creates an n n matrix with
the integer p on the diagonal, the integer q on the upper and lower subdiagonals, and
0s everywhere else.
3. Create a vector vec consisting of 100,000 random real numbers between 0 and 1.
Check that it is indeed a packed array by using Developer`PackedArrayQ. Then
replace one element in vec with an integer. Check that this new vector is not a
packed array. Finally, perform some memory and timing tests on these two vectors.
8.5 Numerical computations
Mathematicas built-in numerical functions are designed to guarantee the accuracy of their
results as much as possible and they are optimized to minimize the work done to generate
those results. Functions such as N, FindRoot, NDSolve, NMinimize, and NIntegrate
use options to allow you to adjust their behavior and get finer control over precision,
accuracy, and other internal aspects of the underlying numerical routines.
In this section we will first look at how to use these options to control the precision
and accuracy of your results. We will then discuss how to incorporate these options into
your own numerical functions. Finally, we will look at a numerical problem that is mathe-
8 Numerics 253
matical in nature, Gaussian elimination, and see how adjusting the underlying algorithms
can help avoid roundoff and division-by-0 errors.
Working with precision and accuracy
When you do computations with Mathematicas numerical functions, results are returned
at the default machine precision.
In[1]:= NIntegrate Sin x
2
, x, 0,
Out[1]= 0.894831
In[2]:= Precision %
Out[2]= MachinePrecision
When you need results with higher precision you will need to change the option
PrecisionGoal, which essentially sets the desired precision of the result (similarly for
accuracy, with AccuracyGoal).
Here is the same computation as above, but asking for 30 digits of precision in the
result.
In[3]:= NIntegrate Sin x
2
, x, 0, , PrecisionGoal 30
NIntegrate::tmap :
NIntegrate is unable to achieve the tolerances specified
by the PrecisionGoal and AccuracyGoal options
because the working precision is insufficient. Try
increasing the setting of the WorkingPrecision option.
Out[3]= 0.894831
Mathematica is complaining that it is unable to produce a result with the requested
precision. If you look at the default value of WorkingPrecision, you will see that it is
set to MachinePrecision. This means that the internal algorithms will work at
machine precision. But, in this example, that was not sufficient to guarantee a result with
much higher precision.
In[4]:= Options NIntegrate
Out[4]= AccuracyGoal , Compiled True,
EvaluationMonitor None, GaussPoints Automatic,
MaxPoints Automatic, MaxRecursion 6, Method Automatic,
MinRecursion 0, PrecisionGoal Automatic,
SingularityDepth 4, WorkingPrecision MachinePrecision
254 An Introduction to Programming with Mathematica
To insure that the PrecisionGoal is met, we need to increase the WorkingPre
cision a bit above the PrecisionGoal.
In[5]:= NIntegrate Sin x
2
, x, 0, ,
PrecisionGoal 30, WorkingPrecision 36
Out[5]= 0.894831469484144958801022013417
In[6]:= Precision %
Out[6]= 30.3742
How much to increase the value of WorkingPrecision above that of Precision
Goal is a bit dependent upon the problem at hand, but a good rule of thumb is to start by
setting WorkingPrecision about 1015% higher than your PrecisionGoal.
Another option to numerical functions that is important to understand is MaxItera
tions. As its name implies, this is the maximum number of iterations that a given iterative
function will perform in doing its computation. For example, the default value of MaxIter
ations in FindRoot is 100.
In[7]:= Options FindRoot
Out[7]= AccuracyGoal Automatic, Compiled True,
DampingFactor 1, EvaluationMonitor None,
Jacobian Automatic, MaxIterations 100,
Method Automatic, PrecisionGoal Automatic,
StepMonitor None, WorkingPrecision MachinePrecision
For many computations, this limit will be sufficient. But with root finding for exam-
ple, a function that is very flat near the desired zero may need a higher number of itera-
tions to find that zero. For example, the function x
11
has a root at 0 of course, but Find
Root has difficulty locating it and is unable to guarantee its precision and accuracy using
the default settings.
In[8]:= FindRoot x
11
, x, 0.5
FindRoot::cvmit :
Failed to converge to the requested accuracy
or precision within 100 iterations. More
Out[8]= x 0.0000362829
If you increase the value of MaxIterations, you will get a more accurate result.
In[9]:= FindRoot x
11
, x, 0.5 , MaxIterations 1000
Out[9]= x 9.84816 10
8
8 Numerics 255
To get even more accuracy, try increasing AccuracyGoal. As discussed above, you
should increase the value of the WorkingPrecision option as well.
In[10]:= FindRoot x
11
, x, 0.5 , AccuracyGoal 30,
WorkingPrecision 36, MaxIterations 1000
Out[10]= x 9.38526423859658090604961544893338306 10
30
One final option to Mathematicas numerical functions that we will explore is Evalua
tionMonitor. This option can be used to evaluate an expression during the computation
of the function for which it is an option. For example, suppose you would like to see all of
the intermediate values that FindRoot comes up with during its computation.
You could simply print the values that x takes on throughout the computation using
a Print statement.
In[11]:= FindRoot Sin x , x, 2.0 , EvaluationMonitor Print x
2.
4.18504
2.46789
3.26619
3.14094
3.14159
3.14159
Out[11]= x 3.14159
This approach suffers from the fact that the Print expression produces no output
and so there is no direct way to access these intermediate values. A better approach would
be to append the intermediate values to a list. In the following example we initialize an
empty list xtemp and use EvaluationMonitor to append values of x to that list
throughout the course of the root-finding computation.
In[12]:= xtemp ;
FindRoot Sin x , x, 2.0 ,
EvaluationMonitor AppendTo xtemp, x
Out[13]= x 3.14159
The intermediate values are now stored in xtemp.
In[14]:= xtemp
Out[14]= 2., 4.18504, 2.46789, 3.26619, 3.14094, 3.14159, 3.14159
256 An Introduction to Programming with Mathematica
Note the use of the delayed rule above with EvaluationMonitor. This ensures
that the right-hand side of the rule is not evaluated before FindRoot starts its
computation.
Newtons method revisited
In Section 5.2 we wrote a program to implement Newtons method for finding roots of
equations.
In[15]:= findRoot fun_, init_, _ :
Module xi init, funxi fun init , df fun' ,
While Abs funxi ,
xi N xi
funxi
df xi
;
funxi fun xi ;
xi
One of the limitations of this implementation is that the user has little control over
the precision or accuracy of the results. In addition, although the loop will continue until
values are within of the root, there is no mechanism for automatically adjusting this
tolerance, nor for controlling the number of iterations that are performed. In this section
we will rewrite this root-finding function to take advantage of the options for numerical
functions that control precision and accuracy.
First we will change the iterative structure from a While loop to a fixed point
iteration. The first argument to FixedPoint is the function that we are iterating, so that
will be the same as the function above, namely, x
i
f x
i
f x
i
. The second argument to Fixed
Point is the initial value for the iteration. The third argument is the number of iterations.
So, using a pure function for the first argument, the Newton iteration will look like this:
FixedPoint #
fun #
fun' #
&, initx, maxIterations
Let us set up the needed options with some default values. We will call our new
program newton.
In[16]:= Options newton
MaxIterations $RecursionLimit,
PrecisionGoal Automatic,
WorkingPrecision Automatic
;
8 Numerics 257
We have set the default value of MaxIterations to be $RecursionLimit
(normally 256) using a delayed rule so that $RecursionLimit is not evaluated until the
option is called. PrecisionGoal and WorkingPrecision are set to Automatic,
which, at the moment, has no value associated with it. In the body of our function, we will
take a value of Automatic for PrecisionGoal to mean a precision that is equal to the
precision of the initial value passed to newton.
If[precisionGoal===Automatic, precisionGoal = Precision[init]];
As we saw in the previous section, we will need to bump up the value of Working
Precision to something a little bigger than PrecisionGoal. We will set it to be ten
more digits than the precision goal.
If[workingPrecision === Automatic,
workingPrecision = precisionGoal + 10];
initx = SetPrecision[init, workingPrecision];
Here then is the definition of newton with these added pieces.
In[17]:= newton fun_, init_?NumericQ, opts___?OptionQ :
Module maxIterations, precisionGoal,
workingPrecision, initx, df fun' ,
maxIterations, precisionGoal, workingPrecision
MaxIterations, PrecisionGoal, WorkingPrecision .
Flatten opts . Options newton ;
If precisionGoal Automatic,
precisionGoal Precision init ;
If workingPrecision Automatic,
workingPrecision precisionGoal 10 ;
initx SetPrecision init, workingPrecision ;
SetPrecision
FixedPoint #
fun #
df #
&, initx, maxIterations ,
precisionGoal
Let us use newton to find the roots of various functions.
In[18]:= f x_ : x
2
2
In[19]:= newton f, 1.0
Out[19]= 1.41421
258 An Introduction to Programming with Mathematica
The precision of this result is the same as the precision of the initial guess.
In[20]:= Precision %
Out[20]= MachinePrecision
Setting PrecisionGoal higher generates a high-precision result.
In[21]:= newton Sin,
14
10
, PrecisionGoal 40
Out[21]= 3.141592653589793238462643383279502884197
In[22]:= %
Out[22]= 0. 10
40
There are still a number of problems that can arise with our implementation of
Newtons method. First is the possibility that the derivative of the function we are working
with might be equal to 0. This will produce a division-by-0 error. Another type of diffi-
culty that can arise in root finding occurs when the derivative of the function in question is
either difficult or impossible to compute. As a very simple example, consider the function
x 3 , which has a root at x 3. Both the built-in function FindRoot and our user-de-
fined newton will fail with this function since a symbolic derivative cannot be computed.
In[23]:= D Abs x 3 , x
Out[23]= Abs 3 x
One way around such problems is to use a numerical derivative (as opposed to an
analytic derivative). The secant method approximates f x
k
using the difference quotient:
f x
k
f x
k 1
x
k
x
k 1
Although this will require two initial values to start, it has the advantage of not
having to compute symbolic derivatives. Here is a simple implementation using a While
loop.
In[24]:= secant f_, a_, b_ :
Module x1 a, x2 b, df , While Abs f x2
1
10
10
,
df
f x2 f x1
x2 x1
;
x1, x2 x2, x2
f x2
df
;
x2
8 Numerics 259
In[25]:= f x_ : Abs x 3
In[26]:= secant f, 3.1, 1.8
Out[26]= 3.
In the exercises, the reader is asked to refine this program by writing it in a func-
tional style and including mechanisms to gain finer control over precision and accuracy in
a manner similar to what we did with the newton function earlier in this section.
Gaussian elimination revisited
When solving the linear system Ax b by numerical techniques, several types of problems
may arise. One problem, roundoff error, sometimes occurs when using machine numbers
as opposed to exact numbers. For many matrices A, there is little propagation of roundoff
error. But for some matrices the error tends to magnify in a startling way and can lead to
highly inaccurate results. Such matrices are called ill-conditioned, and, in this section, we
will identify ill-conditioned matrices and discuss what to do about them when doing
numerical linear algebra.
Another type of problem that can occur was first mentioned in Section 7.5 where we
used Gaussian elimination to solve the system Ax b. In the exercises at the end of that
section, we gave a very brief discussion of the conditions under which the method might
fail, namely, division by 0. In this section we will give a more detailed treatment of the
potential pitfalls with Gaussian elimination.
Since the method of Gaussian elimination is essentially list manipulation involving
additions, subtractions, multiplications, and divisions, clearly one avenue of failure would
be if we were to divide by 0. We formed what are commonly called multipliers (
Ei 1
E1 1
, in
the example below) as follows:
subtractE1 E1_, Ei_ : Rest Ei
Ei 1
E1 1
Rest E1
If the element E1[[1]] were ever equal to 0, the method would fail. Recall the
example from the exercises at the end of Section 7.5.
In[27]:= m 0, 3 , 3, 0 ;
b 5, 6 ;
260 An Introduction to Programming with Mathematica
This simple linear systemm.x b has solution vector x 2,
5
3
.
In[29]:= m. 2,
5
3
Out[29]= 5, 6
Unfortunately, the solve command we developed earlier in Section 7.5 will fail on this
linear system.
In[30]:= solve[m, b]
Power::infy : Infinite expression
1
0
encountered. More
::indet :
Indeterminate expression 0 ComplexInfinity encountered. More
Power::infy : Infinite expression
1
0
encountered. More
Out[30]= Indeterminate, Indeterminate
It is pretty clear that our solve command has not been written to take this situation
into account. The problem can be remedied as suggested in Exercise 1 in Section 7.5, by
interchanging rows (equations) so that the 0 element is not in this pivoting position.
Interchanging rows is equivalent to swapping equations, so this will not change the solu-
tion of the system in any way.
However, there is another problem that can arise when solving systems containing
finite precision coefficients. The imprecision of the coefficients tends to become magnified
in performing the necessary arithmetic. We can see this more clearly with an example.
Suppose we were using six-digit rounded arithmetic on the following system.
0.000001 1.0
1.0 1.0
x
y
1.0
2.0
The augmented matrix would look like the following.
0.000001 1.0 1.0
1.0 1.0 2.0
Gaussian elimination would start solving this system by multiplying the first row by 10
6
(which contains seven digits) and subtracting from the second row. But six-digit rounded
arithmetic would then produce:
8 Numerics 261
0.0000001 1.0 1.0
0.0 1000000. 1000000
Dividing the second row by 1000000. gives the solution for y.
0.000001 1.0 1.0
0.0 1. 1.
The second part of the back substitution gives the solution for x; that is,
1. 1.0 0.000001.
1.00000 0.0 0.0
0.0 1. 1.
This solution x, y 0, 1 , is in fact, not the least bit close to the correct answer. A
much more accurate solution is given by the ordered pair
x, y 1.000001000001, 0.999998999999 .
What has gone wrong? In general, accuracy is lost when the magnitude of the
pivoting position is small compared with the remaining coefficients in that column. Pivot-
ing can be used to avoid two situations. First, it is used to avoid a 0 element, when the
matrix is nonsingular. A square matrix A is said to be nonsingular if it has an inverse; that is,
if there exists a matrix B such that AB I.
Pivoting is also used to minimize the potential for roundoff errors. It does this by
selecting the element from the remaining rows (equations) that is the maximum in absolute
value. This will make the multiplier small and will have the effect of reducing possible
roundoff errors. The following code selects this pivot and reorders the rows of the system
accordingly.
In[31]:= pivot S_ : Module p, ST1 ,
ST1 Abs Transpose S 1 ;
p Position ST1, Max ST1 1, 1 ;
Join S p , Delete S, p
Now the original solve function can be rewritten to pivot on this non-0 element.
The new function is called solvePP (for partial pivot).
In[32]:= subtractE1 E1_, Ei_ : Rest Ei
Ei 1 Rest E1
E1 1
;
In[33]:= elimx1 T_ : Map subtractE1 T 1 , #1 &, Rest T ;
262 An Introduction to Programming with Mathematica
In[34]:= solvep a11_, b1_ :
b1
a11
;
In[35]:= solvep S_ : Module S1 pivot S , E1, a12toa1n, x2toxn ,
x2toxn solvep elimx1 S1 ;
E1 First S1 ;
a12toa1n Drop Rest E1 , 1 ;
Join
Last E1 a12toa1n.x2toxn
First E1
, x2toxn ;
In[36]:= solvePP mat_, b_ :
solvep Transpose Append Transpose mat , b
As we did in Section 7.5, we set things up so that the user can simply pass the matrix
mat and column vector b as arguments, and solvePP will form the augmented matrix in
the call to solvep on the last line above.
We can quickly see how partial pivoting solves our first problem of division by 0.
Solving the system given earlier with this new function now gives the correct result.
In[37]:= m 0, 3 , 3, 0 ;
b 5, 6 ;
In[39]:= solvePP m, b
Out[39]= 2,
5
3
The problem with roundoff error can best be seen by constructing a matrix that
would tend to produce quite large intermediate results relative to its original elements.
One such class of matrices are referred to as ill-conditioned matrices, a complete study of
which is outside the scope of this book. The reader is encouraged to consult Skeel and
Keiper 1993 or Burden and Faires 2000 for a comprehensive discussion of ill-conditioning.
A set of classically ill-conditioned matrices are the Hilbert matrices which arise in
numerical analysis in the solution of what are known as orthogonal polynomials. Recall the
definition of the nth degree Hilbert matrix that we gave in Section 7.5.
In[40]:= HilbertMatrix n_ : Table
1
i j 1
, i, n , j, n
In[41]:= HilbertMatrix 3 MatrixForm
Out[41]//MatrixForm=
1
1
2
1
3
1
2
1
3
1
4
1
3
1
4
1
5
8 Numerics 263
We will use the Hilbert matrices, but, instead of working with exact arithmetic, we
will work with floating point numbers.
In[42]:= N HilbertMatrix 3 MatrixForm
Out[42]//MatrixForm=
1. 0.5 0.333333
0.5 0.333333 0.25
0.333333 0.25 0.2
To compare the simple solver solve and the partial pivoting solver solvePP along
with the built-in LinearSolve we first construct a 25 25 Hilbert matrix and a random
25 1 column vector (and, of course, suppress the display of the 625 elements of the matrix
and 25 elements of the column vector).
In[43]:= h25 N HilbertMatrix 25 ;
In[44]:= b25 Table Random , 25 ;
Now let us use each of these three methods to find the solution vector x of the
system h25.x = b25. We also give a measure of the total error involved in each case by
computing the difference between h25.x and b25.
LinearSolve fails on this linear system.
In[45]:= xLS LinearSolve h25, b25 ;
LinearSolve::luc :
Result for LinearSolve of badly conditioned
matrix 1., 0.5, 0.333333, 0.25, 0.2, 20 ,
0.142857, 0.125, 0.111111, 0.1, 15 , 10
may contain significant numerical errors. More
The solve function (without pivoting) solves the system and produces a total error of
about 161 (this result will vary depending upon the random vector b25).
In[46]:= xGE solve h25, b25 ;
In[47]:= totalerrorGE Total Abs h25.xGE b25
Out[47]= 161.552
Here we compute the solution to this system using partial pivoting.
In[48]:= xPP solvePP h25, b25 ;
In[49]:= totalerrorPP Total Abs h25.xPP b25
Out[49]= 12.8731
It is no surprise that our initial implementation of Gaussian elimination, solve, had
a greater total error than solvepp. As we mentioned above, the Hilbert matrices are very
264 An Introduction to Programming with Mathematica
ill-conditioned and so we would expect that roundoff error would be more significant
without pivoting. (As noted above, results will vary from machine to machine and from
session to session since each evaluation of b25 above will produce a different column
vector.)
The importance of these numbers is that they tell us that there can be a significant
increase in error in using Gaussian elimination without pivoting. We have to be a bit
careful in reading too much into that though. Quite a bit of roundoff error is present in
these results. You should check that this is in fact the case by running the examples with
smaller Hilbert matrices. The exercises outline a method to help reduce such potential
roundoff error.
Exercises
1. The newton function developed in this section suffers from several inefficiencies.
One of them is that if the precision goal is no more than machine precision, all
intermediate computations should be done at the more efficient machine precision as
well. Modify newton so that it will operate at machine precision if the precision goal
is at most machine precision.
2. In the newton program, we added SetPrecision[result,precisionGoal]
at the very end to return the final result at the precision goal, but we have done no
test to insure that the result meets the required precision. Add a test to the end of the
newton function so that, if this condition is not met, an error message is generated
and the current result is output.
3. Some functions tend to cause root-finding methods to converge rather slowly. For
example, the function f x sin x x requires over ten iterations of Newtons
method with an initial guess of x
0
0.1 to get three-place accuracy. Implement the
following acceleration of Newtons method and determine how many iterations of
the function f x sin x x, starting with x
0
0.1, are necessary for six-place
accuracy.
accelNewton x
f x f x
f x
2
f x f x
This accelerated method is particularly useful for functions with multiple roots.
4. Write a functional implementation of the secant method. Your function should
accept as arguments the name of a function and two initial guesses. It should main-
tain the precision of the inputs and it should output the root at the precision of the
8 Numerics 265
initial guess, and the number of iterations required to compute the root. Consider
using the built-in functions FixedPoint or Nest.
5. The norm of a matrix gives some measure of the size of that matrix. The norm of a
matrix A is indicated by A . There are numerous matrix norms, but all share certain
properties. For n n matrices A and B:
(i) A 0
(ii) A 0 if and only if A is the zero matrix
(iii) c A c A for any scalar c
(iv) A B A B
(v) AB A B
One particularly useful norm is the l norm, sometimes referred to as the max norm.
For a vector, this is defined as
x max
1 i n
x
i
The corresponding matrix norm is defined similarly. Hence, for a matrix A a
ij
, we
have
A max
1 i n j 1
n
a
ij
This computes the sum of the absolute values of the elements in each row, and then
takes the maximum of these sums. That is, the l matrix norm is the max of the l
norms of the rows.
Write a function norm[mat,Infinity], which takes a square matrix as an argu-
ment and outputs its norm. Compare your function with the built-in Norm
function.
6. If a matrix A is nonsingular (that is, is invertible), then its condition number c A is
defined as A A
1
. A matrix is called well-conditioned if its condition number is
close to 1 (the condition number of the identity matrix). A matrix is called ill-condi-
tioned if its condition number is significantly larger than 1.
Write a function conditionNumber[mat] that uses the norm you defined in the
previous exercise as an auxiliary function, and outputs the condition number of mat.
Use conditionNumber to compute the condition number of the first ten Hilbert
matrices.
7. An additional technique for solving linear systems of equations is known as scaled
pivoting. Assuming that no column of a matrix mat contains all 0s (in which case there
266 An Introduction to Programming with Mathematica
would be no unique solution), then, for each row, a scale factor is determined by
selecting the element that is the largest in absolute value; that is, in row i, the scale
factor is defined as s
i
max
1 j n
a
ij
. Now a row interchange is determined by
finding the first integer k such that:
a
ki
s
k
max
j 1,2,,n
a
ji
s
j
Once such a k is found, then the ith row and the kth row are interchanged. The
scaling itself is only done for comparison purposes so no additional roundoff error is
introduced by the scaling factor.
Write a function solveSPP that implements scaled partial pivoting using the above
description.
8 Numerics 267
9 Graphics programming
Mathematica contains a rich set of tools for visualizing functions and data. Generally
the built-in graphics functions will provide what you need, but, just like the rest of the
Mathematica programming language, you will periodically find yourself with the need
to create your own plotting and visualization routines. In this chapter we will discuss
how to construct graphical images using Mathematica, and how to write programs that
solve problems that are graphical in nature.
9.1 Structure of graphics
All Mathematica graphics are constructed from objects called graphics primitives. These
primitive elements (Point, Line, Polygon, Circle, etc.) are used by built-in functions
such as Plot to create graphics. Although it is quite straightforward to create images using
Mathematicas built-in functions, you will frequently find yourself having to create a
graphic image for which no Mathematica function exists. This is analogous to the situation
in programming where you often have to write a specialized procedure to solve a particular
problem. We use the basic building blocks and put them together according to the rules
governing the structure of the language and the nature of the problem at hand. In this
section we will look at the building blocks of graphics programming and at how we put
them together to make graphics.
Primitives, directives, and options
Graphics created with functions such as Plot and ListPlot are constructed of lines
connecting points, with options governing the display. We can get some insight into this
process by looking at the internal representation of a plot.
Here is a plot of the sin function.
In[1]:= sinplot Plot Sin x , x, 0, 2
1 2 3 4 5 6
1
0.5
0.5
1
Out[1]= Graphics
Mathematica constructs plots by piecing together various graphics elements. The
InputForm function displays the expression that we could have entered manually to get
the same plot. We use Short to display an abbreviated listing of that expression. (Note:
The formatted output from Short will vary slightly depending upon the width of your
notebook.)
In[2]:= Short InputForm sinplot , 10
Out[2]//Short= Graphics Line 2.617993877991494*^-7,
2.6179938779914644*^-7 , 0.25488992540742256,
0.25213889196341294 , 0.5328694051959508,
0.5080069997492929 , 0.7939393140028285,
0.7131204212611485 , 1.04500937601917, 0.864929243756943 ,
1.1741328775392965, 0.9223551787683757 ,
1.2459531215560486, 0.9477007807106171 ,
1.3122595300248905, 0.9667651045914426 , 2 ,
69 , 5.756221700231303, 0.5029111934098898 ,
6.016521477574974, 0.2635146543573849 ,
6.266821408128109, 0.016363168748098372 ,
6.283185045380199, 2.6179938774695577*^-7 , 25
This graphic consists of a series of coordinates, or points, in the plane connected by
lines of a certain thickness. There are 82 points that are sampled to make this plot the 13
displayed here together with 69 more indicated by the notation <<69>>.
In[3]:= Count InputForm sinplot ,
p_?NumericQ, q_?NumericQ , Infinity
Out[3]= 82
The <<25>> on the bottom indicates options (omitted from this display), such as
PlotRange Automatic. Below we can see that some of these options are immediate
rules and some are delayed.
270 An Introduction to Programming with Mathematica
In[4]:= Count InputForm sinplot , p_Symbol q_, Infinity
Out[4]= 21
In[5]:= Count InputForm sinplot , p_Symbol q_, Infinity
Out[5]= 4
We will examine these graphics elements by constructing a graphic using only primi-
tive elements. In a later section we will look into how the built-in functions such as Plot
construct graphics out of the primitive elements.
In Section 8.1 in the numerics chapter, we displayed a graphic that demonstrated
some of the properties of complex numbers. Let us show how this graphic was created,
using Mathematicas primitive elements.
The following table lists the graphics primitives that we will use in this example
(Point, Line, Circle, and Text) in addition to several other two-dimensional elements
that are available. Note that three-dimensional versions of Point, Line, Polygon, and
Text are also available for constructing three-dimensional graphics.
Graphics elements Usage
Point x, y a point at position x, y
Line x
1
, y
1
, x
2
, y
2
, a line throughthe points x
i
, y
i
Rectangle x
min
, y
min
, x
max
, y
max
a filled rectangle
Polygon x
1
, y
1
, x
2
, y
2
, a filled polygon
Circle x, y , r,
2
,
2
a circular arc of radius r
Disk x, y , r a filled disk of radius r
Raster x
11
, x
12
, , x
21
, x
22
, , a rectangular array of gray levels
Text expr, x, y text centered at x, y
Table 9.1: Graphics primitives
The graphic we will create will contain the following elements:
points in the plane at a complex number a b and its conjugate a b
lines drawn from the origin to each of these points
an arc, indicating the polar angle of the complex number
dashed lines indicating the real and imaginary values
a set of axes in the coordinate plane
9 Graphics programming 271
labels for each of the above elements
First we choose a point in the first quadrant and then construct a line from the origin
to this point.
In[6]:= z 8 3 ;
Line[{{x
1
,y
1
},{x
2
,y
2
},,{x
n
,y
n
}}] is a graphics primitive that creates a line
from the point whose coordinates are x
1
, y
1
to the point x
2
, y
2
, etc..
In[7]:= line1 Line 0, 0 , Re z , Im z ;
Let us also create a point in the plane.
In[8]:= point1 PointSize .02 , Point Re z , Im z ;
We have added the graphics directive PointSize here so that our displayed point
will be reasonably large. A graphics directive works by changing only those objects within its
scope. In this case, that scope is delineated by the curly braces. The form for directives is
{directive, primitive}. Additional primitives can also be placed in the scope of any directive.
dir, prim
1
, prim
2
, , prim
n
The directive dir will affect each of the primitives prim
i
occurring within its scope.
You can place as many primitives as you like within the scope of each directive.
A complete list of the two-dimensional graphics directives, together with usage
statements, is given in Table 9.2.
To display what we have created so far, we first wrap the Graphics function around
the points and lines to turn them into graphics objects. Then we display the list of objects
with the Show function.
In[9]:= Show Graphics line1, point1
Out[9]= Graphics
272 An Introduction to Programming with Mathematica
Admittedly not too exciting, but it is a start. We can add additional graphics ele-
ments indicating the conjugate and a set of axes.
In[10]:= cz Conjugate z ;
line2 Line 0, 0 , Re cz , Im cz ;
point2 PointSize .02 , Point Re cz , Im cz ;
In[13]:= Show Graphics line1, point1, line2, point2
Out[13]= Graphics
Directive Usage
AbsoluteDashing d
1
, d
2
, dashedline segments using absoluteunits
AbsoluteThickness d lines of thickness d measuredin absolute units
CMYKColor c, m, y, b cyan, magenta, yellow, andblack of four
color process
Dashing d
1
, d
2
, dashedline segments of length d
1
, d
2
,
GrayLevel d gray between 0 black and1 white
Hue h, s, b color with hue, saturation, andbrightness
between 0 and1
PointSize r points of radius r given as a fractionof the width
of the entire plot
RGBColor r, g, b color with red, green, andblue components
between 0 and1
Thickness d lines of thickness d given as a fractionof the width
of the entire plot
Table 9.2: Mathematica graphics directives
At this point it would be useful to have axes displayed in our graphic. All of Mathemat-
icas graphics functions have options that allow you to modify some attribute of the entire
graphic. We can get a complete list of those options relevant to Graphics objects by
evaluating the following.
9 Graphics programming 273
In[14]:= Options Graphics
Out[14]= AspectRatio
1
GoldenRatio
, Axes False, AxesLabel None,
AxesOrigin Automatic, AxesStyle Automatic,
Background Automatic, ColorOutput Automatic,
DefaultColor Automatic, DefaultFont $DefaultFont,
DisplayFunction $DisplayFunction, Epilog ,
FormatType $FormatType, Frame False, FrameLabel None,
FrameStyle Automatic, FrameTicks Automatic,
GridLines None, ImageSize Automatic, PlotLabel None,
PlotRange Automatic, PlotRegion Automatic, Prolog ,
RotateLabel True, TextStyle $TextStyle, Ticks Automatic
Notice that each option is specified as a rule with the default value for each option
given on the right-hand side of the rule. In particular, note that Axes is one of the options
for Graphics types and that it is set to False by default.
Options differ from directives in that they affect the entire graphic. Options to func-
tions are placed after any required arguments and are separated by commas. Since Axes is
an option to the Graphics function, it is placed after the graphics elements {line
1
,
point
1
,}. Using the value Automatic for the Axes option is how we ask Mathematica to
figure out the best arrangement for the axes placement and labels, given the elements
present in the graphic.
In[15]:= Show Graphics line1, point1, line2, point2 , Axes Automatic
2 4 6 8
3
2
1
1
2
3
Out[15]= Graphics
Next, let us create dashed lines indicating the real and imaginary components of our
complex number. We use the Dashing directive with Line to get the desired effect.
In[16]:= hline
Dashing 0.04, 0.04 , Line 0, Im z , Re z , Im z ;
vline Dashing 0.04, 0.04 ,
Line Re z , 0 , Re z , Im z ;
Since we were using this graphic to display an arbitrary complex number, we are not
interested in the units on the axes, so we suppress the default value and add our own with
274 An Introduction to Programming with Mathematica
the Ticks option. Ticks {{{Re[z],"a"}},{{Im[z],"b"}}} places tick marks at
Re[z] on the horizontal axis and at Im[z] on the vertical axis and labels them a and b,
respectively. In addition, let us add labels on the axes.
In[18]:= Show Graphics line1, point1, line2, point2, hline, vline ,
Axes Automatic, AxesLabel Re, Im ,
Ticks Re z , "a" , Im z , "b" ;
a
Re
b
Im
Mathematica tries to fit the plot into a region that is similar in shape to your com-
puter screen and uses a ratio of height to width that is known to be pleasing to the eye.
This height to width ratio is known as the AspectRatio and has a default value of
1
,
where is the golden ratio. By setting AspectRatio to Automatic, we will force
Mathematica to use a ratio that is determined from the actual coordinates in the plot.
In[19]:= Show Graphics line1, line2, point2, hline, vline ,
Axes Automatic, AxesLabel Re, Im , Ticks
Re z , "a" , Im z , "b" , AspectRatio Automatic ;
a
Re
b
Im
We now wish to put labels at the two complex numbers and along the line represent-
ing the length Abs[z]. We will use another graphics primitive, Text, to place text where
we need it.
9 Graphics programming 275
Text[expr,{x,y}] will create a text object of the expression expr and center it at (x,
y). So, to create z a b as a piece of text centered at a point a little bit above and to
the left of z, we use:
Text["z = a + b i", {Re[z]-0.75, Im[z]+0.35}]
We are going to add one further element to this graphic object. We would like this
text to use a different font and a different size than the default of Courier, 10 points. Using
StyleForm we can specify any available font and size. In this example we use the Times
font family and set the font size at 9 points. (Names of fonts will vary on different comput-
ers. Users should check their Mathematica documentation for font-naming conventions.)
Text[StyleForm["z = a + b i", FontFamily "Times", FontSize 9]],
{Re[z] - 0.75, Im[z] + 0.35}]
Here then are the labels for the complex number and the length given by the abso-
lute value of the complex number.
In[20]:= text1 Text StyleForm "z a b ",
FontFamily "Times", FontSize 9 ,
Re z .75, Im z .35 ;
In[21]:= text2 Text StyleForm "Abs z ",
FontFamily "Times", FontSize 9 ,
4.2, 2 ;
In[22]:= Show Graphics line1, line2, point1, point2, hline, vline,
text1, text2 , Axes Automatic, AxesLabel Re, Im ,
Ticks Re z , "a" , Im z , "b" ,
AspectRatio Automatic ;
a
Re
b
Im
z a b
Abs z
276 An Introduction to Programming with Mathematica
Lastly, we need to add the arc representing the polar angle and label it. The arc can
be generated with another graphic primitive. Circle[{x,y},r,{a,b}] will draw an arc
of a circle centered at (x, y), of radius r, counterclockwise from an angle of a radians to an
angle of b radians. The arc that we are interested in will have a radius smaller than Abs[z]
and will be drawn from the real (horizontal) axis to the line connecting the origin and z.
Here is the code for the arc and its label, as well as the graphic containing all of the above
elements (we also add the text to label the conjugate).
In[23]:= arc Circle 0, 0 ,
Abs z
3
, 0, Arg z ;
In[24]:= text3 Text StyleForm "Conjugate z a b ",
FontFamily "Times", FontSize 9 ,
Re cz 1.4, Im cz .35 ;
In[25]:= text4 Text StyleForm "Arg z ",
FontFamily "Times", FontSize 9 , 3.5, .5 ;
In[26]:= Show Graphics line1, line2, point1, point2,
hline, vline, text1, text2, text3, text4, arc ,
Axes True, AxesLabel Re, Im , Ticks
Re z , "a" , Im z , "b" , AspectRatio Automatic ;
a
Re
b
Im
z a b
Abs z
Conjugate z a b
Arg z
We have made assignments to many different symbols in this section. Before going
on, it would be a good idea to clear the values associated with all of these symbols. In
Chapter 12 we will talk about contexts in detail, but, for now, you can clear the values
associated with all symbols in the Global` context by evaluating the following.
In[27]:= ClearAll "Global` "
9 Graphics programming 277
Exercises
1. Create a primitive color wheel by coloring successive sectors of a Disk according to
the Hue directive.
2. Create a graphic that contains one each of a circle, a triangle, and a rectangle. Your
graphic should include an identifying label for each object.
3. Create a three-dimensional graphic containing six Cuboid graphics primitives,
randomly placed in the unit cube.
4. Create a graphic that consists of 500 points placed randomly in the unit square. The
points should be of random radii between .01 and 0.1 units, and colored randomly
according to a Hue function.
5. Create a graphic that represents the solution to the following algebraic problem that
appeared in Porta, Davis and Uhl, 1994. Find the positive numbers r such that the
following system has exactly one solution in x and y.
x 1
2
y 1
2
2
x 3
2
y 4
2
r
2
Once you have found the right number r, then plot the resulting circles in true scale
on the same axes, plotting the first circle with solid lines and the two solutions with
dashed lines together in one graphic.
6. Load the package Graphics`Polyhedra` and then display each of the solids
defined in the package, including Tetrahedron, Octahedron, Icosahedron,
Cuboid, and the Dodecahedron.
7. Create a graphic of the sin function over the interval (0, 2 ) that displays vertical
lines at each point calculated by the Plot function to produce its plot.
278 An Introduction to Programming with Mathematica
9.2 Graphics programming
Up until this point, we have looked at the tools that are available to construct relatively
simple graphics in Mathematica. This has allowed us to create images by using the graphics
building blocks primitives, directives, and options. In this section we consider problems
that are more involved or whose solution requires geometric insight as we construct our
programs. We will begin with two examples that create specialized plotting functions, the
first for plotting roots on a given interval and the second for plotting data. The second of
these will give a good introduction to incorporating error messages and options into your
functions. The last two examples are more mathematical in nature. The first is a purely
geometric problem on simple closed paths. The last example shows how to construct
graphics from programming work we did in Chapter 7, the display of binary trees.
Root plotting
In this section we will use our knowledge of built-in graphics functions together with
various programming techniques from previous chapters to write a program that plots a
function together with all of its roots in a given interval. The basic idea, using Cases to
extract the points in a plot and Split to identify sign changes, is due to Paul Abbott from
his article in The Mathematica Journal (Abbott 1998).
In Exercise 7 of Section 9.1, we used Cases to extract coordinate pairs from the data
in sinplot. In this section, we will use a function with a few more roots in the specified
interval to work through the details of the problem.
In[1]:= sinplot Plot Sin 2 x , x, 1, 7 ;
2 4 6
1
0.5
0.5
1
This finds all those Line expressions from sinplot and extracts only their argu-
ments, the point coordinates. Note the need for as a third argument to Cases so that
the pattern matching goes down to the deepest nested expression in sinplot.
In[2]:= pts Cases sinplot, Line x__ x, ;
9 Graphics programming 279
In[3]:= Shallow pts
Out[3]//Shallow= 1., 0.909298 , 0.960965, 0.938983 ,
0.924601, 0.961495 , 0.884699, 0.980344 ,
0.862349, 0.98818 , 0.842015, 0.993596 ,
0.831653, 0.995724 , 0.820668, 0.997513 ,
0.81129, 0.99866 , 0.800961, 0.999516 , 147
From the above list of points, we select each pair that exhibits a sign change in the
y-coordinate (Last[]).
In[4]:= Select Split pts, Sign Last #2 Sign Last #1 & ,
Length #1 2 &
Out[4]= 0.00256681, 0.0051336 , 0.17505, 0.342993 ,
1.32279, 0.475915 , 1.65979, 0.177051 ,
3.0015, 0.276542 , 3.34379, 0.39347 ,
4.65091, 0.12264 , 4.82866, 0.230455 ,
6.01028, 0.519114 , 6.34258, 0.118509
A sign change occurs between each of the first and second points, the third and
fourth, the fifth and sixth. FindRoot will use the bisection method if we pass it two initial
values, so using the first two x-coordinates in each of these three pairs should give us the
roots we are after.
In[5]:= Map First, %, 2
Out[5]= 0.00256681, 0.17505 , 1.32279, 1.65979 ,
3.0015, 3.34379 , 4.65091, 4.82866 , 6.01028, 6.34258
In[6]:= Map FindRoot Sin 2 x 0, x, # 1 , # 2 &, %
Out[6]= x 4.12702 10
19
, x 1.5708 ,
x 3.14159 , x 4.71239 , x 6.28319
Now we can turn these roots into graphics objects and combine them with the
original plot.
In[7]:= roots x . %
Out[7]= 4.12702 10
19
, 1.5708, 3.14159, 4.71239, 6.28319
In[8]:= pts Map Point #, 0 &, roots
Out[8]= Point 4.12702 10
19
, 0 , Point 1.5708, 0 ,
Point 3.14159, 0 , Point 4.71239, 0 , Point 6.28319, 0
280 An Introduction to Programming with Mathematica
In[9]:= Show sinplot,
Epilog RGBColor 0, 0, 1 , PointSize .02 , pts ;
2 4 6
1
0.5
0.5
1
Here then is a function that combines all of these steps.
In[10]:= RootPlot fun_, x_, xmin_, xmax_ :
Module z, fplot, pts, spts, roots,
points, f Function x, Evaluate fun ,
fplot Plot f x , x, xmin, xmax ,
DisplayFunction Identity ;
pts Cases fplot, Line z__ z, ;
spts Map First,
Select Split pts, Sign Last #2 Sign Last #1 & ,
Length #1 2 & , 2 ;
roots Map FindRoot f x 0, x, # 1 , # 2 &, spts ;
points Map Point #, 0 &, x . roots ;
Show fplot, DisplayFunction $DisplayFunction,
Epilog RGBColor 0, 0, 1 , PointSize .02 , points ;
roots
In[11]:= RootPlot Sin z 2 Sin z , z, , 3
2 2 4 6 8
1
0.5
0.5
1
Out[11]= z 1.75004 , z 4.21345 10
24
, z 1.75004 ,
z 3.14159 , z 4.53315 , z 6.28319 , z 8.03322
9 Graphics programming 281
In[12]:= Chop %
Out[12]= z 1.75004 , z 0 , z 1.75004 , z 3.14159 ,
z 4.53315 , z 6.28319 , z 8.03322
The exercises at the end of this section contain suggestions for passing options from
RootPlot to the auxiliary fplot by means of Utilities`FilterOptions`.
Plotting data
In this section we will create a function from graphics primitives that overcomes a minor
inconvenience of ListPlot. ListPlot normally plots a vector or matrix of data, display-
ing each piece of data as a Point object. When the option PlotJoined is set to True,
the data points are connected by Line primitives, but the original Point primitives are
not displayed. For example, here are ten points in the plane.
In[13]:= data2D 0.043, 0.575 ,
0.151, 0.120 , 0.234, 0.001 , 0.283, 0.930 ,
0.343, 0.569 , 0.416, 0.768 , 0.465, 0.675 ,
0.539, 0.528 , 0.786, 0.856 , 0.914, 0.794 ;
Here is a plot of the points in the plane. We make the points a little larger with the
PlotStyle option.
In[14]:= ListPlot data2D, PlotStyle PointSize .02 ;
0.4 0.6 0.8
0.2
0.4
0.6
0.8
When given the PlotJoined option, ListPlot simply connects the points with
lines, but the points themselves are omitted.
In[15]:= ListPlot data2D, PlotJoined True ;
0.4 0.6 0.8
0.2
0.4
0.6
0.8
282 An Introduction to Programming with Mathematica
A simplistic first approach would be to make a function that grabs the data and then
stuffs them into Point primitives. Note the use of conditional definitions so that Show
Points handles both one- and two-dimensional data sets. In the case of a one-dimen-
sional data vector, each point is indexed by its position in the vector. In the case of
two-dimensional data input, we assume that each data point maps to its coordinates in the
plane.
In[16]:= ShowPoints data_, s_: 0.02 : PointSize s ,
MapIndexed Point #2 1 , #1 &, data ; VectorQ data ;
In[17]:= ShowPoints data_, s_: 0.02 :
PointSize s , Map Point, data ; Dimensions data 2 2;
Here, Epilog is used to add the points after the data have been plotted.
In[18]:= ListPlot data2D, PlotJoined True,
Epilog ShowPoints data2D ;
0.4 0.6 0.8
0.2
0.4
0.6
0.8
Here is a one-dimensional example of the use of ShowPoints.
In[19]:= data1D Table Random Integer, 1, 10 , 8
Out[19]= 8, 7, 5, 9, 1, 8, 9, 2
In[20]:= ListPlot data1D, PlotJoined True,
Epilog ShowPoints data1D ;
2 3 4 5 6 7 8
4
6
8
There are several disadvantages to this approach. Epilog is not a commonly used
method of modifying graphics and so users might not expect that this would be the way to
display the points. Secondly, it is difficult to modify the style of the Point objects with
9 Graphics programming 283
this approach you would have to make PlotStyle PointSize[] an available
option to ShowPoints and that makes things too complicated for the user who would
have to think about options to options.
A better approach, more consistent with established Mathematica programming style,
would be to create a function that plots the data much like ListPlot, has a Plot
Joined option, but does not omit the Point graphics objects from the plot. Although a
function already exists that does much of this (see the Standard Add-ons package
Graphics`MultipleListPlot`), it is instructive to create such a function from
scratch in order to demonstrate how to use graphics primitives, options, and error-check-
ing in writing functions.
First, let us deal with the shape of the data. If the data are given as a two-dimensional
list we will assume that each data point, consisting of a pair of numbers, gives the horizon-
tal and vertical coordinates directly. In this case, the data can be passed directly to the
graphics primitives.
If the data are given as a one-dimensional list, we will put them into a two-dimen-
sional form by indexing each data point.
In[21]:= data1D Table Random Integer, 1, 10 , 8
Out[21]= 3, 5, 7, 7, 10, 6, 10, 10
In[22]:= pts MapIndexed #2 1 , #1 &, data1D
Out[22]= 1, 3 , 2, 5 , 3, 7 , 4, 7 , 5, 10 , 6, 6 , 7, 10 , 8, 10
The plot will be constructed of graphics primitives directly. For example to simply
plot the points, we could do the following.
In[23]:= Show Graphics PointSize .02 , Map Point, pts ,
Axes Automatic
2 3 4 5 6 7 8
4
5
6
7
8
9
10
Out[23]= Graphics
284 An Introduction to Programming with Mathematica
Our function will be named DataPlot. We start by giving it the same options as
those of ListPlot.
In[24]:= Options DataPlot Options ListPlot
Out[24]= AspectRatio
1
GoldenRatio
, Axes Automatic, AxesLabel None,
AxesOrigin Automatic, AxesStyle Automatic,
Background Automatic, ColorOutput Automatic,
DefaultColor Automatic, DefaultFont $DefaultFont,
DisplayFunction $DisplayFunction, Epilog ,
FormatType $FormatType, Frame False, FrameLabel None,
FrameStyle Automatic, FrameTicks Automatic,
GridLines None, ImageSize Automatic, PlotJoined False,
PlotLabel None, PlotRange Automatic, PlotRegion Automatic,
PlotStyle Automatic, Prolog , RotateLabel True,
TextStyle $TextStyle, Ticks Automatic
Next, we need a way of passing the option PlotJoined to the DataPlot. This is
accomplished by the following construction.
pjQ = PlotJoined /. Flatten[{opts, Options[DataPlot]}]
Read from right to left, first the options that are passed to DataPlot are combined
in a list with the options defined for DataPlot above. Then that list is flattened to
remove any nested lists of options. Then the value for the rule PlotJoined val is
extracted and assigned to the symbol pjQ. So, for example, if the user evaluates Data
Plot[data,PlotJoined True], then inside the body of DataPlot, pjQ will be
assigned the value True.
Finally, here are all the pieces put together in our first construction of DataPlot.
Note the use of the package Utilities`FilterOptions`. This allows us to pass the
options for Graphics directly into our function DataPlot inside Show. FilterOp
tions will insure that only valid Graphics options are passed.
In[25]:= Needs "Utilities`FilterOptions`"
In[26]:= Options DataPlot Options ListPlot ;
9 Graphics programming 285
In[27]:= DataPlot data_, opts___ : Module pjQ, pts ,
pjQ PlotJoined . Flatten opts, Options DataPlot ;
pts Which
VectorQ data , MapIndexed #2 1 , #1 &, data ,
Dimensions data 2 2, data ;
If pjQ,
Show Graphics PointSize .02 , Point pts, Line pts ,
FilterOptions Graphics, opts , Axes Automatic ,
Show Graphics PointSize .02 , Point pts ,
FilterOptions Graphics, opts , Axes Automatic
In[28]:= data2D 0.043, 0.575 , 0.151, 0.120 , 0.234, 0.001 ,
0.283, 0.930 , 0.343, 0.569 , 0.416, 0.768 ,
0.465, 0.675 , 0.539, 0.528 , 0.786, 0.856 ,
0.914, 0.794 ;
In[29]:= DataPlot data2D, PlotJoined True ;
0.2 0.4 0.6 0.8
0.2
0.4
0.6
0.8
In[30]:= data1D Table Random Integer, 1, 10 , 8
Out[30]= 2, 7, 7, 7, 3, 3, 6, 1
In[31]:= DataPlot data1D ;
2 3 4 5 6 7 8
2
3
4
5
6
7
The exercises contain several examples of modifications and improvements to
DataPlot.
286 An Introduction to Programming with Mathematica
Simple closed paths
Our next example of a programming problem that involves the use of graphics solves a
very simplified variation of what are known as traveling salesman problems. A closed path is
one that travels to every point and returns to the original point. The traveling salesman
problem asks for the shortest closed path that connects an arbitrary set of points.
The traveling salesman problem is one of great theoretical, as well as practical,
importance. Airline routing and telephone cable wiring over large regions are examples of
problems that could benefit from a solution to the traveling salesman problem.
From a theoretical point of view, the traveling salesman problem is part of a large
class of problems that are known as NP-complete problems. These are problems that can be
solved in polynomial time using nondeterministic algorithms. A nondeterministic algorithm
has the ability to choose among many options when faced with numerous choices, and
then to verify that the solution is correct. The outstanding problem in computer science at
present is known as the problem. This equation says that any problem that can be
solved by a nondeterministic algorithm in polynomial time ( ) can be solved by a
deterministic algorithm in polynomial time ( ). It is widely believed that and
considerable effort has gone into solving this problem. (The interested reader should
consult Lawler et al 1985 or Pemmaraju and Skiena 2003.)
Our focus will be on a solvable problem that is a substantial simplification of the
traveling salesman problem. We will find a simple closed path a closed path that does not
intersect itself through a set of n points.
We will demonstrate a graphical solution to the problem by working with a small
value of n and then generalizing to arbitrary values of n. Let us first create a set of ten pairs
of points (n 10) in the unit square.
In[32]:= coords Table Random , 10 , 2
Out[32]= 0.429717, 0.94548 , 0.154498, 0.333952 ,
0.829465, 0.187126 , 0.185409, 0.208253 ,
0.253829, 0.432073 , 0.36397, 0.603652 ,
0.0593643, 0.774766 , 0.804412, 0.766921 ,
0.898828, 0.920331 , 0.858976, 0.829739
9 Graphics programming 287
Here we have created a table of ten pairs of numbers (the coordinates of our points in
the plane), and then created graphics primitives by mapping Point over each pair.
In[33]:= points Map Point, coords
Out[33]= Point 0.429717, 0.94548 , Point 0.154498, 0.333952 ,
Point 0.829465, 0.187126 , Point 0.185409, 0.208253 ,
Point 0.253829, 0.432073 , Point 0.36397, 0.603652 ,
Point 0.0593643, 0.774766 , Point 0.804412, 0.766921 ,
Point 0.898828, 0.920331 , Point 0.858976, 0.829739
We can show the points alone.
In[34]:= Show Graphics PointSize .02 , points
Out[34]= Graphics
Or we can show the points connected by lines.
In[35]:= lines Line coords
Out[35]= Line 0.429717, 0.94548 , 0.154498, 0.333952 ,
0.829465, 0.187126 , 0.185409, 0.208253 ,
0.253829, 0.432073 , 0.36397, 0.603652 ,
0.0593643, 0.774766 , 0.804412, 0.766921 ,
0.898828, 0.920331 , 0.858976, 0.829739
In[36]:= Show Graphics PointSize .02 , points, lines
Out[36]= Graphics
288 An Introduction to Programming with Mathematica
Let us create a utility function for plotting a set of points in the plane together with
lines connecting them in order.
In[37]:= PointPlot coords_List :
Show Graphics
Line coords ,
PointSize .02 , RGBColor 1, 0, 0 , Map Point, coords
In[38]:= PointPlot coords ;
At this stage, it is apparent that there are two problems. First, the path is not closed;
that is, the last point visited is not the point we started from. The Line primitive connects
the first point to the second, the second to the third, etc., in the sequence that the points
are presented to it. So we need to connect the last point to the first point to close the path.
This can be accomplished by appending the first point to the end of the list of coordinates.
In[39]:= path coords . a_, b__ a, b, a ;
In[40]:= PointPlot path ;
The second problem the fact that our path is not simple is geometric in nature.
To find an algorithm that will insure that our path does not cross itself for any set of points
in the plane, we will first pick a point from our set at random and call this the base point.
In[41]:= base coords Random Integer, 1, Length coords
Out[41]= 0.253829, 0.432073
9 Graphics programming 289
The path problem can be solved by first computing the counterclockwise (polar)
angle between a horizontal line and each of the remaining points, using the base point as
the vertex of the angle. Then, sorting the points according to this angle and connecting the
points in this order will produce the desired result.
Base
p1
p2
p3
First we compute the angle between two points a and b. (You should verify the
trigonometric analysis necessary to find this angle in the various cases. Note that we are
computing the polar angle between two points and, hence, we need the ArcTan function.)
In[42]:= angle a_List, b_List : Apply ArcTan, b a
We can use this function to compute the angle between our base point and each of
the points in the list coords. We need to make sure that we do not try to compute the
angle between the base point and itself as this will evaluate to ArcTan[0,0], which is
undefined. This situation can be avoided by removing the base point from our list of
coordinates when computing the angles.
In[43]:= remain Complement coords, base ;
In[44]:= Map angle base, # &, remain
Out[44]= 2.08695, 2.36232, 1.86747, 1.00012,
1.24074, 0.546405, 0.402315, 0.581378, 0.64796
Instead of computing the angles explicitly, we will just use the angle function as an
ordering function on our list of coordinates. Sort[list,rule] will sort list according to
rule, which is a two-argument predicate. We wish to sort coords according to our order-
ing function on the angles between each point and the base point. The following code
accomplishes this.
In[45]:= s Sort remain, angle base, #1 angle base, #2 &
Out[45]= 0.154498, 0.333952 ,
0.185409, 0.208253 , 0.829465, 0.187126 ,
0.804412, 0.766921 , 0.858976, 0.829739 ,
0.898828, 0.920331 , 0.36397, 0.603652 ,
0.429717, 0.94548 , 0.0593643, 0.774766
290 An Introduction to Programming with Mathematica
This is our list of coordinates sorted according to the polar angle between each point
and the base point. In order to start and end with the base point, we Join three separate
lists and then display the graphic.
In[46]:= path Join base , s, base
Out[46]= 0.253829, 0.432073 , 0.154498, 0.333952 ,
0.185409, 0.208253 , 0.829465, 0.187126 ,
0.804412, 0.766921 , 0.858976, 0.829739 ,
0.898828, 0.920331 , 0.36397, 0.603652 , 0.429717, 0.94548 ,
0.0593643, 0.774766 , 0.253829, 0.432073
In[47]:= PointPlot path ;
If we collect the above commands into a program simpleClosedPath, then we
can find such paths for arbitrary sets of coordinates.
In[48]:= simpleClosedPath lis_ : Module base, angle, sorted ,
base lis Random Integer, 1, Length lis ;
angle a_, b_ : Apply ArcTan, b a ;
sorted Sort Complement lis, base ,
angle base, #1 angle base, #2 & ;
Join base , sorted, base
Now we can create large sets of points and find the corresponding simple closed path
readily.
In[49]:= data Table Random , 25 , 2 ;
In[50]:= PointPlot simpleClosedPath data ;
9 Graphics programming 291
In[51]:= data Table Random , 100 , 2 ;
In[52]:= PointPlot simpleClosedPath data ;
Although the algorithm we have developed in this section for computing simple
closed paths seems to work fairly well, there are certain conditions under which it will still
fail. The exercises at the end of this section investigate some of those conditions and walk
you through how best to work around them.
Drawing trees
The trees drawn in Chapter 7 were drawn using a Mathematica program. We will develop a
simpler version of the program here; the full version is developed in the exercises. Here,
trees are drawn without their labels with just a disk at each node and, more impor-
tantly, the placement of nodes is not as good (aesthetically speaking). Still, it is a good
example of using recursion to create a line drawing.
When drawing trees, the central question is: How far should the children of a given
node be separated? For example, in Figure 9.1, the separation of the children of node 2 is
much greater than that of the children of node 1. That is because the total width of the
trees below node 2 is so great that they require such a separation; or, rather, the total width
of the right side of the left subtree and the left side of the right subtree requires that
separation.
1 2
Figure 9.1: A tree with different separations
292 An Introduction to Programming with Mathematica
To illustrate this point, consider the trees in Figures 9.2(a) and 9.2(b). The subtrees
of the root are the same, but in a different order; the result is that in Figure 9.2(a), the
children of the root must be separated much more.
a b
Figure 9.2: Trees whose children have different separations
Thus, to properly place subtrees, we need to know, for each one, its total width to
the left and to the right of its root. Then, the two trees will be separated by an amount
equal to the right width of the left subtree plus the left width of the right subtree, plus
some arbitrary additional separation. This is illustrated in Figure 9.3. lw1 represents the
left width of the left subtree, rw1 the right width of the left subtree, and lw2 and rw2
represent the corresponding widths for the right subtree. minsep is the additional separa-
tion always added between subtrees, and sep is the separation eventually computed for
these two subtrees.
sep
lw1 rw1 minsep lw2 rw2
Figure 9.3: Calculation of the separation between children
The function placeTree is given a binary tree (represented as in Section 7.5) and
returns a list of three things:
1. A separation tree: a tree having the same shape as the argument, labelled at each
interior node with a number, the separation of that nodes children.
2. The left width of the tree: the distance it extends to the left from its root.
3. The right width of the tree.
9 Graphics programming 293
Now, computing placeTree[{lab},lc,rc}] is accomplished in these steps:
Recursively compute placeTree[lc] and placeTree[rc]; suppose the
results are {st1,lw1,rw1} and {st2,lw2,rw2}, respectively.
The separation of lc and rc is equal to the right width of lc (rw1) plus the left
width of rc (lw2), plus the additional separation. Call the total separation thus
computed sep.
The left width of the total tree is sep/2 +lw1, and its right width is sep/2
+ rw2.
This leads to the following code.
In[53]:= placeTree _ : , 0, 0
placeTree _, lc_, rc_ : Module left placeTree lc ,
right placeTree rc , minsep 1.0, sep ,
sep left 3 right 2 minsep;
sep, left 1 , right 1 , left 2
sep
2
, right 3
sep
2
Given a list {st,lw,rw} produced by placeTree, we no longer need lw or rw to
draw the tree: the separation tree st suffices. Transforming st into a drawing is straightfor-
ward (the Disk primitive draws a filled circle with given center and radius).
In[55]:= drawSepTree , lev_, xaxis_ :
Disk xaxis, lev , 0.1
drawSepTree sep_, lc_, rc_ , lev_, xaxis_ :
Join Disk xaxis, lev , 0.1 ,
Line xaxis, lev , xaxis sep, lev 1 ,
Line xaxis, lev , xaxis sep, lev 1 ,
drawSepTree lc, lev 1, xaxis sep ,
drawSepTree rc, lev 1, xaxis sep
Thus, to draw a tree tree, enter:
placeTree[tree];
drawSepTree[%[[1]], 0, 0];
Show[Graphics[%]]
Alternately, we can create a function to automate this process.
In[57]:= showTree tree_, opts___ :
Show Graphics drawSepTree placeTree tree 1 , 0, 0 , opts
294 An Introduction to Programming with Mathematica
Here is a simple example.
In[58]:= tree1 a, b , a, c, e, g , f , d , b ;
In[59]:= showTree tree1, AspectRatio Automatic ;
Exercises
1. Create a function ComplexListPlot that plots a list of complex numbers using
the ListPlot function. Set initial options so that the PlotStyle is red, the
PointSize is a little larger than the default, and the horizontal and vertical axes are
labeled Re and Im, respectively. Set it up so that options to ComplexListPlot
are passed to ListPlot.
2. Create a function RootPlot that plots the complex solutions to a polynomial in the
plane. Use your implementation of ComplexListPlot that you developed in the
previous exercise.
3. Modify the function RootPlot so that you can pass the options fromPlot to the
auxiliary function fplot. You will need to use the Utilities`FilterOp
tions` package to pass these options.
4. Add some error checking to DataPlot so that a message is returned if the data that
are passed are not a one- or two-dimensional list. Your message should be of the
following form:
In[1]:= DataPlot::baddim "The data used by DataPlot must be
in the form of a one or two dimensional list.";
Then modify the Which statement inside DataPlot so that it continues to do the
right thing if the data that are passed are a one- or two-dimensional list, but, if not,
the baddim message above is returned. For example, something like the following
will work.
9 Graphics programming 295
In[2]:= pts Which
VectorQ data , MapIndexed #2 1 , #1 &, data ,
Dimensions data 2 2, data,
True, Message DataPlot::baddim ; $Failed ;
5. Although the program simpleClosedPath works well, there are conditions under
which it will occasionally fail. Experiment by repeatedly computing simpleClosed
Path for a set of ten points until you see the failure. Determine the conditions that
must be imposed on the selection of the base point for the program to work
consistently.
6. Modify simpleClosedPath so that the point with the smallest x-coordinate of the
list of data is chosen as the base point.
7. Modify simpleClosedPath so that the point that has the largest y-coordinate is
chosen as the base point.
8. Write a function triangleArea that computes the area of any triangle in the plane
given a list of the three coordinate points that describe that triangle.
9. Write a function pointInPolygonQ that tests whether a given point is inside a
specified polygon. For example, the origin is inside the polygon formed by joining
the four unit vectors:
pointInPolygonQ 1, 0 , 0, 1 , 1, 0 , 0, 1 , 0, 0
True
10. A polygon is called convex if a line connecting any two points inside the polygon lies
completely inside the polygon. Most of the simple closed polygons we computed in
this section are nonconvex. For a given set of n points, find those points which form a
convex polygon that is a boundary for the entire point set. (The smallest such bound-
ary is called the convex hull of the set of points.) That is, given a set of points in the
plane
296 An Introduction to Programming with Mathematica
write a function convex that outputs a graph such as the following.
11. Another way of finding a simple closed path is to start with any closed path and
progressively make it simpler by finding intersections and changing the path to avoid
them. Prove that this process ends, and that it ends with a closed path. Write a
program to implement this procedure and then compare the paths given by your
function with those of simpleClosedPath given in the text.
12. The tree-drawing code we have presented is not the same code we used in drawing
the trees in Chapter 7. The two trees drawn in Figure 9.4 show the difference:
drawing (a) is the one produced by placeTree, and (b) is the one produced by the
algorithm used in Chapter 7. That algorithm is due to Reingold and Tilford (1981),
and basically what it does is just this: instead of basing the separation of subtrees on
their total width, it does a level-by-level comparison, and separates them only as far
as needed at any particular level.
a b
Figure 9.4: Results from different tree-drawing algorithms
Program this tree-drawing algorithm. There is one tricky part to it, which we will
leave you to discover for yourself, except to say this: your program should draw the
tree shown in Figure 9.5 roughly as you see it here.
9 Graphics programming 297
Figure 9.5: A tricky tree to draw
13. Another difference between the tree-drawing code we have shown here and the code
that was used in Chapter 7 is that the algorithm there was able to draw trees with
labels at the nodes. Extend your algorithm from Exercise 13 to add labels; your trees
should have strings as their labels. You need to take the width of the labels into
account when computing the separation tree (this is a change to placeTree), and
make sure the lines do not intersect the labels (this is a change to drawSepTree).
Unfortunately, there is no way to compute the exact width of a text string as it will
appear in a Mathematica graphic; just approximate using the number of characters in
the label.
9.3 Sound
The sound of mathematics
We hear sound when the air around our ears compresses and expands the air near the
eardrum. Depending upon how the eardrum vibrates, different signals are sent to the brain
via the auditory nerves in the inner ear. These signals are then interpreted in the brain as
various sounds. Musical tones compress and expand the air periodically according to sine
waves. The human ear is able to hear these waves when the frequency is between 20 and
20,000 oscillations per second, or hertz.
298 An Introduction to Programming with Mathematica
Recall that one oscillation of sin x occurs between 0 and 2 .
In[1]:= Plot Sin x , x, 0, 2 ;
1 2 3 4 5 6
1
0.5
0.5
1
The function sin 4 x oscillates four times in the same interval.
In[2]:= Plot Sin 4 x , x, 0, 2 ;
1 2 3 4 5 6
1
0.5
0.5
1
Mathematica is able to take a function such as sin and sample its amplitudes roughly
8000 times per second, and then send corresponding voltages to the speaker on your
computer, if it has one, to produce the sound of the sine wave. The function that accom-
plishes this is Play, which has the same syntax as the Plot command.
In[3]:= ?Play
Play f, t, tmin, tmax plays a sound
whose amplitude is given by f as a function of
time t in seconds between tmin and tmax. More
The function Sin[256t] oscillates 256 times each 2 units, so, if we want to play
a function that oscillates 256 times per second, we want Sin[256 t (2 )]. This plays
the function for two seconds.
9 Graphics programming 299
In[4]:= Play Sin 256 t 2 , t, 0, 2
Out[4]= Sound
If your computer has sound capabilities, you should hear a C, one octave below
middle C, played for two seconds. The graphic that Mathematica outputs with the Sound
object is a somewhat primitive attempt to display the waveform. Since it does not contain
very useful information, we will occasionally omit it from the display.
The Play function samples functions at a rate of about 8,000 times per second, or
hertz. This is good to keep in mind as anomalies can occur when playing a function whose
periodicity is very close to the sample rate. Listen to the quite surprising result that follows
(users will have to check the SampleRate on their computers and adjust the following
code accordingly).
In[5]:= Options Play, SampleRate
Out[5]= SampleRate 8000
In[6]:= Play Sin 8000 2 t , t, 0, 1
Out[6]= Sound
Although we would expect a tone at 8,000 hertz, we get something quite different.
You are encouraged to try other frequencies that are close to the sample rate on your
computer. Play is sampling the function sin 8000 2 t 8,000 times. Since the function
oscillates 8,000 times on this interval, the samples appear to be about the same and so
300 An Introduction to Programming with Mathematica
Play misses the periodic nature of this function. If Play did adaptive sampling, much like
Plot does, then it would avoid this particular problem.
Sounds that are generally thought to be pleasant to the human ear are modeled by
periodic functions. Noise consists of random amplitudes. We can use these notions to find
periodicity in sequences of numbers.
For example, recall that a rational number can be expressed as a finite or repeating
decimal, whereas an irrational number cannot be so represented. If we were to play the
digits of a rational number, its periodic nature should be apparent as a discernible tone.
Playing the digits of an irrational number should result in noise.
The following displays the first 20 digits of the decimal expansion of
1
19
.
In[7]:= RealDigits N
1
19
, 20
Out[7]= 5, 2, 6, 3, 1, 5, 7, 8, 9, 4, 7, 3, 6, 8, 4, 2, 1, 0, 5, 3 , 1
The 1 at the end of the above list indicates the number of places to the left of the
decimal point where the first non-0 digit occurs. Since the first digit of this real number is
one place to the right of the decimal point, this is indicated with a negative number.
The periodic nature of this number is not apparent from such a short list. We can
lengthen the list and pull off only the decimal digits as follows. We suppress the display of
the output using the semicolon.
In[8]:= digits = First[RealDigits[N[1/19, 1000]]];
Now we can play this list of digits. ListPlay will play a sound where the ampli-
tudes are given by the numbers in our list. (Mathematica scales the amplitudes to fit in a
range that ListPlay can work with, and that is audible.)
In[9]:= ListPlay digits
Out[9]= Sound
Clearly (from listening to the resulting tone), this sequence is periodic, whereas the
following sequence of digits is not.
In[10]:= irratdigits First RealDigits N , 1000 ;
In[11]:= ListPlay irratdigits
Out[11]= Sound
As the reader is probably well aware at this point, Play and ListPlay are audio
analogues of Plot and ListPlot. This analogy will allow us to do audio programming
in much the same way as we approached graphics programming earlier in this chapter. The
next section contains a discussion of some ideas in sound synthesis.
9 Graphics programming 301
White noise, white music
Imagine playing a recording of a certain sound at different speeds. Normally you would
expect the character of the resulting sound to be quite different than the original. Speeding
up a recording of your voice makes it sound cartoon-like, and if sped up fast enough,
unintelligible. Slowing down a recording of the first few bars of Gershwins Rhapsody in
Blue would make the clarinet solo sound like a rumble.
There are some sounds though that sound roughly the same when played at different
speeds. Benit Mandelbrot of the IBM Thomas J. Watson Research Center described
these sounds as scaling noises. White noise is probably the most common example of a
scaling noise. If you tuned your radio in between stations, recorded the noise, and then
played the recording at different speeds, you would hear roughly the same sound, although
you would have to adjust the volume to get this effect.
Mandelbrot additionally characterized white noise as having zero auto-correlation.
This means that the fluctuations in such a sound at any moment are completely unrelated
to any previous fluctuations.
In his book, Fractal Music, Hypercards, and More Martin Gardner describes an
algorithm for generating white tunes, those having no correlation between notes (Gard-
ner 1992). In this section we will implement his algorithms in Mathematica and compose
such tunes. We will then see how to generate tunes that have varying degrees of correla-
tion among the notes.
A simple melody with no correlation can be generated by randomly selecting notes
from a scale. First we generate the frequencies of the 12 semitones from an equal-tempered
C major scale. This is just a chromatic scale beginning with middle C.
In[12]:= Cmajor Table N 261.62558 2
j 12
, j, 0, 11
Out[12]= 261.626, 277.183, 293.665, 311.127, 329.628, 349.228,
369.994, 391.995, 415.305, 440., 466.164, 493.883
This plays the entire scale.
In[13]:= Timing Do Play Sin Cmajor j 2 t , t, 0, 1 ,
j, 1, Length Cmajor
Out[13]= 1.622 Second, Null
The reader who executes the above code will certainly notice that it is a bit slow.
Since we will be generating many sounds below, we will need to speed up the execution of
multiple sounds. The reason for the slowness has to do with how Play handles the func-
tions on which it operates. Normally, Play will compile the function that appears as its
argument, but it does not do this if what appears is only the name of a function defined
elsewhere. Cmajor was defined elsewhere, so it is not compiled. The following function
302 An Introduction to Programming with Mathematica
PlayTones will speed the evaluation immensely. Note the time for execution of the same
scale as compared with the Do loop above. (We have made the PlayTones function
Listable so that it will automatically map across lists of frequencies. Otherwise, we
would have to manually Map it across such lists.)
In[14]:= SetAttributes PlayTones, Listable
In[15]:= PlayTones freq_, time_: 0.5 :
Play Sin 2 t freq , t, 0, time
In[16]:= Timing PlayTones Cmajor, 1 ;
Out[16]= 0.21 Second, Null
Now we can quickly generate the tune. First, we randomly generate 20 frequencies
from the list Cmajor (we have suppressed the display of the graphics images).
In[17]:= randomnotes =
Table[Cmajor[[Random[Integer, {1,12}]]], {20}]
Out[17]= 466.164, 311.127, 391.995, 440., 277.183, 293.665, 293.665,
440., 391.995, 391.995, 329.628, 440., 311.127, 311.127,
493.883, 369.994, 369.994, 311.127, 261.626, 349.228
This plays the list of frequencies for half-second intervals each.
In[18]:= PlayTones randomnotes, 0.5
Out[18]= Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound
A listener would be hard-pressed to find a pattern or any autocorrelation in this
tune and the music is quite uninteresting as a result. Melodies generated using this
scaling are referred to as 1 f
0
, where the 0 loosely refers to the level of correlation.
We leave as an exercise the writing of more sophisticated white melodies one where
the duration of each note varies randomly, and another where the likelihood of a note
being chosen obeys a certain probability distribution.
Brownian music
We now move in the other direction and generate melodies that are overly correlated. We
will essentially perform a random walk through the C major scale. Music generated in
such a way is called Brownian because it behaves much like the movement of particles
suspended in liquid Brownian motion.
9 Graphics programming 303
Our melody will be constructed as follows: each note will be generated by randomly
moving up or down a few semitones from the previous note. When a sequence gets to one
end of the scale, we will simplify matters by having it wrap around to the other end.
We first create a function step that will randomly choose an integer from 2 to 2.
These steps will determine how many semitones to move up or down.
In[19]:= step : Random Integer, 2, 2
Instead of alternating between choosing a step size and moving up and down the
scale, we will first create a list of the steps in entirety. We will choose 20 steps correspond-
ing to 20 notes.
In[20]:= SeedRandom 0 ;
s20 Table step, 20
Out[21]= 1, 1, 1, 2, 0, 0, 2, 2, 1, 0, 2, 1, 1, 0, 1, 2, 2, 2, 1, 1
This list will correspond to first moving one step up, then two steps down, then two
steps up, etc. So, starting (arbitrarily) with the sixth element of the list Cmajor, the
following gives the positions of the notes to play.
In[22]:= FoldList Plus, 6, s20
Out[22]= 6, 7, 6, 7, 5, 5, 5, 3, 5, 4, 4, 6, 7, 6, 6, 7, 5, 3, 5, 6, 5
There is one problem with this approach. If we get to the end of the list (12th
position), and have to add two steps say, we would be stuck.
In[23]:= Cmajor 14
Part::partw :
Part 14 of 261.626, 277.183, 293.665, 6 , 440., 2
does not exist. More
Out[23]= 261.626, 277.183, 293.665, 311.127, 329.628, 349.228,
369.994, 391.995, 415.305, 440., 466.164, 493.883 14
The way around this is to use modular arithmetic. This will have the effect of wrap-
ping around to the opposite end of the list whenever you reach one boundary. Since the
list Cmajor is 12 elements long, we will use mod 11 and add 1. This will give us positions
1 through 12, as opposed to 0 through 11 if we used mod 12 alone. (Recall that Part[list,
0] gives the Head of list.)
In[24]:= pos Mod FoldList Plus, 4, s20 , 11 1
Out[24]= 5, 6, 5, 6, 4, 4, 4, 2, 4, 3, 3, 5, 6, 5, 5, 6, 4, 2, 4, 5, 4
304 An Introduction to Programming with Mathematica
Finally, we create a list of those frequencies from Cmajor at the positions given by
the above list pos.
In[25]:= brown Cmajor pos
Out[25]= 329.628, 349.228, 329.628, 349.228, 311.127, 311.127, 311.127,
277.183, 311.127, 293.665, 293.665, 329.628, 349.228, 329.628,
329.628, 349.228, 311.127, 277.183, 311.127, 329.628, 311.127
Here, then, is a function for generating the tones from a Brownian walk across the C
major scale. This function is set up so that the default range of steps is 2 to 2 (r_:2).
In[26]:= BrownMusic n_Integer, r_ : 2 : Module cmajor, steps ,
cmajor Table N 261.62558 2
j 12
, j, 0, 11 ;
steps Table Random Integer, r, r , n ;
cmajor Mod FoldList Plus, 4, steps , 11 1
This plays the tones with half-second intervals.
In[27]:= PlayTones BrownMusic 20 , 0.5
Out[27]= Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound , Sound
This melody has a different character from the 1 f
0
melody produced above. In fact,
it is so over-correlated that it is often referred to as 1 f
2
music as a result of a computed
spectral density of 1 f
2
. Although different in character from 1 f
0
music, it is just as
monotonous. The melody meanders up and down the scale aimlessly without any central
theme. The exercises contain a discussion of 1 f music (or noise); that is, music that is
moderately correlated. 1 f noise is quite widespread in nature and is intimately tied to
areas of science that study fractal behavior. John Casti, in his book Reality Rules: I, Picturing
the World in Mathematics gives the following characterization of 1 f noise: If an electrical
engineer were to compute the power spectrum (the squared magnitude of the Fourier
transform) f x of the relative frequency intervals x between successive notes in Bachs
Brandenburg Concerto, it would be found that over a large range f x c x, where c is some
constant. Thus Bachs music is characterized by the kind of noise that engineers call 1 f
noise. (The interested reader should consult Casti 1992 or Mandelbrot 1982.)
9 Graphics programming 305
Exercises
1. Evaluate Play[Sin[1000/x],{x,-2,2}]. Explain the dynamics of the sound
generated from this function.
2. Experiment with the Play function by creating arithmetic combinations of sin
functions. For example, you might try the following.
In[1]:= Play
Sin 440 2 t
Sin 660 2 t
, t, 0, 1
Out[1]= Sound
3. Create a tone that doubles in frequency each second.
4. A square wave consists of the addition of sine waves, each an odd multiple of a funda-
mental frequency; that is, it consists of the sum of sine waves having frequencies f
0
,
3 f
0
, 5 f
0
, 7 f
0
, etc. Create a square wave with a fundamental frequency of 440 hertz.
The more overtones you include, the squarer the wave.
5. Create a square wave consisting of the sum of sine waves with frequencies f
0
, 3 f
0
,
5 f
0
, 7 f
0
, etc., and amplitudes 1,
1
3
,
1
5
,
1
7
, etc. This is actually a truer square wave
than that produced in the previous exercise.
6. Create a square wave consisting of overtones that are randomly out of phase. How
does this wave differ from the previous two?
7. A sawtooth wave consists of the sum of both odd- and even-numbered overtones ( f
0
,
2 f
0
, 3 f
0
, 4 f
0
, etc. with amplitudes in the ratios 1,
1
2
,
1
3
,
1
4
, etc.) Create a sawtooth
wave and compare its tonal qualities with the square wave.
8. A wide variety of sounds can be generated using FM (frequency modulation) synthesis.
The basic idea of FM synthesis is to use functions of the form
a sin 2 F
c
, t mod sin 2 F
m
t
where a is the peak amplitude, F
c
is the carrier frequency in hertz, mod is the modula-
tion index, and F
m
is the modulating frequency in hertz.
Determine what effect varying the parameters has on the resulting tones by creating
a series of FM synthesized tones. First, create a function FM[Amp,Fc,mod,
Fm,time] that implements the above formula and generates a tone using the Play
function. Then you should try several examples to see what effect varying the parame-
ters has on the resulting tones. For example, you can generate a tone with strong
306 An Introduction to Programming with Mathematica
vibrato at a carrier frequency at middle A for one second by evaluating
FM[1,440,45,5,1].
9. Write a function pentatonic that generates 1 f
2
music choosing notes from a
five-tone scale. A pentatonic scale can be played on a piano by beginning with C ,
and then playing only the black keys: C , E , F , A , C . The pentatonic scale is
common to Chinese, Celtic, and Native American music.
10. Modify the routine for generating 1 f
0
music so that frequencies are chosen accord-
ing to a specified probability distribution. For example, you might use the following
distribution that indicates a note and its probability of being chosen: C 5%, C
5%, D 5%, E 10%, E 10%, F 10%, F 10%, G 10%, A 10%, A 10%,
B 5%, B 5%, C 5%. (Hint: Try the Which function.)
11. Modify the routine for generating 1 f
0
music so that the durations of the notes obey
1 f
0
scaling. Write a function tonesAndTimes that creates a two-dimensional list
of frequencies and time durations. Consider using the function MapThread.
12. If you read musical notation, take a musical composition such as one of Bachs
Brandenburg Concertos and write down a list of the frequency intervals x between
successive notes. Then find a function that interpolates the power spectrum of these
frequency intervals and determine if this function is of the formf x c x for some
constant c. (Hint: To get the power spectrum, you will need to square the magnitude
of the Fourier transform: take Abs[Fourier[]]^2 of your data.) Compute the
power spectra of different types of music using this procedure.
13. Modify the routine for generating 1 f
2
music so that the durations of the notes obey
1 f
2
scaling.
14. The following series of exercises are designed to create 1 f music music that is
mildly correlated.
a. Write a function Cmajor16 that extends Cmajor to 16 consecutive semitones.
b. Write three functions red, green, and blue that simulate rolling 3 six-sided
dice. The first note from cmajor16 is picked by rolling the dice and choosing
the note in the position given by the sum (mod 16) + 1.
c. To generate the next eight notes, think of the numbers 0 through 7 in binary. Let
red correspond to the 1s digit, green to the 2s digit, and blue to the 4s digit.
Starting from 0, and going to 1, only the 1s digit changes. So only the red die is
retossed, the blue and green are left alone. This new sum (mod 16) of the red,
green and blue is the next position from the list Cmajor16. The third roll is
obtained by noticing that in going from 1 to 2, both the 1s digit and the 2s digit
9 Graphics programming 307
change. Hence, reroll the red and green die, leaving the blue alone. The new
sum of the three dice is the position of the next note. Continue in this fashion,
rolling only those dice that correspond to digit changes when moving through
the numbers 07, base 2. Finally, generate the tones corresponding to these
frequencies.
d. Extend the above algorithm to include four dice to produce 16 notes from a
21-tone scale. If you have a sufficiently powerful computer with lots of memory
and disk space, try ten dice to produce 1,024 notes from a 55-tone scale.
308 An Introduction to Programming with Mathematica
10 Front end programming
In this chapter we extend the programming concepts we have covered thus far to the
objects that comprise the user interface, or front end. Because the objects that the
Mathematica user interacts with are themselves Mathematica expressions, all of the
tools that you use to do computations can also be used to create, manipulate, and alter
cells and notebooks themselves. We will first look at the underlying structure of these
objects and then discuss ways of manipulating them directly from within Mathematica.
10.1 Introduction
Up until this point, we have been primarily concerned with learning about programming
constructs and styles so that we can write programs to manipulate data or solve problems
from science, engineering, or mathematics. We have taken for granted that the space in
which we do our experimenting, prototyping, and documenting has been the Mathematica
notebook, an interface that has some similarities to a word processor document.
It is not uncommon now to add interactive elements to your documents to make
them more useful for yourself or the intended reader of your documents. With programs,
documentation, and papers all being created and used in electronic format, Mathematica
provides a seamless and well-integrated interface to these elements.
Another tool that is useful, especially for educators, are buttons that allow you to
hide your program code behind a familiar and easy-to-use interface element the button.
The user clicks on a button and an action happens that is determined by the underlying
code. For example, you might want to have calculus students quickly plot Taylor polyno-
mial approximations to a function together with the original function but do not want
them to spend time learning the syntax of such commands in Mathematica. You could
easily program an interface that would only require them to fill in a few parameters before
clicking a button to produce the desired plot.
In this chapter we will discuss the structure of cell and notebook expressions, look at
a few basic functions for manipulating these expressions, and then create several simple
examples that give a flavor of the kind of things that can be done with front end
programming.
Before we begin we should mention that this chapter is not intended as a complete
discussion of front end programming. An entire book could certainly be written on this
topic alone. This book is intended to give you an introduction to the many aspects of
programming with Mathematica and front end programming is certainly an appropriate
topic for that introduction. But there are several areas that cannot be included here, either
because of space limitations or because they do not fit under the introductory nature of
this book. These topics include front end options and front end tokens. An understanding
of each of these topics is quite important for more advanced front end programming. The
interested reader can delve further into this subject by looking in the Front End category
of the Help Browser or by searching the Mathematica Information Center online at
library.wolfram.com/infocenter.
10.2 The structure of cells and notebooks
We have spent a lot of time in this book focusing on the structure of Mathematica expres-
sions. In Chapter 2 we indicated that Mathematica expressions are of the form h[e
1
,e
2
,]
where h is the head of the expression and the e
i
are the elements which may themselves be
Mathematica expressions. We even went so far as to say that everything in Mathematica is
an expression. In this section we will learn that this statement extends to elements of the
front end, specifically to notebooks and cells.
Notebook expressions
Notebooks are ASCII files, meaning that you can open them in a text editor and view their
contents directly. If you were to do that, you would see that the underlying expression is a
Mathematica function called a Notebook. The notebook would look like this:
Notebook[{
Cell[string,style,options],
Cell[string,style,options],

},
options]
In other words, the Notebook is a function whose first argument is a list of one or
more Cell objects, followed by some options. The Mathematica kernel does not do
310 An Introduction to Programming with Mathematica
anything with this practically. It is the Mathematica front end that knows how to render
this expression as the familiar notebook.
For example, here is a very simple notebook that you could write in a text editor (of
course there is no reason to do that).
Notebook[{
Cell["Demo notebook", "Section"],
Cell["This is a text cell.","Text"],
Cell["1+2+3", "Input"]
}]
The Mathematica front end renders this expression in the familiar manner, a window.
Let us create the notebook from scratch using a kernel command, NotebookPut.
NotebookPut[expr] will create a notebook corresponding to expr in the front end and
make it the currently selected notebook.
In[1]:= nb NotebookPut
Notebook
Cell "Demo notebook", "Section" ,
Cell "This is a text cell", "Text" ,
Cell "1 2 3", "Input"
Out[1]= NotebookObject Untitled 1
10 Front end programming 311
Here is the notebook as viewed in the front end.
There is actually quite a lot going on behind the scenes here in terms of the interac-
tion between the kernel and the front end. As stated in Chapter 1, the kernel and the front
end are two separate programs that communicate with each other through a protocol
called MathLink. For purposes of efficiency, MathLink itself does not store the notebook in
memory but instead refers to it by means of a handle. These handles are called notebook
objects and are given as NotebookObject[fe,id], where fe is an object that refers to the
entire front end and id is an integer that is a unique identifier for that notebook. In the
example above, looking at the InputForm displays this information stored with the
notebook object.
In[2]:= InputForm nb
Out[2]//InputForm=
NotebookObject[FrontEndObject[LinkObject["3v8_shm",
1, 1]], 28]
Since we have assigned a symbol, nb, to this object, we can refer to it through this
symbol. NotebookGet gets the expression corresponding to this notebook and reads it
into the kernel. You should think of it as analogous to Get for packages.
In[3]:= NotebookGet nb
Out[3]= Notebook
Cell CellGroupData Cell Demo notebook, Section , Cell
This is a text cell, Text , Cell 1 2 3, Input , Open ,
FrontEndVersion 5.0 for Microsoft Windows,
ScreenRectangle 0., 1024. , 0., 681.
Notice that the front end has added two options to this notebook: FrontEndVer
sion and ScreenRectangle. It has also added some grouping information for the cells.
312 An Introduction to Programming with Mathematica
These are default behaviors of the front end and may vary from one front end to another.
They are also user-settable.
Manipulating notebooks
NotebookPut and NotebookGet are general functions for dealing with entire note-
books at once. There are a host of additional functions for manipulating parts of note-
books. You might first think that we can simply use functions like Part to extract a
particular part of a notebook we are interested in. There are several reasons why this is not
generally practical. First, because a notebook can contain many, many cells, it is often
quite difficult to determine precisely which part you want to work on. Secondly, since the
notebook resides in the front end, not the kernel, it is often not very efficient to manipu-
late the notebook directly by the kernel (although, if the notebook is small enough, this is
certainly possible).
As it turns out, there is a way around these issues and that is through something
referred to as the current selection, which is essentially a reference to the notebook
object. You could then think of the notebook manipulation functions as operating on
streams.
To see a list of the open notebooks, use Notebooks[].
In[4]:= Notebooks
Out[4]= NotebookObject Untitled 1 ,
NotebookObject 10FEProgramming.nb ,
NotebookObject Messages
Again, using InputForm, you can see the actual handles to each of the notebooks.
In[5]:= Notebooks InputForm
Out[5]//InputForm=
{NotebookObject[FrontEndObject[LinkObject["3v8_shm",
1, 1]], 28], NotebookObject[
FrontEndObject[LinkObject["3v8_shm", 1, 1]], 27],
NotebookObject[FrontEndObject[LinkObject["3v8_shm",
1, 1]], 7]}
Let us walk through some of the most common notebook operations you should
learn about. The first is NotebookCreate. As its name implies, this function will create a
new untitled notebook in the front end. We assign nb to be the handle to this notebook.
10 Front end programming 313
In[6]:= nb NotebookCreate
Out[6]= NotebookObject Untitled 2
Now let us write to the notebook. NotebookWrite takes two arguments: the first
argument is the notebook object that we are writing to; the second argument is what we
are writing. We will create a few different examples below.
A Cell is an expression with two arguments. The first argument is the contents of
the cell; the second argument is the cell style, a listing of which is under the Format Style
menu in the front end.
In[7]:= NotebookWrite nb, Cell "Here is some text.", "Text"
314 An Introduction to Programming with Mathematica
Adding options to Cell allows us to change some of the properties of the cell. For
example, here are several of the options that you can add.
In[8]:= Take Options Cell , 10, 15
Out[8]= Deletable True, PageWidth WindowWidth, Visible True,
CellFrame False, CellDingbat None, ShowCellBracket True
In[9]:= NotebookWrite nb,
Cell "Here is some more text.", "Text",
CellFrame True, CellDingbat
If we simply give a string as the second argument to NotebookWrite, Mathematica
will use the default cell type, Input.
In[10]:= NotebookWrite nb, "Here is some text."
10 Front end programming 315
Now suppose we wanted to insert an input cell with the expression 2
100
in it.
In[11]:= 2
100
Out[11]= 1267650600228229401496703205376
If you were to look at the underlying expression of the above cell (under the Format
menu, choose Show Expression), it would look like this:
Cell[BoxData[
SuperscriptBox["2", "100"]], "Input"]
We will talk about BoxData in just a moment, but we should be able to insert a cell
like this directly into our notebook object. Before we do this, notice that the insertion
point has been left inside the Input cell after the last NotebookWrite. To move the cell
insertion bar after the current cell, we will use SelectionMove which takes three argu-
ments: the notebook we are operating on, the direction to move, and the unit by which we
should move. The direction can be any of Next, Previous, After, Before, All. The
units are things like Word, Cell, CellGroup, Notebook (see the Help Browser under
SelectionMove for a complete description).
So, in our example, we want to move the selection just after the present cell.
In[12]:= SelectionMove nb, After, Cell
316 An Introduction to Programming with Mathematica
Now we can write the input cell to the notebook.
In[13]:= NotebookWrite nb,
Cell BoxData SuperscriptBox "2", "100" , "Input"
Notice that at the end of each NotebookWrite, the cell insertion bar was placed
just after the cell that was written, except in the case of writing input cells. Oftentimes, you
will need to move around within the notebook or select a particular cell (or other expres-
sion) and perform some operation on it. For example, suppose we would like to select the
previous cell (the one containing the 2
100
) in nb and evaluate it. We can do this with the
SelectionMove function.
In[14]:= SelectionMove nb, Previous, Cell
10 Front end programming 317
To evaluate the currently selected expression, use SelectionEvaluate.
In[15]:= SelectionEvaluate nb
Let us put a few of these pieces together and create a function that will evaluate the
next input cell. In Section 10.5 we will turn this code into a button.
For this example we will operate in the current notebook. We can refer to the
notebook in which these commands are being evaluated by EvaluationNotebook[].
First we select the current unit; that is, the cell in which the following code lives.
SelectionMove[EvaluationNotebook[],All,Cell]
Then we move the selection insertion to the next cell (at the moment, this code will only
work if it is immediately followed by an input cell).
SelectionMove[EvaluationNotebook[],Next,Cell]
Finally, we evaluate the currently selected input.
SelectionEvaluate[EvaluationNotebook[]]
318 An Introduction to Programming with Mathematica
Here we bundle this code up into the function EvaluateNext.
In[16]:= EvaluateNext :
SelectionMove EvaluationNotebook , All, Cell ;
SelectionMove EvaluationNotebook , Next, Cell ;
SelectionEvaluate EvaluationNotebook ;
Evaluating the cell containing EvaluateNext causes the immediately following cell to be
evaluated.
In[17]:= EvaluateNext
In[18]:= 2 2
Out[18]= 4
Exercises
1. Using NotebookPut, create a notebook with one Title cell, one Section cell, one
Text cell and two Input cells.
2. Use NotebookGet to read the notebook you created in Exercise 1 into the kernel.
Then programmatically change the Section cells to Subsection cells either using
Cases or an appropriate rule.
3. Take either of the notebooks you created in the above exercises and use Selection
Move and SelectionEvaluate to evaluate all of the Input cells in the notebook.
10.3 Cell data types
The cells in your notebooks often contain different kinds of data. Sometimes they will only
contain text. Other times they may contain formatted mathematical expressions, or possi-
bly a graphical object. Since the Cell data object has to handle each one of these kinds of
data, there is a mechanism that enables the front end to deal with these objects in a consis-
tent manner cell data types. We will look at a few of the most important and useful cell
data types in the next few sections.
10 Front end programming 319
TextData
Let us first look at a text cell that contains no special formatting.
Cell["Here is some text.", "Text"]
The formatted version of this cell looks like this:
Here is some text.
Adding some formatting to this cell causes a TextData wrapper to be added.
Cell[TextData[{
"Here is some ", StyleBox["italicized", FontSlant->"Italic"],
" text."
}], "Text"]
The formatted version of this cell looks like this:
Here is some italicized text.
Cells with TextData can contain a number of other data objects embedded in the cell.
For example, here is a text cell that contains a ValueBox. ValueBoxes provide a means
of embedding evaluations inside of your text cells.
Cell[TextData[{
"The current version is: ", ValueBox["$Version"]
}], "Text"]
The formatted version of this cell looks like this:
The current version is: 5.1 for Microsoft Windows
A listing of all of the possible ValueBox names that can be used can be found
choosing Create Value Display Object from the Input menu. Looking under the list of
global variables that can be used as the argument to ValueBox, you will see Date, for
example.
320 An Introduction to Programming with Mathematica
Cell[TextData[{
"The current date is: ",
ValueBox["DateLong"]
}], "Text"]
The formatted version of this cell looks like this:
The current date is: Friday, October 8, 2004
BoxData
Many of your cells in Mathematica will contain formatted mathematical expressions.
Whenever you work with these two-dimensional typeset objects, a different editor is
invoked, called the math editor. This is indicated in the front end by a pink background in
Text cell style on the typeset expression (you can enter a math typeset expression by
pressing Control-9). This is also indicated in the underlying cell structure by means of the
BoxData wrapper. For example, consider the following cell containing a superscript
expression.
Cell[BoxData[
RowBox[{
SuperscriptBox["x", "2"], "+", "y"}]], "Input"]
The formatted version of this cell looks like this:
x
2
y
There are several things to note here. First, we see that Mathematica has automati-
cally placed the elements x
2
, + and y all in something called a RowBox. This is how Mathe-
matica represents box objects or a series of strings.
Secondly, the x
2
object is represented internally as another box object, specifically
SuperscriptBox[x,2]. You can use DisplayForm to print box expressions in an
explicit two-dimensional form.
In[1]:= SuperscriptBox x, 2 DisplayForm
Out[1]//DisplayForm=
x
2
10 Front end programming 321
There are many different box objects in Mathematica. Below are just a few commonly
used box objects.
Cell[BoxData[
SqrtBox["2"]], "Input"]
The formatted version of this cell looks like this:
2
Cell[BoxData[
FractionBox["x", "y"]], "Input"]
The formatted version of this cell looks like this:
x
y
Cell[BoxData[
RowBox[{
SubsuperscriptBox[" ", "a", "b"],
RowBox[{"x", " ",
RowBox[{" ", "x"}]
}]
}]
], "Input"]
The formatted version of this cell looks like this:
a
b
x x
GraphicsData
Another type of data wrapper that you will encounter is GraphicsData, used to indicate
a graphical object in the cell. For example, creating a graphics object in the front end
displays a plot.
322 An Introduction to Programming with Mathematica
In[2]:= Plot Sin x , x, 0, 2 ;
1 2 3 4 5 6
1
0.5
0.5
1
If you unformat the graphics cell, the first few lines would look like the following:
Cell[GraphicsData["PostScript", "\<\
%!
%%Creator: Mathematica
%%AspectRatio: .61803
MathPictureStart
...
Normally you will not create graphics objects from scratch so it would seem as if
there is not too much you could do with GraphicsData objects manually. But suppose
you were interested in displaying your graphics to a notebook other than the one in which
you evaluate the graphics input. For example, we could use NotebookPut to write out a
new notebook containing a graphics cell object as follows:
In[3]:= MyDisplayChannel gr_ :
NotebookPut Notebook Cell GraphicsData
"PostScript", DisplayString gr , "Graphics"
This is now used by giving MyDisplayChannel as the value of DisplayFunc
tion for any plot you create.
In[4]:= Plot3D Sin x y , x, 0, 2 , y, 0, 2 ,
DisplayFunction MyDisplayChannel
Out[4]= NotebookObject Untitled 5
Evaluating the above expression will cause a new notebook window to be created in
your front end containing just the output of the Plot3D command, a graphic of the
surface sin x y .
10 Front end programming 323
Exercises
1. Using NotebookPut, create a notebook with several Text cells each containing a
ValueBox such as $Version, $OperatingSystem, and $UserName.
2. Using NotebookPut, create a notebook with an Input cell containing the integral
1
1 x
3
x. Then evaluate the integral using SelectionMove and SelectionEval
uate.
10.4 GridBoxes
ShowTable
Whenever you create a two-dimensional expression consisting of some number of rows
and columns, Mathematica represents that expression as a GridBox object. For example, if
you used the BasicInput palette to create a 2 2 matrix, it would be represented as follows:
Cell[BoxData[GridBox[{
{"a", "b"},
{"c", "d"}
}]], "Input"]
The formatted version of this cell looks like this:
a b
c d
Looking at the GridBox object, you should see that it is identical (structurally) to a
matrix in Mathematica, which is really just a list of lists.
In[1]:= FullForm
a b
c d
Out[1]//FullForm=
List List a, b , List c, d
In[2]:= a, b , c, d MatrixForm
Out[2]//MatrixForm=
a b
c d
324 An Introduction to Programming with Mathematica
Using GridBoxes, let us create a function for displaying arrays of data in a format-
ted table. First we create some sample data.
In[3]:= data " ", " ", " " ,
1.234, 2.3451, 3.4567801 , SqrtBox " " , "
x
y
", " n " ;
We can put this data into a GridBox and immediately print it in a two-dimensional
grid using DisplayForm.
In[4]:= GridBox data DisplayForm
Out[4]//DisplayForm=
1.234 2.3451 3.4567801
x
y
n
GridBox can be given several options that control its appearance.
In[5]:= Options GridBox
Out[5]= GridBaseline Axis, RowSpacings 1., ColumnSpacings 0.8,
ColumnWidths Automatic, RowAlignments Baseline,
ColumnAlignments Center , GridFrame False,
GridFrameMargins 0.4, 0.4 , 0.5, 0.5 ,
RowLines False, ColumnLines False, RowMinHeight 1.,
RowsEqual False, ColumnsEqual False,
AutoDelete True, AllowScriptLevelChange True,
MultilineFunction None, GridDefaultElement
Let us add a frame, make the margins around each grid element a bit larger than the
default, and add some lines between the rows and columns. Usually you will set the values
for GridFrame, RowLines, and ColumnLines to either True or False to enable or
disable these elements. Giving an explicit number as the value of each of these options
gives the thickness of the line that is drawn for that object.
In[6]:= GridBox data,
GridFrame 1.2, GridFrameMargins 1, 1 , 1, 1 ,
RowLines 1, ColumnLines 1 DisplayForm
Out[6]//DisplayForm=
1.234 2.3451 3.4567801
x
y
n
Now we can bundle up this code and turn all of it into a function, ShowTable. If we
wish, we can add some formatting, but to do so we have to wrap the GridBox in a Style
10 Front end programming 325
Box. FontSize, FontFamily, Background, and SingleLetterItalics are all
options to StyleBox.
In[7]:= ShowTable data_ : DisplayForm StyleBox
GridBox data,
GridFrame 1.2, GridFrameMargins 1, 1 , 1, 1 ,
RowLines 1, ColumnLines 1 ,
FontFamily "Times",
Background GrayLevel .8 , SingleLetterItalics True
In[8]:= ShowTable data
Out[8]//DisplayForm=
1.234 2.3451 3.4567801
x
y
n
Sometimes the data you work with will need to be manipulated in some way to
display it. The following is another example of the use of ShowTable, but one for which
we first need to think about the dimensions of our data. Consider displaying a table of
reciprocals of rep units, numbers consisting entirely of 1s.
In[9]:= RepUnit n_?Positive : Nest 10 #1 1 &, 1, n 1
In[10]:= expr Map
1
RepUnit #
&, Range 12
Out[10]= 1,
1
11
,
1
111
,
1
1111
,
1
11111
,
1
111111
,
1
1111111
,
1
11111111
,
1
111111111
,
1
1111111111
,
1
11111111111
,
1
111111111111
Since the above output contains 12 expressions, we need to explicitly partition it to
be rectangular. First we partition the data into rows of three elements (columns) each.
In[11]:= ShowTable Partition expr, 3
Out[11]//DisplayForm=
1
1
11
1
111
1
1111
1
11111
1
111111
1
1111111
1
11111111
1
111111111
1
1111111111
1
11111111111
1
111111111111
326 An Introduction to Programming with Mathematica
Here we partition the data into rows of four elements each.
In[12]:= ShowTable Partition expr, 4
Out[12]//DisplayForm=
1
1
11
1
111
1
1111
1
11111
1
111111
1
1111111
1
11111111
1
111111111
1
1111111111
1
11111111111
1
111111111111
In the above tables, we are manually partitioning the rows and columns into sublists
that will be rectangular when they are put into the table. It would be good programming
style to take that task from the user and do it automatically. We leave this as an exercise.
TriangleForm
In this section we will use GridBox to develop a function for displaying an array in a
triangular format. Such a function is quite useful for displaying the elements of Pascals
triangle in the familiar triangular array.
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
1 6 15 20 15 6 1
1 7 21 35 35 21 7 1
1 8 28 56 70 56 28 8 1
First let us create a function for generating the first n rows of Pascals triangle.
In[13]:= PascalTable rows_ :
Table Binomial n, m , n, 0, rows , m, 0, n
Here are the first four rows (including the 0th row).
In[14]:= expr PascalTable 3
Out[14]= 1 , 1, 1 , 1, 2, 1 , 1, 3, 3, 1
10 Front end programming 327
If we put empty strings around the elements in the appropriate places we can see
what the grid should look like.
In[15]:= GridBox
"", "", 1, "", "" ,
"", 1, "", 1, "" ,
1, "", 2, "", 1
DisplayForm
Out[15]//DisplayForm=
1
1 1
1 2 1
So we need to develop a function to insert these empty strings between each element
in each row and we also need to pad out each row to the length of the longest row in the
entire table. First we write the function to pad each row.
In[16]:= pad lis_ : PadLeft lis, 2 Length expr 1,
"", Round 2 Length expr 1 Length lis 2
In[17]:= pad expr 1
Out[17]= , , , 1, , ,
In[18]:= pad expr 2
Out[18]= , , , 1, 1, ,
Now to insert the appropriate number of empty strings between elements, let us first
manually insert space in a few rows.
In[19]:= Insert expr 2 , "", 2
Out[19]= 1, , 1
In[20]:= Insert expr 3 , "", 2 , 3
Out[20]= 1, , 2, , 1
In[21]:= Insert expr 4 , "", 2 , 3 , 4
Out[21]= 1, , 3, , 3, , 1
Here is the function to create the third argument for Insert.
In[22]:= Map List, Rest Range Length 1, 3, 3, 1
Out[22]= 2 , 3 , 4
328 An Introduction to Programming with Mathematica
Here is the function to add the appropriate amount of space between elements in
each row.
In[23]:= addspace lis_ :
Insert lis, "", Map List, Rest Range Length lis
In[24]:= addspace expr 3
Out[24]= 1, , 2, , 1
In[25]:= addspace expr 1
Out[25]= 1
This maps the addspace function across each row of the Pascal table.
In[26]:= expr Map addspace, PascalTable 3
Out[26]= 1 , 1, , 1 , 1, , 2, , 1 , 1, , 3, , 3, , 1
Then we pad out each row using our pad function developed above.
In[27]:= Map pad, expr
Out[27]= , , , 1, , , , , , 1, , 1, , ,
, 1, , 2, , 1, , 1, , 3, , 3, , 1
Finally we put this expression into a GridBox and display it.
In[28]:= GridBox % DisplayForm
Out[28]//DisplayForm=
1
1 1
1 2 1
1 3 3 1
Here is the TriangleForm function then consisting of the above pieces.
In[29]:= TriangleForm lis_List :
Module addspace, expr, len Length lis ,
addspace l_ :
Insert l, "", Map List, Rest Range Length l ;
expr Map addspace, lis ;
DisplayForm GridBox Map PadLeft #, 2 len 1,
"", Round
1
2
2 len 1 Length # &, expr
10 Front end programming 329
In[30]:= PascalTable 5 TriangleForm
Out[30]//DisplayForm=
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
Exercises
1. Modify ShowTable so that it can display a user-specified heading in the first row of
the grid. Include formatting to set the style of the strings in the heading to be differ-
ent than the rest of the elements displayed by ShowTable.
2. Modify ShowTable so that it automatically partitions the list it is passed to be
rectangular, with the number of rows and columns as close to each other as possible.
3. Create a function TruthTable[expr,vars] that displays the logical expression expr
together with all the possible truth values for the variables in the list vars. For exam-
ple, here is the truth table for the expression A B C.
In[1]:= TruthTable Implies A B, C , A, B, C
Out[1]//DisplayForm=
A B C A B C
T T T T
T T F F
T F T T
T F F F
F T T T
F T F F
F F T T
F F F T
You will first need to create a list of all possible truth value assignments for the
variables, A, B, C in this case. One approach would be to use Distribute. So,
essentially, this is the left-hand side, or first three columns of the above table (not
counting the first row containing the table headings).
330 An Introduction to Programming with Mathematica
In[2]:= vars A, B, C ;
len Length vars ; ins
Distribute Table True, False , len , List, List, List
Out[3]= True, True, True , True, True, False , True, False, True ,
True, False, False , False, True, True , False, True, False ,
False, False, True , False, False, False
You can then create a list of rules associating each of these triples of truth values with
a triple of variables.
In[4]:= Map Thread vars #1 &, ins
Out[4]= A True, B True, C True ,
A True, B True, C False , A True, B False, C True ,
A True, B False, C False , A False, B True, C True ,
A False, B True, C False , A False, B False, C True ,
A False, B False, C False
Substituting these rules into the logical expression produces a truth value for each of
the above rows.
In[5]:= Implies A B, C . Map Thread vars #1 &, ins
Out[5]= True, False, True, False, True, False, True, True
Your task is to put all these pieces together in a GridBox with appropriate
formatting.
10.5 Buttons
Buttons are very user-friendly objects whose functionality is familiar to any computer user.
From the programmers point of view, they allow you to hide your code behind a graphical
element, the button. Instead of writing a function and evaluating it by pressing Shift-Enter
from the keyboard, you pass the mouse cursor over the button and simply click. Whatever
code is hidden underneath the button is then evaluated.
In this section we will first look at the structure of ButtonBoxes and then create
some examples to demonstrate the variety of tasks that can be accomplished with buttons.
10 Front end programming 331
Making buttons the easy way
The simplest way to create buttons is to select an expression in your Mathematica note-
book, choose Create Button from the Input menu, and then activate your button. Let us
walk through these steps to create a button that pastes an expression into your notebook.
Suppose you were writing a paper in which you are discussing sequences and you
need to use an expression such as the following repeatedly in your notebook:
a
1
, a
2
, , a
n
. To create a button that would allow you to paste this expression into your
notebook by simply clicking that button, we first write down the expression we will work
with below in a regular input cell.
a
1
, a
2
, , a
n
Now select the entire expression and choose Create Button Paste from the Input
menu.
a
1
, a
2
, , a
n
Finally, to activate the button so that you can click it to have an action occur, select
the cell in which the button occurs and then choose Cell Properties Cell Active from the
Cell menu.
a
1
, a
2
, , a
n
Clicking the above button will paste the following at the insertion point:
a
1
, a
2
, , a
n
.
If you wished, you could create a free-standing palette from this button by choosing
Generate Palette from Selection from the File menu.
Although the above procedure for creating buttons is quite straightforward, it is only
convenient for fairly simple buttons. For more complicated buttons you will find that you
need a good understanding of the structure of buttons and the various options that control
their actions and display. We turn to those topics in the next few sections.
The structure of buttons
Buttons are created with the ButtonBox function in Mathematica. ButtonBox takes one
argument and by default, that argument is pasted at the current selection point.
In the examples that follow, we will use DisplayForm to display the button as an
interactive element. If you were to unformat your button (Show Expression from the
Format menu), you would see essentially all that precedes the DisplayForm below.
332 An Introduction to Programming with Mathematica
In[1]:= ButtonBox "some text", Active True DisplayForm
Out[1]//DisplayForm=
some text
Note that we have added the option Active True. This makes the resulting
button uneditable, one that is clickable. You will need to add this option to all your but-
tons to activate them. Clicking this button causes the following to be pasted at the current
selection point.
some text
Let us create a button that can serve as a template for a definite integral.
In[2]:= ButtonBox["Integrate[fun,{x,xmin,xmax}]", Active->True]
//DisplayForm
Out[2]//DisplayForm=
Integrate fun, x, xmin, xmax
Clicking the button causes the following to be pasted in.
Integrate fun, x, xmin, xmax
We can use placeholders in our template button so that the user can move from one
placeholder to the next by pressing the Tab key. The placeholder character can be
entered either from the Complete Characters palette (look under Letter-like Forms and
then Keyboard Forms), or directly from the keyboard by typing -sp- (pressing the
Escape key, then the characters s and then p, and finally the closing Escape key).
In[3]:= ButtonBox["Integrate[ ,{ , , }]", Active->True]
//DisplayForm
Out[3]//DisplayForm=
Integrate , , ,
Clicking on this button causes the following expression to be pasted. You can move
from one placeholder to another by pressing the Tab key.
Integrate , , ,
10 Front end programming 333
ButtonStyle
Although having buttons that paste their contents at the current selection point is useful,
there is much more that buttons can do. For example, they can wrap the contents of the
ButtonBox around a selected expression and then evaluate that expression. To change
the default behavior of buttons from simply pasting their contents to other actions, we
have to use the ButtonStyle option. ButtonStyle is used to control both the style
and the actions associated with your buttons. In the following example, ButtonStyle is
set to CopyEvaluateCell.
In[4]:= ButtonBox "Integrate ,x ", Active True,
ButtonStyle "CopyEvaluateCell" DisplayForm
The character is entered either from palettes or directly from the keyboard by
typing -spl- . Evaluating the above input produces the cell below. Selecting the input
cell containing Cos x
2
x
5
and then clicking the button causes the template to be
wrapped around the selected expression and then it is evaluated.
Out[4]//DisplayForm=
Integrate , x
In[5]:= Cos x
2
x
5
In[6]:= Integrate Cos x
2
x
5
, x
Out[6]=
x
6
6 2
FresnelC
2
x
If you were to use ButtonStyle EvaluateCell instead of CopyEvaluate
Cell, the button action would erase the selection and replace it with the new input and
the result.
Another very useful ButtonStyle is Hyperlink. Making a hyperlink is accom-
plished by creating a button out of some expression and setting the ButtonStyle option
to Hyperlink and adding the ButtonData option.
Cell[TextData[{
"Search for button on ",
ButtonBox["Google",
ButtonData:>{
URL[ "http://www.google.com"], None},
ButtonStyle->"Hyperlink"]
}], "Text"]
334 An Introduction to Programming with Mathematica
The formatted version of this cell looks like this:
Search for button on Google
Setting ButtonStyle to Hyperlink sets the button action to jump to some
location. That location is specified as the value of the option ButtonData. In this exam-
ple, that is set to URL["http://www.google.com"]. ButtonData set to a URL will
cause your web browser to be launched and opened to the location given as the argument
to the URL in this case http://www.google.com.
A list of all the possible ButtonStyle values is displayed in Table 10.1.
ButtonStylevalues Action
Paste pastes the contents default
Evaluate pastes, then evaluates in place
EvaluateCell paste, then evaluate entire cell
CopyEvaluate copy current selectioninto newcell,
then paste andevaluate
CopyEvaluateCell copy current selectioninto newcell,
then paste andevaluate cell
Hyperlink jumpto different location
Table 10.1: Possible ButtonStyles and associated actions
ButtonFunction
Whenever you need to put some Mathematica code inside your button, you will need to do
so as the value of the option ButtonFunction. You will also need to explicitly set the
option ButtonEvaluator which is set to None by default. The ButtonEvaluator
option tells the front end what program it should communicate with to process the con-
tents of the button function. Setting it to None tells the front end to communicate with
itself which is fine for operations like copying and pasting. But for operations that need to
communicate with a kernel, you will have to specify that explicitly. A value of Automatic
sends the code to the default kernel for the current notebook. If you had other kernels set
up, you could direct the button function at one of those.
10 Front end programming 335
In[7]:= ButtonBox "Compute 5 ",
Active True,
ButtonFunction Factorial 5 ,
ButtonEvaluator Automatic DisplayForm
Out[7]//DisplayForm=
Compute 5
Clicking this button will not cause any output to be displayed. This is because these
buttons are not evaluated in the kernel in the usual way as part of the main loop. In this
case, you can use Print to see the side effect of this computation.
In[8]:= ButtonBox "Compute 5 ",
Active True,
ButtonFunction Print Factorial 5 ,
ButtonEvaluator Automatic DisplayForm
Out[8]//DisplayForm=
Compute 5
120
You can use any Mathematica function you wish as the value of the ButtonFunc
tion option. But, in addition to the above issue with displaying output, you should be
aware of another important issue. As it turns out, the front end does not know how to
parse the special shorthand notation we often use for arithmetic and other operations. You
will be forced to use the FullForm of such expressions inside of your ButtonFunc
tion. So instead of 2+2, use Plus[2,2]; instead of {<<Graphics`;LogPlot[
Exp[x],{x,1,2}]} use CompoundExpression[Get["Graphics`", LogPlot[
Exp[x],{x,1,2}]]. Fortunately, the parser for the front end can recognize the short-
hand notation for List, Rule, and RuleDelayed, so you can use the shorthand nota-
tions {}, , and , respectively.
As a final example, we will create a button that loads a package and then performs a
computation with some functions from that package. Here is the code that we want to
encapsulate in our button.
In[9]:= Needs "Graphics`Polyhedra`"
336 An Introduction to Programming with Mathematica
In[10]:= Show Graphics3D Stellate Icosahedron ;
Here is the button code. Note that we have also added an option to ButtonBox to
set the background and set the entire cell to use the Times font family.
Cell[BoxData[
ButtonBox[RowBox[{"Stellate"," ","Icosahedron"}],
ButtonFunction->
CompoundExpression[Needs["Graphics`Polyhedra`"],
Show[Graphics3D[Stellate[Icosahedron[]]]]],
ButtonEvaluator->Automatic,
Background->GrayLevel[.5]]],
"Input",Active->True,
FontFamily->"Times",
FontColor->GrayLevel[1]]
And here is the button with a result of clicking it just below.
In[11]:= Stellate Icosahedron
10 Front end programming 337
Example: an evaluate button
At the end of Section 10.2 we created a function EvaluateNext, which evaluated the
immediately following input cell. In this section we will turn the code from that function
into a button.
Here was the code we developed in that section.
EvaluateNext:=(
SelectionMove[EvaluationNotebook[],All,Cell];
SelectionMove[EvaluationNotebook[],Next,Cell];
SelectionEvaluate[EvaluationNotebook[]];
)
To put this code inside a button, we need to make a few modifications. First, remem-
ber that the front end does not know how to parse shorthand notation such as ;. Instead
we need to use CompoundExpression. Second, instead of EvaluationNotebook, we
will use ButtonNotebook, which gives the notebook in which the current button lives.
Finally, we need to use ButtonCell to refer to the cell containing the button itself.
Putting all these pieces together, here is the ButtonFunction.
In[13]:= ButtonFunction CompoundExpression
SelectionMove ButtonNotebook , All, ButtonCell ,
SelectionMove ButtonNotebook , Next, Cell ,
SelectionEvaluate ButtonNotebook ;
Here then is the code to generate our evaluate button.
Cell[TextData[{
Cell[BoxData[
ButtonBox["EVALUATE",
ButtonFunction:>CompoundExpression[ {
SelectionMove[
ButtonNotebook[ ], All, ButtonCell],
SelectionMove[
ButtonNotebook[ ], Next, Cell],
SelectionEvaluate[
ButtonNotebook[ ]]}],
Active->True]]],
" MATHEMATICA INPUT"
}], "Text"]
338 An Introduction to Programming with Mathematica
And here is the formatted button. Clicking the Evaluate button causes the cell just
below the button cell to be evaluated.
EVALUATE MATHEMATICA INPUT
In[14]:= 2 2
Out[14]= 4
Finally, let us add some formatting to make this cell a little nicer looking.
Cell[TextData[{
Cell[BoxData[
ButtonBox[
StyleBox["EVALUATE",
FontFamily->"Helvetica",
FontSize->10,
FontWeight->"Bold"],
ButtonFunction:>CompoundExpression[ {
SelectionMove[
ButtonNotebook[ ], All, ButtonCell],
SelectionMove[
ButtonNotebook[ ], Next, Cell],
SelectionEvaluate[
ButtonNotebook[ ]]}],
Active->True,
Background->GrayLevel[0.500008]]]],
StyleBox[" MATHEMATICA INPUT",
FontFamily->"Helvetica",
FontSize->10,
FontWeight->"Bold",
FontSlant->"Italic",
FontColor->GrayLevel[1]]
}], "Text",
Background->GrayLevel[0.500008]]
Here is the formatted version of this code with the result of clicking the button.
EVALUATE MATHEMATICA INPUT
In[15]:= 2 5
Out[15]= 7
There is a little inefficiency in our code as we are calling the kernel several times (two
instances of SelectionMove and one of SelectionEvaluate) for what are essen-
10 Front end programming 339
tially front end operations, moving and selecting. You can send these sorts of commands
directly to the front end by wrapping them in FrontEndExecute. To distinguish
between the kernel command and the front end command you also need to append the
FrontEnd` context to the function. So, for example, instead of using Selection
Move[] in the kernel, you can send it directly to the front end with the following.
FrontEndExecute[FrontEnd`SelectionMove[]]
With this in mind, the EVALUATE button can be rewritten by only changing the
ButtonFunction.
ButtonFunction:>FrontEndExecute[ {
FrontEnd`SelectionMove[
ButtonNotebook[ ], All, ButtonCell],
FrontEnd`SelectionMove[
ButtonNotebook[ ], Next, Cell],
FrontEnd`SelectionEvaluate[
ButtonNotebook[ ]]}]
Another method of directly accessing front end commands is via front end tokens.
These tokens allow you to perform any menu command directly from the kernel. We will
not discuss them here, but for a detailed discussion of front end tokens, see the Front End
category of the Help Browser.
Exercises
1. Create a button that will serve as a template for the Plot3D function.
2. Create a button that will wrap Expand[] around any selected expression and
evaluate that expression.
3. Using GridBox, create a palette of buttons that operate on polynomials like that in
Exercise 2. Include in your palette a button for each of Expand, Factor, Apart,
and Together.
340 An Introduction to Programming with Mathematica
11 Examples and applications
The development of larger-scale Mathematica programming projects is discussed and
illustrated in this chapter. Each of the examples in this chapter contain numerous tasks
that need to work together and also integrate well with Mathematica. When you
develop such applications, it is important to think about how your functions work with
each other as well as how well they integrate with the rest of Mathematica. The user's
interface to your programs should be as close as possible to the built-in functions of
Mathematica so that users can more easily learn the syntax and usage. Features such as
options, argument checking, messaging, and documentation are all discussed in the
context of larger applications that are developed using all of the tools that were
developed in earlier chapters.
11.1 Manipulating data files
Introduction
One of the most common tasks for scientists and engineers is working with data sets that
have been generated by some external process or collector. If they are lucky, the data are
stored in a file that has a standard format and can then be read into other programs such as
Mathematica with ease using that programs importing functionality. Oftentimes, however,
data are stored in files with nonstandard formats and reading that file into a program such
as Mathematica requires some manual processing of that file to extract the required parts.
In this section we will walk through the steps of reading, manipulating, and visualiz-
ing a dataset that consists of solar radiation data collected by the Renewable Resource Data
Center, an organization that is managed by the US Department of Energy (interested
readers should visit rredc.nrel.gov). A copy of one such dataset has been placed in the
IPM3 files that are available for this book (see the Preface for details).
Getting the data into Mathematica
Our first task is to read the data into Mathematica. There are several functions that are
useful for getting data into Mathematica. One of them is Import, which is a good function
to try if you know your data are in one of the standard file formats supported by Import.
In this example, we will take a more general approach, one that can be used for somewhat
arbitrary file formats, with appropriate modifications.
When working with files that you need to read into Mathematica, you have several
options for how to deal with file locations. One option is to put your data file anywhere on
your system and then point to it in your Mathematica. For example, suppose you place a
data file testdata.txt in the following directory:
C:\Work\Project42\DataFiles\
In your notebook you could then create a path to this file as follows. (Note, this does
not read the file into Mathematica, it simply creates a path to the file.)
In[1]:= file ToFileName
"C:", "Work", "Project42", "DataFiles" , "testdata.txt"
Out[1]= C:\Work\Project42\DataFiles\testdata.txt
Any cells, such as the above, that you need to evaluate before doing any other work
in a notebook, can be turned into initialization cells by selecting those cell brackets and
then selecting Cell Properties Initialization Cell from the Cell menu.
Another strategy for setting up your work environment would be to put any Mathe-
matica commands such as that above in an init.m file that will then be read into the kernel
whenever the kernel is first started up. For more information on this approach, see the
subsection Creating Help Browser documentation in Section 11.2.
It is a common convention to put user applications and packages in one of several
Applications subdirectories. The two places to consider are given by the following. (Al-
though the input will be the same on all operating systems, the output will reflect the
directory structure of your operating system.)
In[2]:= ToFileName $BaseDirectory, "Applications"
Out[2]= C:\Documents and Settings\All Users\
Application Data\Mathematica\Applications\
342 An Introduction to Programming with Mathematica
In[3]:= ToFileName $UserBaseDirectory, "Applications"
Out[3]= C:\Documents and Settings\Paul Wellin\
Application Data\Mathematica\Applications\
$BaseDirectory is only writable by users who have administrative privileges.
Typically you put your applications there if you want to make them available to all users of
your computer. $UserBaseDirectory is only writable by the currently logged-in user
of you computer. This is the place to put your files if you do not have administrative
privileges or if you simply wish to keep your files accessible only to yourself.
The full list of the directories on Mathematicas search path is given by $Path.
In[4]:= $Path TableForm
Out[4]//TableForm=
C:\Program Files\Wolfram Research\Mathematica\5.0\AddOns\JLink
C:\Program Files\Wolfram Research\Mathematica\5.0\AddOns\NETLink
C:\Documents and Settings\Paul Wellin\Application Data\Mathematica\Kernel
C:\Documents and Settings\Paul Wellin\Application Data\Mathematica\Autoload
C:\Documents and Settings\Paul Wellin\Application Data\Mathematica\Applications
C:\Documents and Settings\All Users\Application Data\Mathematica\Kernel
C:\Documents and Settings\All Users\Application Data\Mathematica\Autoload
C:\Documents and Settings\All Users\Application Data\Mathematica\Applications
.
C:\Documents and Settings\Paul Wellin
C:\Program Files\Wolfram Research\Mathematica\5.0\AddOns\StandardPackages
C:\Program Files\Wolfram Research\Mathematica\5.0\AddOns\StandardPackages\StartUp
C:\Program Files\Wolfram Research\Mathematica\5.0\AddOns\Autoload
C:\Program Files\Wolfram Research\Mathematica\5.0\AddOns\Applications
C:\Program Files\Wolfram Research\Mathematica\5.0\AddOns\ExtraPackages
C:\Program Files\Wolfram Research\Mathematica\5.0\SystemFiles\Graphics\Packages
C:\Program Files\Wolfram Research\Mathematica\5.0\Configuration\Kernel
The files associated with this book are all stored in a directory IPM3 which should
live in the Applications directory in either $BaseDirectory or $UserBaseDirec
tory. Since both are on the path, the following designation sets up a system-independent
file name that is relative to the path given by $Path.
In[5]:= datafile ToFileName "IPM3", "DataFiles" , "23232.txt"
Out[5]= IPM3\DataFiles\23232.txt
Now suppose you have looked at this text file 23232.txt in a text editor (see display of
the first few lines of this file below) and noted that it contains strings and numbers and that
elements are separated by commas.
11 Examples and applications 343
"City: ","SACRAMENTO "
"State: ","CA"
"WBAN No: ", 23232
"Lat(N): ", 38.52
"Long(W): ",121.50
"Elev(m): ", 8
"Pres(mb):", 1015
"Stn Type:","Secondary"
"MONTHLY SOLAR RADIATION (kWh/m2/day)"
"COLUMN A: Year"
"COLUMN B: Month"
"COLUMN C: Flat-Plate Collector Facing South at Fixed Tilt=0"
"COLUMN D: Flat-Plate Collector Facing South at Fixed Tilt=Lat-15"
Program Listing 11.1: Display of first few lines of file 23232.txt
Back in Mathematica, you can use ReadList to read the file using some assumptions
about the data in the file.
In[6]:= rawdata ReadList datafile, Word,
WordSeparators ",",
RecordLists True,
RecordSeparators "\r\n", "\n", "\r" ;
ReadList takes two arguments: the first argument is the file that we are reading, in
this case, datafile; the second argument specifies the type of objects that are contained
in the file. Since we have a mix of strings and numbers in this file, we will simply assume
each entry is of type Word and manipulate the entries afterwards.
In addition to the two arguments to ReadList, we have also used several options
that state some assumptions about the form of the data and file. WordSeparators ","
indicates that elements in the file are assumed to be separated by commas. Record
Lists True indicates that each line of data from the file should be put in a separate
sublist in Mathematica. RecordSeparators {"\r\n","\n","\r"} specifies that any
of the three ways to end lines in text files (Windows, Unix, and Macintosh Classic) should
be considered. This is particularly useful if you do not know the origin of the operating
system on which your file was created or if you are unsure of how it has been transported
between operating systems.
344 An Introduction to Programming with Mathematica
Examining the data file
Now that we have read the data file into Mathematica, let us try to get a sense of its shape
and its contents so that we can start to determine in which parts we will be most interested.
Here is an abbreviated view showing the first few lines, an indication that 377 lines
are not displayed, and then the last few lines.
In[7]:= Short rawdata, 10
Out[7]//Short= "City: ", "SACRAMENTO " ,
"State: ", "CA" , "WBAN No: ", 23232 ,
"Lat N : ", 38.52 , "Long W : ", 121.50 ,
377 , 90, 9, 5.9, 7.0, 7.1, 6.8, 4.6, 8.7,
9.6, 9.7, 9.5, 9.7, 5.6, 6.9, 7.7, 7.7 ,
90, 10, 4.3, 5.8, 6.3, 6.4, 5.2, 6.5, 7.6,
7.9, 8.0, 8.1, 4.8, 4.9, 6.1, 6.2 ,
90, 11, 2.9, 4.2, 4.8, 5.1, 4.5, 4.2, 5.3,
5.7, 5.9, 6.0, 3.7, 2.9, 4.2, 4.4 ,
90, 12, 2.2, 3.5, 4.1, 4.5, 4.2, 3.3, 4.3,
4.8, 5.1, 5.2, 3.3, 2.3, 3.5, 3.8
Note that the data set contains 386 records, or lines. In Mathematica, we should think
of these records as sublists since ReadList reads each record in as a list.
In[8]:= Dimensions rawdata
Out[8]= 386
Here are the first nine lines of the data set. They give information about where the
solar data were collected. Note that these lines are strings of text.
In[9]:= Take rawdata, 9 TableForm
Out[9]//TableForm=
"City: " "SACRAMENTO
"State: " "CA"
"WBAN No: " 23232
"Lat N : " 38.52
"Long W : " 121.50
"Elev m : " 8
"Pres mb :" 1015
"Stn Type:" "Secondary"
"MONTHLY SOLAR RADIATION kWh m2 day "
The next several lines contain information about each of the columns later in the file
that contain the actual data. Again, these lines are strings. Although lines 10 through 25
contain this metadata about the columns, here we only display the first several.
11 Examples and applications 345
In[10]:= Take rawdata, 10, 16 TableForm
Out[10]//TableForm=
"COLUMN A: Year"
"COLUMN B: Month"
"COLUMN C: Flat Plate Collector Facing South at Fixed Tilt 0"
"COLUMN D: Flat Plate Collector Facing South at Fixed Tilt Lat 15"
"COLUMN E: Flat Plate Collector Facing South at Fixed Tilt Lat"
"COLUMN F: Flat Plate Collector Facing South at Fixed Tilt Lat 15"
"COLUMN G: Flat Plate Collector Facing South at Fixed Tilt 90"
The 26th record of this file is simply a column identifier for the data that follows.
In[11]:= Take rawdata, 26
Out[11]= "COL A", "COL B", "COL C", "COL D", "COL E",
"COL F", "COL G", "COL H", "COL I", "COL J", "COL K",
"COL L", "COL M", "COL N", "COL O", "COL P"
Starting at row 27, we have our actual data first the year, then the month, and then
several columns with numbers that represent solar radiation collected by different collec-
tors, measured in kilowatt hours per square meter per day.
In[12]:= Take rawdata, 27, 29
Out[12]= 61, 1, 1.6, 1.9, 2.0, 2.0, 1.6, 1.7,
2.0, 2.0, 2.1, 2.1, 0.7, 0.5, 0.8, 0.8 ,
61, 2, 3.0, 4.0, 4.3, 4.4, 3.7, 4.2, 4.9,
5.2, 5.3, 5.3, 2.8, 2.6, 3.4, 3.5 ,
61, 3, 4.3, 5.1, 5.2, 5.1, 3.7, 5.8, 6.4,
6.6, 6.5, 6.6, 3.2, 3.6, 4.2, 4.2
The data in these rows are still strings as a result of using the Word data type in
ReadList when we read in the file.
In[13]:= Map Head, Take rawdata, 27 , 2
Out[13]= String, String, String, String,
String, String, String, String, String, String,
String, String, String, String, String, String
To convert each of these elements to numbers, we need to map ToExpression
across each element.
In[14]:= Map ToExpression, Take rawdata, 27 , 2
Out[14]= 61, 1, 1.6, 1.9, 2., 2., 1.6,
1.7, 2., 2., 2.1, 2.1, 0.7, 0.5, 0.8, 0.8
346 An Introduction to Programming with Mathematica
In[15]:= Map Head, %, 2
Out[15]= Integer, Integer, Real, Real, Real, Real, Real,
Real, Real, Real, Real, Real, Real, Real, Real, Real
Extracting and converting data
We only wish to work with the actual data that represent the solar radiation values col-
lected on various dates. To extract only those rows that contain these numbers, we will
select those rows from rawdata that do not begin with a quote character.
In[16]:= data Select rawdata, StringTake # 1 , 1 "\"" & ;
Now we can turn each of the elements in data into a number using
ToExpression.
In[17]:= cleandata Map ToExpression, data, 2 ;
Here we can see the results of these operations by looking at the first two rows of
cleandata.
In[18]:= Take cleandata, 1, 2
Out[18]= 61, 1, 1.6, 1.9, 2., 2., 1.6, 1.7, 2., 2., 2.1,
2.1, 0.7, 0.5, 0.8, 0.8 , 61, 2, 3., 4., 4.3, 4.4,
3.7, 4.2, 4.9, 5.2, 5.3, 5.3, 2.8, 2.6, 3.4, 3.5
In[19]:= Map Head, cleandata 1
Out[19]= Integer, Integer, Real, Real, Real, Real, Real,
Real, Real, Real, Real, Real, Real, Real, Real, Real
Each of these rows in cleandata represent a year, a month, and a set of solar
radiation values collected during that month. We can use Select to pick out the row
whose first element (year) is 61 and whose second element (month) is 2; in other words, to
get the data corresponding to February 1961.
In[20]:= Select cleandata, # 1 61 && # 2 2 &
Out[20]= 61, 2, 3., 4., 4.3, 4.4, 3.7,
4.2, 4.9, 5.2, 5.3, 5.3, 2.8, 2.6, 3.4, 3.5
11 Examples and applications 347
Here is how we would pick out all records between August 1961 and January 1962.
In[21]:= Select cleandata,
# 1 61 && # 2 8 # 1 62 && # 2 1 &
Out[21]= 61, 8, 6.7, 7.1, 6.8, 6.2, 3.5, 9.5, 9.9, 9.7,
9.3, 9.9, 5.4, 7.2, 7.3, 7.6 , 61, 9, 5.8, 6.8, 7.,
6.8, 4.6, 8.6, 9.4, 9.6, 9.4, 9.6, 5.4, 6.8, 7.5, 7.5 ,
61, 10, 4.1, 5.4, 5.8, 5.9, 4.7, 6., 6.9, 7.3, 7.3,
7.4, 4.3, 4.3, 5.4, 5.5 , 61, 11, 2.5, 3.6, 4., 4.2,
3.7, 3.5, 4.3, 4.7, 4.8, 4.9, 2.9, 2.3, 3.2, 3.4 ,
61, 12, 1.6, 2.1, 2.3, 2.3, 2.1, 1.9, 2.3, 2.5, 2.5,
2.6, 1.2, 0.8, 1.3, 1.4 , 62, 1, 2.3, 3.3, 3.8,
4., 3.6, 3.2, 4., 4.3, 4.6, 4.6, 2.8, 2., 3., 3.2
It will be useful to have a more natural interface for selecting data based on this date
criteria. Here then is a function that we can use to easily select those records between two
dates, each given by a month and year.
In[22]:= GetData dat_, m1_, y1_ , m2_, y2_ : Select dat,
# 1 y1 && # 2 m1 # 1 y2 && # 2 m2 &
Using GetData, this picks out the data from August 1970 through January 1971.
In[23]:= GetData cleandata, 8, 70 , 1, 71
Out[23]= 70, 8, 7.4, 7.9, 7.6, 6.9, 3.7, 10.8, 11.2, 11.1,
10.6, 11.3, 6.6, 8.8, 8.9, 9.2 , 70, 9, 6.1, 7.2, 7.3,
7.1, 4.8, 9., 9.9, 10., 9.8, 10., 5.8, 7.2, 8., 8. ,
70, 10, 4.1, 5.4, 5.8, 5.9, 4.7, 6., 6.9, 7.3, 7.3,
7.3, 4.3, 4.4, 5.4, 5.5 , 70, 11, 2.2, 2.9, 3.2, 3.3,
2.8, 2.8, 3.4, 3.6, 3.7, 3.7, 1.9, 1.5, 2.1, 2.2 ,
70, 12, 1.4, 1.8, 1.9, 1.9, 1.6, 1.6, 1.9, 2., 2.,
2., 0.7, 0.5, 0.8, 0.9 , 71, 1, 2.1, 2.9, 3.3, 3.5,
3.1, 2.8, 3.4, 3.7, 3.9, 3.9, 2.2, 1.6, 2.3, 2.5
Visualizing the data
The third through 16th columns of rawdata contain solar radiation values that come
from different collectors, or perhaps one collector set at a different angle to the sun. Let us
take a look at just one of these.
In[24]:= Take rawdata, 15
Out[24]= "COLUMN F: Flat Plate
Collector Facing South at Fixed Tilt Lat 15"
348 An Introduction to Programming with Mathematica
Using GetData we can extract all those values for this particular collector (the sixth
column, referred to as "COL F") taken from January 1980 through December 1980,
In[25]:= d1 Part GetData cleandata, 1, 80 , 12, 80 , All, 6
Out[25]= 2.7, 3.6, 5.9, 5.7, 5.9, 5.9, 6.1, 6.7, 6.8, 6., 4.7, 3.1
and similarly for 1981.
In[26]:= d2 Part GetData cleandata, 1, 81 , 12, 81 , All, 6
Out[26]= 2.4, 4.3, 4.7, 6.2, 6.2, 6.2, 6.3, 6.8, 6.8, 5.8, 3.7, 2.2
Using MultipleListPlot (defined in Graphics`MultipleListPlot`), we
can quickly view these two datasets together.
In[27]:= Graphics`MultipleListPlot`
In[28]:= MultipleListPlot d1, d2, PlotJoined True,
AspectRatio Automatic, AxesLabel None, "kWh m
2
day" ;
2 4 6 8 10 12
4
5
6
kWh m
2
day
It would be easy to modify this plot in a variety of ways. For example, we could give
explicit month text for the horizontal axis tick marks.
In[29]:= months 1, "Jan" , 2, "Feb" , 3, "Mar" , 4, "Apr" ,
5, "May" , 6, "Jun" , 7, "Jul" , 8, "Aug" ,
9, "Sep" , 10, "Oct" , 11, "Nov" , 12, "Dec" ;
In[30]:= MultipleListPlot d1, d2, PlotJoined True,
AspectRatio Automatic, Ticks months, Automatic ,
AxesLabel None, "kWh m
2
day" ;
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
4
5
6
kWh m
2
day
11 Examples and applications 349
Of course lots of additional information could be added to the plot.
In[31]:= MultipleListPlot d1, d2, PlotJoined True,
AspectRatio Automatic, Ticks months, Automatic ,
AxesLabel None, "kWh m
2
day" ,
PlotLegend "1980", "1981" , LegendTextSpace 5,
LegendLabel "Lat N : 38.52, \nLong W : 121.50" ;
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
4
5
6
kWh m
2
day
1981
1980
Lat N : 38.52,
Long W : 121.50
The exercises ask you to bundle up the code developed in this section into several
functions and also to modify the graphical content and display it in several useful ways.
Exercises
1. Create a function ReadSolarData[file,opts] that reads in a solar data file such as
23232.txt using ReadList, strips out any line beginning with a quote character, and
returns the remaining lines with each element converted to a number. You should set
explicit options to ReadSolarData that borrow fromReadList.
2. Create a function that computes the total solar radiation for a given year from one
collector (your choice) from the data file 23232.txt. Make a plot comparing these
yearly radiation values for the history of the dataset.
3. Create a function PlotSolarData[dat
1
,dat
2
,opts] that uses MultipleList
Plot as in the previous section to plot datasets dat
1
and dat
2
. Your function should
include customized tick information, axes labels, and a legend that displays the
latitude and longitude of the collector in the LegendLabel. In addition, you
function should be able to accept options similar to MultipleListPlot and pass
them directly to MultipleListPlot.
4. Overload PlotSolarData so that when evaluated as PlotSolar
Data[month,opts], and month is one of January, February,, December, it will
produce a plot comparing the solar radiation for that month across all years of the
dataset.
350 An Introduction to Programming with Mathematica
11.2 Random walks
Introduction
Random walks are widely used to represent random processes in nature; physicists model
the transport of molecules, biologists model the locomotion of organisms, engineers model
heat conduction, and economists model the time behavior of financial markets, all with the
random walk model. This model can be envisioned by thinking of a person taking a
succession of steps, which are randomly oriented with respect to one another. It is a good
application of Mathematica to a problem that involves a diverse set of computing tasks.
In this section, we will develop a program for executing a random walk. Then we will
run the program and create a visualization of the walk that is created. In the course of our
application development, we discuss options, defaults, messaging, and documentation
issues.
The one-dimensional random walk
The simplest random walk model consists of n steps of equal length, back-and-forth along
a line. A step increment (or step) in the positive x direction corresponds to a value of 1 and
a step increment in the negative x direction corresponds to a value of 1. A list of the
successive step increments of an n-step random walk in one dimension is therefore a list of
n randomly selected 1s and 1s. This list can be generated in many ways. Here is one
straightforward implementation, generating a list of ten steps.
In[1]:= Table 1
Random Integer
, 10
Out[1]= 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
Using FoldList, we can generate a list of the n 1 locations of a one-dimensional
n-step walk, which starts at the origin.
In[2]:= FoldList Plus, 0, %
Out[2]= 0, 1, 2, 1, 0, 1, 0, 1, 2, 3, 4
We can now write the program walk1D to generate a list of the step locations of an
n-step random walk, originating at the origin.
In[3]:= walk1D n_ : FoldList Plus, 0, Table 1
Random Integer
, n
Here is a ten-step one-dimensional random walk using this walk1D program.
In[4]:= walk1D 10
Out[4]= 0, 1, 2, 1, 2, 1, 2, 3, 2, 3, 2
11 Examples and applications 351
A list of the step locations can also be generated without first creating a list of the
step increments, using the nesting operation.
In[5]:= walk1D2 n_ : NestList # 1
Random Integer
&, 0, n
This is just slightly faster than the previous approach.
In[6]:= n 10
6
;
Timing walk1D n ; , Timing walk1D2 n ;
Out[7]= 0.461 Second, Null , 0.31 Second, Null
Finally, we can plot the random walk using ListPlot.
In[8]:= ListPlot walk1D 1000 , PlotJoined True ;
200 400 600 800 1000
10
20
30
40
The two-dimensional random walk
The random walk model in two or more dimensions is more complicated than the random
walk in one dimension. Although each step of a one-dimensional walk is at 0 degrees
(forward) or 180 degrees (backward) with respect to the preceding step, in higher dimen-
sions each step can take a range of orientations with respect to the previous step.
We will first consider a random walk on a lattice. Specifically, we will look at a lattice
walk on the two-dimensional square lattice. This walk consists of steps of uniform length,
randomly taken in the North, East, South, or West direction. The list of the possible step
increments in this walk is given by the compass directions.
In[9]:= NSEW 0, 1 , 1, 0 , 0, 1 , 1, 0
Out[9]= 0, 1 , 1, 0 , 0, 1 , 1, 0
Here is a list of five step increments.
In[10]:= n 5;
NSEW Table Random Integer, 1, 4 , n
Out[11]= 0, 1 , 0, 1 , 1, 0 , 0, 1 , 1, 0
352 An Introduction to Programming with Mathematica
Here then is a program, called walk2D that generates a list of the step locations of
an n-step lattice walk starting at the origin {0, 0}.
In[12]:= walk2D n_ :
Module NSEW 0, 1 , 1, 0 , 0, 1 , 1, 0 ,
FoldList Plus, 0, 0 ,
NSEW Table Random Integer, 1, 4 , n
Here is a ten-step lattice walk in two dimensions.
In[13]:= walk2D 10
Out[13]= 0, 0 , 1, 0 , 0, 0 , 0, 1 , 1, 1 ,
0, 1 , 1, 1 , 1, 2 , 2, 2 , 2, 1 , 2, 2
Finally, here is a short function to generate a two-dimensional off-lattice walk. A
random angle, , is chosen between 0 and 2 and then a pair consisting of {cos , sin } is
generated. FoldList then iterates the process of adding one pair to the previous as above.
In[14]:= walk2DOffLattice n_ :
FoldList Plus, 0, 0 ,
Map Cos # , Sin # &, Table Random Real, 0, 2 , n
Visualizing the random walk
We will create a snapshot of the path of the two-dimensional walk using the graphics
primitive Line to draw lines between successive points in the walk.
In[15]:= ShowWalk2D coords_, opts___ :
Show Graphics Line coords , opts, AspectRatio Automatic
Here we have set the value of the AspectRatio option to Automatic so that
steps in the x and y directions will appear to have equal lengths in the plot. This option can
be overwritten by specifying a different value in the list of options given by opts. Note
the use of the triple blank in the definition of ShowWalk2D. The pattern opts___
matches any sequence (possibly empty) of rules which are used here to govern the display
of the graphic by changing certain options to the Graphics function. It is important that
opts appears before the option AspectRatio. This will allow you to override this (or
any) option value. If Mathematica sees an option listed more than once in a list of options,
it will only use the first such option. If opts had come at the end of this function, you
would not be able to change the value for AspectRatio.
11 Examples and applications 353
Here is a 125-step off-lattice walk.
In[16]:= ShowWalk2D walk2DOffLattice 125 , Axes Automatic ;
6 4 2 2 4 6
6
4
2
2
And here is a 25-step two-dimensional lattice walk.
In[17]:= ShowWalk2D walk2D 25 ;
A lattice walk repeatedly revisits sites that have been previously visited in the course
of its meandering. As a result, it is difficult, and usually impossible, to discern the history
of the walk from a snapshot of the path. The best way to see the entire evolution process of
the walk in an unobstructed fashion is to create an animation.
Creating an animation of a random walk in Mathematica is straightforward. This can
be explained using a short lattice walk as an example.
In[18]:= walk walk2D 10
Out[18]= 0, 0 , 1, 0 , 1, 1 , 1, 2 , 0, 2 ,
0, 1 , 0, 0 , 1, 0 , 1, 1 , 2, 1 , 1, 1
The animation consists of a sequence of graphics cells where the first cell shows the
first step of the walk (consisting of a line drawn between the first two elements in walk for
example) and each succeeding cell shows one more step than the previous cell. In general
then, the mth cell is drawn using the Line function and the first m 1 elements in walk.
All of the graphics cells are drawn by mapping Show across the walk list.
Map[(Show[Graphics[Line[Take[walk, #]]]])&,
Range[2, Length[walk] ]]
In general, objects in a graphics cell are scaled to fill the monitor screen. Therefore,
if we simply create cells, each containing a different number of steps of the walk using the
354 An Introduction to Programming with Mathematica
above graphics command, steps in one cell will appear to be of a different length than the
same steps in other cells. This will result in a jerky looking animation.
We can make all of the step lengths in all of the graphics cells uniform by using the
PlotRange option with the ordered pair of the minimum and maximum values of the
components of the random walk in each direction, {{xmin,xmax},{ymin,ymax}}. This
quantity can be determined by separating the x and y components of the walk using
Transpose and then mapping an anonymous function containing Min and Max on to it.
In[19]:= Map Min # , Max # &, Transpose walk
Out[19]= 2, 0 , 0, 2
Here then, is the overall program for creating the animation.
In[20]:= AnimateWalk2D coords_, opts___ :
Map
Show Graphics
Line Take coords, # ,
opts, AspectRatio Automatic,
PlotRange
Map Min # 1, Max # 1 &, Transpose coords &,
Range 2, Length coords
Note: We have added 1 to the maximum x and y values and subtracted 1 from the
minimum x and y values in order to enhance the display by making the graphics a little
smaller inside its bounding box. You might also wish to replace Map[Show[]] with
Scan[Show[]]. Scan is quite similar to Map but its main difference is that it does not
return an expression, so the Show is essentially a side effect of this computation.
While we can not see the random walk animation run in a book, we can look at the
graphics cells in the animation by creating a graphics array.
In[21]:= Show GraphicsArray Partition
AnimateWalk2D walk, DisplayFunction Identity , 5 ;
The option DisplayFunction Identity is used to suppress the display of the
individual graphics cells created by the AnimateWalk function (the GraphicsArray
has its own DisplayFunction function option whose default value is $DisplayFunc
tion) and the Partition function is used to specify the number of graphics in each row
of the GraphicsArray picture.
11 Examples and applications 355
The three-dimensional random walk
For a three-dimensional random lattice walk, we will use the vertices of a cube as our
directional vectors. We could input them manually, but they are defined in the
Graphics`Polyhedra` package so we may as well use those.
In[22]:= <<Graphics`Polyhedra`
In[23]:= NSEW3 Vertices Cube
Out[23]=
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
,
1
2
Here then is the lattice walk in three dimensions.
In[24]:= walk3D n_ : FoldList Plus, 0, 0, 0 ,
NSEW3 Table Random Integer, 1, 8 , n
In[25]:= walk3D 5
Out[25]= 0, 0, 0 ,
1
2
,
1
2
,
1
2
, 0, 0, 0 ,
1
2
,
1
2
,
1
2
,
2 , 2 , 2 ,
1
2
2 ,
1
2
2 ,
1
2
2
We can visualize this with only a slight modification to the ShowWalk2D function.
In[26]:= ShowWalk3D coords_, opts___ :
Show Graphics3D Line coords , opts, AspectRatio Automatic
In[27]:= ShowWalk3D walk3D 1200 ;
356 An Introduction to Programming with Mathematica
Finally, we create an off-lattice walk in three dimensions.
In[28]:= walk3DOffLattice n_ :
FoldList Plus, 0, 0, 0 , Map Cos # , Sin # ,
#
2
&,
Table Random Real, 2 , 2 , n
In[29]:= ShowWalk3D walk3DOffLattice 1000 ;
Adding options and defaults
When writing your own programs, it is often difficult to predict how a user will want to
use your functions. Programmers usually try to provide a variety of ways to use their
functions (allowing for different types of input, for example), or sometimes they write
separate functions to handle special cases. The problem with having a separate function for
each special case is that the user can soon become overloaded with the variety of functions
to learn.
In this section, we will show how to write options for your functions so that they
behave like the built-in options in Mathematica. (In Chapter 12 we will add some error-trap-
ping and messaging and we will see how to incorporate the use of options into a full-
fledged package.)
The use of options and defaults in your programs allows you to minimize the use of
many parameters and function names for the user to remember. For example, the built-in
function FactorInteger has an option GaussianIntegers, which, when set to
True, will factor a number over the Gaussian integers.
In[30]:= FactorInteger 5, GaussianIntegers True
Out[30]= , 1 , 1 2 , 1 , 2 , 1
11 Examples and applications 357
The alternative to such an option would be to have a separate function, say Factor
GaussianInteger, that the user would have to use. Since the main process here is
factorization of numbers, it makes sense to have one function that covers various situations
allowing for factorization over different domains by specifying different options.
On the other hand, polynomial factorization is a fundamentally different operation
from integer factorization, and so a different function is used for that.
In[31]:= Factor 27 x
5
81 x
4
y 9 x
3
y
2
73 x
2
y
3
32 x y
4
4 y
5
Out[31]= 3 x y
3
x 2 y
2
In the previous sections we developed five separate functions, walk1D, walk2D,
walk2DOffLattice, walk3D, and walk3DOffLattice that each generated random
walks, the only differences being in the dimension of the walk or whether the walk was on
the lattice or not. It is not practical to expect the user to remember five different function
names for what is essentially the same process. It would be far easier to create only one
function RandomWalk and set the dimension or lattice walk through the use of options.
We will define two options to RandomWalk, Dimension and LatticeWalk. The
LatticeWalk option will be specified as a rule and when set to True, it will generate a
lattice walk; when set to False, it will generate an off-lattice walk. The following defini-
tion both defines options for the RandomWalk function and specifies their default values.
In[32]:= Options RandomWalk LatticeWalk True, Dimension 2
Out[32]= LatticeWalk True, Dimension 2
If you were now to ask for information about the RandomWalk function, you would
see these new options listed.
In[33]:= ?RandomWalk
Global`RandomWalk
Options RandomWalk LatticeWalk True, Dimension 2
As far as the LatticeWalk option is concerned, we will use this option in the
RandomWalk function by branching to either a lattice walk or an off-lattice walk, depend-
ing upon the value of this new option. We will need to extract the value of this option
inside the RandomWalk function, which we do as follows:
latticeQ = LatticeWalk/.Flatten[{opts,Options[RandomWalk]}];
358 An Introduction to Programming with Mathematica
From right to left, the values of the options to RandomWalk are substituted into
opts; then these (rules) are substituted to extract the value of LatticeWalk. This value
is then assigned to the symbol latticeQ.
Similarly, we will extract the value of the option Dimension. But we want to use the
definitions given in the previous sections to branch appropriately, depending upon the
value of Dimension; that is, depending upon whether we wish a 1D, 2D, or 3D random
walk. The Which function is perfect for this task.
Which[
dim == 1, use walk1D definition,
dim == 2, use walk2D definition,
dim == 3, use walk3D definition]
Here then is the full function RandomWalk, using this option structure.
In[34]:= <<Graphics`Polyhedra`
In[35]:= RandomWalk n_Integer, opts___?OptionQ :
Module dim, latticeQ ,
latticeQ, dim LatticeWalk, Dimension .
Flatten opts, Options RandomWalk ;
Which
dim 1, walk1D n ,
dim 2, If latticeQ, walk2D n , walk2DOffLattice n ,
dim 3, If latticeQ, walk3D n , walk3DOffLattice n
Notice that if the LatticeWalk option has been set to True, then the first branch
of the If statement is followed, giving the lattice walk. If LatticeWalk has any other
value (False for example), then the off-lattice definition is used.
This uses the default value of LatticeWalk and the default value of Dimension
to create five steps of a two-dimensional lattice walk.
In[36]:= RandomWalk 5
Out[36]= 0, 0 , 0, 1 , 1, 1 , 1, 0 , 0, 0 , 0, 1
This creates an off-lattice walk.
In[37]:= RandomWalk 4, LatticeWalk False
Out[37]= 0, 0 , 0.282568, 0.959247 , 0.584254, 0.460629 ,
1.13491, 0.374105 , 1.82222, 1.10047
11 Examples and applications 359
Here is a three-dimensional lattice walk.
In[38]:= RandomWalk 4, Dimension 3
Out[38]= 0, 0, 0 ,
1
2
,
1
2
,
1
2
, 2 , 2 , 0 ,
1
2
2 ,
1
2
2 ,
1
2
, 2 , 0, 0
And here is a three-dimensional off-lattice walk.
In[39]:= RandomWalk 4, LatticeWalk False, Dimension 3
Out[39]= 0, 0, 0 , 0.895264, 0.445536, 0.426506 ,
1.53545, 1.21375, 0.212921 ,
1.47703, 2.21204, 0.0277754 ,
2.30801, 2.76835, 0.566116
Just as we have combined our various random walks into one function, so should we
combine the functions to visualize these walks, using Which to determine which branch to
take.
In[40]:= ShowWalk coords_, opts___ : Which
Dimensions coords 2 2,
Show Graphics Line coords , opts, AspectRatio Automatic ,
Dimensions coords 2 3, Show
Graphics3D Line coords , opts, AspectRatio Automatic
Here then are several examples of these functions.
In[41]:= ShowWalk RandomWalk 10
4
, Dimension 2, LatticeWalk False ;
360 An Introduction to Programming with Mathematica
In[42]:= ShowWalk RandomWalk 10
3
, Dimension 3, LatticeWalk True ;
In[43]:= ShowWalk RandomWalk 10
4
, Dimension 2 ,
Frame True, Background GrayLevel 0.9 ;
40 20 0 20 40 60 80 100
10
0
10
20
30
40
50
In the next chapter we will see how to bundle up all of these functions into a self-con-
tained package with various implementation details hidden from the user.
Error-trapping and messaging
In addition to error-trapping, messaging, and usage messages, another (even more impor-
tant) way to make user-defined functions behave like built-in functions is to check the
arguments to each function carefully and issue error messages when appropriate.
Good programming practice dictates that we try to anticipate how a user of our
programs will interact with them. In particular, it is good programming style to try and
catch any errors the user may make and respond with an appropriate message. For exam-
ple, the built-in Sin function will report an error and give you a warning message if you
give it the wrong number of arguments.
In[44]:= Sin 1.2, 3.4
Sin::argx :
Sin called with 2 arguments; 1 argument is expected. More
Out[44]= Sin 1.2, 3.4
One of the conditions we might want to check for with our RandomWalk function is
that the user enters a positive integer as its first argument. Let us first write the warning
message.
11 Examples and applications 361
In[45]:= RandomWalk::rwn "Argument `1` is not a positive integer.";
We can put a simple trap for the condition in the body of RandomWalk:
If[Not[IntegerQ[n]&&n>0], Message[RandomWalk::rwn,n],]
If the first argument to RandomWalk passes the test in this If statement (that is, if it
fails to be an integer or fails to be greater than 0), then a message will be generated substi-
tuting the argument n for `1` in the rwn message above.
Here is the rewritten RandomWalk function with the error trap included.
In[46]:= Clear RandomWalk
In[47]:= Options RandomWalk LatticeWalk True, Dimension 2 ;
In[48]:= <<Graphics`Polyhedra`
In[49]:= RandomWalk n_, opts___?OptionQ : Module dim, latticeQ ,
If Not IntegerQ n && n 0 , Message RandomWalk::rwn, n ,
latticeQ, dim LatticeWalk, Dimension .
Flatten opts, Options RandomWalk ;
Which
dim 1, walk1D n ,
dim 2, If latticeQ, walk2D n , walk2DOffLattice n ,
dim 3, If latticeQ, walk3D n , walk3DOffLattice n
Now if we pass a noninteger or negative argument to RandomWalk, the warning will
be triggered.
In[50]:= RandomWalk 6
RandomWalk::rwn : Argument 6 is not a positive integer.
In[51]:= RandomWalk 10.5
RandomWalk::rwn : Argument 10.5` is not a positive integer.
362 An Introduction to Programming with Mathematica
Creating Help Browser documentation
Whenever you distribute any Mathematica applications, users will appreciate the inclusion
of a set of help files that describe your application in some detail. You can incorporate this
documentation into Mathematicas Help Browser so that users access your documentation
in the same manner as they access the documentation that comes with Mathematica. In this
section we will describe how to go about doing that.
The directory and file structure of your application is referred to as its layout. Your
documentation, which will consist of Mathematica notebooks, can be viewed in the Help
Browser by creating some specific directories and placing certain files in these directories.
The top-level directory name we will use for our random walk application will be
RandomWalks. Let us first identify the directory on your computer system where you
should place this RandomWalks directory. It is common convention to put user-defined
applications in one of several Applications directories. The full path to these directories on
your computer can be given as follows:
In[52]:= ToFileName $BaseDirectory, "Applications"
Out[52]= C:\Documents and Settings\All Users\
Application Data\Mathematica\Applications\
In[53]:= ToFileName $UserBaseDirectory, "Applications"
Out[53]= C:\Documents and Settings\Paul Wellin\
Application Data\Mathematica\Applications\
$BaseDirectory is writable by anyone with administrative privileges on your
computer and is readable by everyone on your computer. $UserBaseDirectory is only
writable and readable by the currently logged in user of your computer. For our purposes
here we will use $BaseDirectory, but either location is fine.
Inside the base directory there should be an Applications directory. If it does not
already exist you will have to create it. Then the following directories for our random
walks application should be created inside the applications directory.
RandomWalks
Documentation
English
FrontEnd
Palettes
StyleSheets
Kernel
11 Examples and applications 363
So the directory StyleSheets should be created inside the FrontEnd directory which
should be a subdirectory of RandomWalks. Any style sheets that you define for your
application should go in the FrontEnd/StyleSheets directory. Similarly for any palettes
that you want to be associated with your application. The kernel directory can contain an
init.m file that may have some Mathematica commands that you want to be evaluated every
time your application is loaded.
All of the notebooks that you want to appear in the Help Browser need to be placed
in the Documentation/English directory. For example, in our random walks application,
we have placed the notebook RandomWalks.nb inside of the directory RandomWalks/-
Documentation/English.
Finally, you must create the text file BrowserCategories.m in the directory
Documentation/English. This file will identify precisely how your documentation will
appear in the Help Browser. Here is a very simple BrowserCategories.m file from the
RandomWalks example.
BrowserCategory["Random Walks", None, {
Item["Introduction", "RandomWalks.nb", CopyTag->"rw:1"],
Item["The One-Dimensional Random Walk", "RandomWalks.nb",
CopyTag->"rw:2"],
Item["The Two-Dimensional Random Walk", "RandomWalks.nb",
CopyTag->"rw:3"],
Item["Visualizing the Random Walk", "RandomWalks.nb", CopyTag->"rw:4"],
Item["The Three-Dimensional Random Walk", "RandomWalks.nb",
CopyTag->"rw:5"],
Item["Adding Options and Defaults", "RandomWalks.nb", CopyTag->"rw:6"],
Item["Error-Trapping and Messaging", "RandomWalks.nb",
CopyTag->"rw:6"],
Item["Creating Help Browser Documentation", "RandomWalks.nb",
CopyTag->"rw:7"]
}]
The opening BrowserCategory takes three arguments. The first is the name of
the category as it will appear at the top level in the Help Browser. The second argument is
the name of the subdirectory in which your notebook source files live. If it is the same as
the directory in which the BrowserCategories.m lives (which is typically where it is), then
use None as the name. Finally, the third argument is a list of Item commands. The Item
function takes the following form: Item[name, filename, options]. The name gives the
subcategory name (typically section or subsection names in your source notebook), the
filename is the file in which the documentation is found, and the options typically are
tagging commands.
364 An Introduction to Programming with Mathematica
In our example BrowserCategories.m file we have used CopyTag to identify the
specific set of cells within the RandomWalks.nb notebook that are associated with a
particular Item. In the notebook itself, you will have to tag the corresponding cells using
the Add/Remove Cell Tags command in the Find menu or using a tool such as AuthorTools
(an application itself that comes with Mathematica) to assist you with doing this somewhat
automatically.
Once your BrowserCategories.m file is created and placed in the Documentation/En-
glish directory of your application, you will need to rebuild the Help index of the Help
Browser by choosing that item under the Help menu. The documentation for your applica-
tion should then appear in the Help Browser under the Add-ons & Links category.
You should feel free to modify the RandomWalk example application that comes
with the IPM3 materials by editing the BrowserCategories.m file or using it as a template
for your own applications.
Exercises
1. The version of RandomWalk developed in this section generates one-dimensional
walks of unit step. Modify RandomWalk so that the step size is a uniformly distrib-
uted random number between 1 and 1.
2. Modify ShowWalk so that it produces a ListPlot when passed something of the
form RandomWalk[n,Dimension 1].
3. The RandomWalk program developed in this section is not set up properly to take
unit steps in three dimensions on the off-lattice walk. The following formulas can be
used to represent a point parametrically on the unit sphere.
x , cos 1 cos
2
y , sin 1 cos
2
z cos
Use these formulas to rewrite walk3DOffLattice so that off-lattice three-dimen-
sional walks take unit steps.
4. The square end-to-end distance of a two-dimensional walk is defined as
x
f
x
i
2
y
f
y
i
2
, where x
i
, y
i
and x
f
, y
f
are the initial and final locations of the
walk, respectively. Assuming the initial point is the origin, then this simplifies to
x
f
2
y
f
2
. Write a function SquareDistance that takes a two-dimensional walk as an
11 Examples and applications 365
argument and computes the square end-to-end distance. Write a usage message and
include this function as a publicly exported function in the RandomWalks.m package.
5. Create a function AnimateWalk that takes RandomWalk[n,Dimension 2] as
an argument and produces a series of graphics that can be animated by displaying in
quick succession. Include in your graphics a red disk that moves to the current
position in the walk. The viewer will then see this red disk moving along on the
random walk as the animation is played.
6. Modify AnimateWalk from the previous exercise so that it can also accept the
output from RandomWalk[n,Dimension 3].
11.3 The Game of Life
A cellular automaton is a system of discrete lattice sites, each of which has a value (usually
an integer) associated with it. The values of the sites change simultaneously, in a succession
of discrete time steps, by applying rules that depend on the values of a site and the sites in
its vicinity.
Cellular automata have been used to model various physical, chemical, biological,
and social phenomena (Gaylord and Wellin 1995). In principle, any process that can be
described by an algorithm or program can be modeled by a cellular automaton.
The Game of Life, created by the British mathematician John Conway, is the most
well-known cellular automaton. It is the forerunner of so-called artificial life (or a-life)
systems and it was the first program run on the first parallel processing computer. It has
been estimated that more computer time has been spent (or wasted, depending on your
point of view) running the Game of Life program than any other program.
We will show how to program the Game of Life in Mathematica, so that it is opti-
mized for efficiency (run speed). This is a good application to work on at this point as its
implementation covers many of the topics from earlier chapters of the book: functional vs.
procedural programming, rule-based programming, setting attributes, and many more.
The Game of Life is played on a two-dimensional square Boolean lattice where sites
have values of either 0 or 1. A site with value 1 is said to be alive and a site with value 0 is
said to be dead. To illustrate the computations involved in the Game of Life program, we
will use the following small lattice system.
366 An Introduction to Programming with Mathematica
In[1]:= GameBoard Table Random Integer , 4 , 4 ;
TableForm GameBoard
Out[2]//TableForm=
1 0 0 0
1 0 0 1
0 0 1 1
0 0 1 1
In order to update a site of GameBoard, the sum of the values of the sites in its
neighborhood must be determined.
The neighborhood of a site in GameBoard consists of the site and the eight nearest
neighbor sites lying North, Northeast, East, Southeast, South, Southwest, West, and
Northwest of the site.
The neighborhood of a site located in the interior of the lattice is obvious. For
example, the nearest neighbor sites to the {2,3} site (which lies in the second row, third
column of GameBoard) are the {1,3}, {1,4}, {2,4}, {3,4}, {3,3}, {3,2}, {2,2},
and {1,2} sites.
The neighborhood of a site lying on one of the borders of the lattice is less apparent.
Employing what are known as periodic boundary conditions, some of the nearest neighbor
sites are taken from the opposing borders. A non-corner site located in the first or last row
(column) of the board has the corresponding site in the last or first row (column) as a
nearest neighbor site, respectively, and a corner site has the two sites in the opposing
corner as two of its nearest neighbor sites. For example, the nearest neighbor sites to the
{2,4} site (which lies in the second row, last column of GameBoard) are the {1,4},
{1,1}, {2,1}, {3,1}, {3,4}, {3,3}, {2,3}, and {1,3} sites.
The 16 neighborhoods of the sites in the lattice system can be generated in two steps.
An expanded matrix is created by first copying the first element in each row on to the
end of the row and copying the last element in each row on to the front of the row, and
then copying the first row on to the end of the list of rows and copying the last row on to
the front of the list of rows. The following anonymous function can be used to perform
this operation.
In[3]:= wrap Join Last #1 , #1, First #1 &;
11 Examples and applications 367
The application of the wrap function to GameBoard is shown below.
In[4]:= wrap Map wrap, GameBoard TableForm
Out[4]//TableForm=
1 0 0 1 1 0
0 1 0 0 0 1
1 1 0 0 1 1
1 0 0 1 1 0
1 0 0 1 1 0
0 1 0 0 0 1
The expanded matrix created by applying the wrap function to the lattice can be
partitioned into overlapping three-by-three matrices to create a list of the neighborhoods
of the sites in the lattice.
In[5]:= Neighborhoods Partition
wrap Map wrap, GameBoard , 3, 3 , 1, 1 TableForm
Out[5]//TableForm=
1 0 0
0 1 0
1 1 0
0 0 1
1 0 0
1 0 0
0 1 1
0 0 0
0 0 1
1 1 0
0 0 1
0 1 1
0 1 0
1 1 0
1 0 0
1 0 0
1 0 0
0 0 1
0 0 0
0 0 1
0 1 1
0 0 1
0 1 1
1 1 0
1 1 0
1 0 0
1 0 0
1 0 0
0 0 1
0 0 1
0 0 1
0 1 1
0 1 1
0 1 1
1 1 0
1 1 0
1 0 0
1 0 0
0 1 0
0 0 1
0 0 1
1 0 0
0 1 1
0 1 1
0 0 0
1 1 0
1 1 0
0 0 1
Given the neighborhoods of the sites on the lattice, we can determine whether a site
is alive or dead and how many of its nearest neighbor sites are alive. These are the two
quantities which appear in the rules used to update a site.
The three life and death rules for updating a site in the lattice are:
1. A living site (a site with value 1) with exactly two living nearest neighbor sites
remains alive (its value is updated to 1).
2. Any site (a site with value 0 or 1) with three living nearest neighbor sites stays
alive or is born (its value is updated to 1).
3. Any other site (a site with value 0 or 1) remains dead or dies (its value is updated
to 0).
368 An Introduction to Programming with Mathematica
A conditional function which, given the neighborhood of a site, applies the appropri-
ate rule is given below.
In[6]:= LiveOrDie lis_ : Module neighbors ,
neighbors Count lis, 1, 2 ;
If lis 2, 2 1 && neighbors 4 neighbors 3, 1, 0
Applying the LiveOrDie function to the neighborhoods of GameBoard yields the
updated GameBoard.
In[7]:= Map LiveOrDie, Neighborhoods, 2 TableForm
Out[7]//TableForm=
1 1 1 0
1 1 1 0
0 1 0 0
1 1 1 0
Finally, the evolution of the lattice over t time steps, or until it stops changing, is
carried out using FixedPointList.
FixedPointList[Map[LiveOrDie,
Partition[wrap[Map[wrap,#]],{3,3},{1,1}],{2}]&,GameBoard,t]
The code fragments developed above can be used to construct a program for the
Game of Life. However, while this program will work, it is unduly slow. A much more
efficient (faster running) program for the Game of Life can be developed by following
some general Mathematica programming guidelines.
The most efficient way to program in Mathematica is to utilize the following
approaches as much as possible:
avoid looping
minimize conditional branching
manipulate data structures in their entirety
employ built-in Mathematica functions
use anonymous functions, higher-order functions, and nested function calls
create look-up tables
11 Examples and applications 369
The use of these principles is well illustrated in the Game of Life program we will
now develop.
A matrix whose elements are the number of living, nearest neighbor sites to the
corresponding sites in the Game of Life lattice can be computed directly from the lattice
without having to first create the neighborhoods of the lattice, using the following
function.
In[8]:= liveNeighbors mat_ :
Apply Plus, Map RotateRight mat, # &, 1, 1 , 1, 0 ,
1, 1 , 0, 1 , 0, 1 , 1, 1 , 1, 0 , 1, 1
The liveNeighbors function makes use of the fact that Mathematica adds lists by
vector addition, adding the corresponding elements of the lists. Applying the function to
the GameBoard example, we get
In[9]:= liveNeighbors[GameBoard]//TableForm
Out[9]//TableForm=
3 3 3 5
3 3 3 4
4 3 4 5
3 3 3 4
Comparing this output with the Neighborhoods matrix created earlier, we can see
that each element in liveNeighbors[GameBoard] is the number of living nearest
neighbor sites to the corresponding site in GameBoard.
We can write down site update rules, whose two arguments are the value of a site and
the sum of the values of the nearest neighbor sites in its neighborhood. These rules are a
direct translation of the life and death rules from words to code.
In[10]:= update 1, 2 : 1
update _, 3 : 1
update _, _ : 0
SetAttributes update, Listable ;
The update rule is given the Listable attribute, so, when it is applied to a matrix
of site values and also to a matrix of the number of living neighbors to these sites, a matrix
is created whose elements are obtained by applying the update function to the correspond-
ing elements of the two matrices. This behavior can be demonstrated using a general
function, g, with the GameBoard and liveNeighbors[GameBoard] matrices.
In[14]:= SetAttributes g, Listable ;
370 An Introduction to Programming with Mathematica
In[15]:= g GameBoard, liveNeighbors GameBoard
Out[15]= g 1, 3 , g 0, 3 , g 0, 3 , g 0, 5 ,
g 1, 3 , g 0, 3 , g 0, 3 , g 1, 4 ,
g 0, 4 , g 0, 3 , g 1, 4 , g 1, 5 ,
g 0, 3 , g 0, 3 , g 1, 3 , g 1, 4
Using the update rules with the GameBoard and liveNeighbors matrices, and
comparing the result obtained earlier by applying the LiveOrDie function to the Neigh
borhoods of GameBoard, we see that each site in the board has been correctly updated.
In[16]:= update GameBoard, liveNeighbors GameBoard TableForm
Out[16]//TableForm=
1 1 1 0
1 1 1 0
0 1 0 0
1 1 1 0
Note: While the three update rules overlap with one another, there is no confusion as
to when each rule is used because Mathematica applies more specific rules before more
general rules. Thus, while a site with value 1 and 2 nearest neighbor sites with value 1 will
satisfy both the first and third rules, the first rule is used because it is the most specific
applicable rule. Similarly, while a site having three nearest neighbor sites with value 1 will
satisfy both the second and third rules, the second rule is used because it is the most
specific applicable rule. The third rule is more general than the other two rules and hence
is only used if neither of the other rules can be used.
The evolution of the lattice over t time steps can be carried out using an anonymous
function, where # represents the lattice configuration in FixedPointList.
update[#, liveNghbrs[#]]&
Using the GameBoard example and three time steps to illustrate this, gives
In[17]:= FixedPointList update #, liveNeighbors # &, GameBoard, 3
Out[17]= 1, 0, 0, 0 , 1, 0, 0, 1 , 0, 0, 1, 1 , 0, 0, 1, 1 ,
1, 1, 1, 0 , 1, 1, 1, 0 , 0, 1, 0, 0 , 1, 1, 1, 0 ,
0, 0, 0, 0 , 0, 0, 0, 0 , 0, 0, 0, 0 , 0, 0, 0, 0 ,
0, 0, 0, 0 , 0, 0, 0, 0 , 0, 0, 0, 0 , 0, 0, 0, 0
11 Examples and applications 371
Let us take the Transpose of this result in order to interchange rows and columns
and facilitate a comparison with our previous results.
In[18]:= Map Transpose, %
Out[18]= 1, 1, 0, 0 , 0, 0, 0, 0 , 0, 0, 1, 1 , 0, 1, 1, 1 ,
1, 1, 0, 1 , 1, 1, 1, 1 , 1, 1, 0, 1 , 0, 0, 0, 0 ,
0, 0, 0, 0 , 0, 0, 0, 0 , 0, 0, 0, 0 , 0, 0, 0, 0 ,
0, 0, 0, 0 , 0, 0, 0, 0 , 0, 0, 0, 0 , 0, 0, 0, 0
The code fragments given above are combined into the Game of Life program.
In[19]:= LifeGame n_Integer?Positive, steps_ :
Module gameboard, liveNeighbors, update ,
gameboard Table Random Integer , n , n ;
liveNeighbors mat_ :
Apply Plus, Map RotateRight mat, # &,
1, 1 , 1, 0 , 1, 1 , 0, 1 ,
0, 1 , 1, 1 , 1, 0 , 1, 1 ;
update 1, 2 : 1;
update _, 3 : 1;
update _, _ : 0;
SetAttributes update, Listable ;
FixedPointList
update #, liveNeighbors # &, gameboard, steps
The input parameters, n and steps, are, respectively, the linear size of the lattice
and the maximum number of time steps carried out.
Finally, the focus in playing the Game of Life is on identifying various patterns of 1s
amongst the 0s, and observing their behaviors. This is best done using a graphical, rather
than numerical, display.
First we generate a Game of Life on a 100 100 board, and run it for 150 generations.
In[20]:= g LifeGame 100, 150 ;
372 An Introduction to Programming with Mathematica
ArrayPlot is well-suited for taking arrays of numbers and making plots, explicitly
specifying how to color alive and dead sites. This shows only the last frame from the game.
In[21]:= ArrayPlot Last g , ColorRules 0 Black, 1 Red ;
Below is the code to generate an array showing only every 25th iteration.
In[22]:= garray Map
ArrayPlot #, ColorRules 0 Black, 1 Red ,
DisplayFunction Identity &,
Table g i , i, 1, 150, 25
Out[22]= Graphics , Graphics , Graphics ,
Graphics , Graphics , Graphics
In[23]:= Show GraphicsArray garray
Out[23]= GraphicsArray
The following generates an animation consisting of 100 iterations of the Game of
Life on an initial 75 75 gameboard. We can not show the animation in a printed book, of
course, so we just indicate the input to evaluate.
In[24]:= AnimateLife lis_List :
Scan ArrayPlot #, ColorRules 0 Black, 1 Red &, lis
In[25]:= AnimateLife LifeGame 75, 100 ;
11 Examples and applications 373
Exercises
1. Define a new graphics function LifeGraphics that creates a raster array of color
values. Set it up so that it takes an option Colors defined to have default values for
1 and 0 and such that you can give it your own coloring scheme. Then you will
display Life games with Show[LifeGraphics[LifeGame[n,steps]]].
2. The Game of Life is most interesting to watch when persistent patterns, known as
life-forms, occur during the evolution process. One pattern that has been extensively
studied is known as the glider, which is defined by
glider x_, y_ :
x, y , x 1, y , x 2, y , x 2, y 1 , x 1, y 2
Modify the program for the Game of Life so that the lattice can be seeded with life
forms and observe the behavior of a glider (it should appear, disappear, and then
reappear in a shifted position every fifth generation). To better understand the use of
the periodic boundary conditions, note what happens when a glider pattern moves
beyond a border of the game board.
11.4 Implementing languages
Introduction to PDL
The Mathematica programming language is just one example of a computer language.
There are many, many others, including C and Fortran for general-purpose programming,
SQL for database queries, T
E
X and PostScript for typesetting, and on and on. The process-
ing of these languages shares some basic methods, which we will illustrate in this section by
implementing a mini Picture-Description Language, PDL.
PDL will be used to describe pictures consisting of simple shapes either contained in
or next to one another. An example of such a picture is shown in Figure 11.1; it is
described by the following picture specification.
374 An Introduction to Programming with Mathematica
square (5)
containing n/n (clear rectangle (5, 2)
containing w/w (circle (2))
containing c/c (circle (2))
containing e/e (circle (2)))
containing c/n (oval (3, 1)
connecting sw/ne (square (1))
connecting se/nw (square (1))
connecting s/c (circle (1)))
The picture in Figure 11.1 contains one large and two small squares, one rectangle
(but it is clear; that is, invisible), four circles and an oval. The rectangle is contained in
the square, and in turn contains three circles; the oval is contained in the square and has
the two small squares and a circle connected to it. The numbers in parentheses give the
sizes of the shapes, and the odd-looking notations like n/n and se/nw indicate how two
shapes are connected.
Figure 11.1: A picture produced by PDL
For example, the n/n notation on the second line says that the top (or north) of
the rectangle is positioned next to the top of the square that contains it; the se/nw nota-
tion on the second to last line indicates that the upper-left (northwest) point of the
square is placed next to the lower-right (southeast) point of the oval; on the last line, the
south point of the oval connects to the center of the circle.
We will write a function PDL that will take such a description (as a character string)
and convert it into Mathematica graphics primitives for display.
11 Examples and applications 375
Syntax
The first property all modern computer languages share is that their syntax can be for-
mally defined. The formal definition guides the implementation in a direct and simple way.
The formal definition is given as a context-free grammar, in which a set of rules, called
productions, are used to define both the allowable picture specifications and the syntactic
structure of those specifications. The formal syntax of PDL is given in Table 11.1.
In the PDL grammar, the names given in slanted or italic font are called variables. The
variables generate sets of strings; the legal picture specifications are all the strings gener-
ated by the variable picture. The items written in typewriter font appear literally in
specifications. Aside from integer (which, by definition, generates all the integers) and
direction (which, by definition, generates the strings n, e, s, w, c, ne, se, sw, nw), the
variables generate strings in the following way: starting with a variable, replace it by the
right-hand side of any rule in which it appears on the left-hand side; then continue to
replace variables by the right-hand sides of rules for those variables (or, in the case of
integer and direction, by any integer or direction) until a string without variables is obtained.
(When production 2 or 8 is applied, the variable just disappears.)
1. picture shape associations
2. associations
3. associations connection associations
4. associations containment associations
5. connection connectingdirection direction picture
6. containment containingdirection direction picture
7. shape color primitive size
8. color
9. color clear
10. primitive square
11. primitive circle
12. primitive oval
13. primitive rectangle
14. size integer size2
15. size2
16. size2 , integer
Table 11.1: Formal syntax of PDL
376 An Introduction to Programming with Mathematica
Consider the picture specification which produces the picture shown in Figure 11.2.
square (20) containing c/w (oval (9, 18))
Figure 11.2: A simpler picture produced by PDL.
It is generated from picture in this way (where we have indicated the number of the
production being used in each case).
picture
1
shape associations
7
color primitive size associations
8
primitive size associations
10
squaresize associations
14
square integer size2 associations
square 20size2 associations
15
square 20 associations
4
square 20 containment associations
11 Examples and applications 377
Parsing
A crucial observation is that the derivation of a string from a variable can be represented as
a tree, called a parse tree. The derivation above corresponds to the following tree.
picture 1
shape 7 associations 4
color 8 primitive 10 size 14 containment associations
square integer size2 15
20
Figure 11.3: A parse tree
Notice that there is no need to include the variable at each node; the production
number immediately determines the variable.
We will use the tree representation of the input or a very similar representation,
omitting uninteresting things like parentheses extensively. The goal of parsing is to
transform the sequence of characters in the input into a parse tree. Given that form, we can
do the real work: finding the location of each shape and generating the Mathematica
graphics primitives to draw the picture.
The parsing phase is divided into two steps, lexical analysis and parsing, and the
remainder of the processing is also divided into two steps, computing information about
each shape in the picture (especially, its location) and converting this information into
graphics primitives. Thus, the function PDL is given by
In[1]:= IPM3`PDL`
In[2]:= ShowPicture p_ : Show Graphics p , AspectRatio Automatic
In[3]:= PDL input_ :
ShowPicture ConvertShapes ComputeShapes Parse Lex input
378 An Introduction to Programming with Mathematica
For example, the following produces the graphic in Figure11.2.
In[4]:= PDL "square 20 containing c w oval 9, 18 " ;
Before delving into programming details, we will finish our brief users guide
begun earlier. As we have seen, shapes can be clear, in which case they are not drawn, or
regular, in which case they are drawn in black. Each shape has a size (one integer for
squares and circles, two for ovals and rectangles) with these meanings:
square : length of a side
circle : diameter
oval : horizontal and vertical diameters
rectangle : width and height
A shape can contain or be connected to any number of other shapes. (Since every
shape has an explicit size, a contained shape is not necessarily completely contained.)
The most complicated aspect of the language is determining where shapes go, depending
upon the points at which they are connected. Each shape has a center and eight compass
points. These are shown for each shape in Figure 11.4. When a shape is connected to or
contained in another shape, the two directions given in the connecting or containing
phrase match up. For example, Figure 11.5 shows the picture specified by square (4)
connected se/n (circle (2)).
11 Examples and applications 379
center
north
south
east west
northeast
southeast southwest
northwest
center
north
south
east west
northeast
southeast southwest
northwest
center
north
south
east west
northeast
southeast southwest
northwest
center
north
south
east west
northeast
southeast southwest
northwest
Figure 11.4: Compass points for the four types of shapes
In[5]:= PDL "square 4 connecting se n circle 2 " ;
The top (north) of the circle is next to the lower-right (southeast) corner of the
square. Similarly, Figure 11.6 shows an oval containing a rectangle.
In[6]:= PDL "oval 10,7 containing n n rectangle 1,3 " ;
The top of the oval touches the top of the rectangle. However, in both cases, the two
figures do not exactly touch; rather, a gap of size 0.1 is left between them. The difference
between connecting and containing is simply the direction of the gap.
380 An Introduction to Programming with Mathematica
picture
1
connecting d
1
d
2
picture
2
In the above code, direction d
2
of picture
2
is placed at a point determined by finding
the d
1
direction of picture
1
and then moving 0.1 units away from the center of picture
1
. If
connecting is replaced by containing, the correction is 0.1 units toward the center of
picture
1
.
Finally, the rules about correcting by 0.1 do not apply if either direction is c, for
center. If the connecting directions are d/c or c/d, whatever d is, it is not adjusted by 0.1
in either direction. (Thus, in this case it does not matter whether picture
2
is connected or
contained.) The reader is urged to try some examples using the code provided in the
IPM3`PDL` package before continuing.
Lexical analysis
To return to the programming of the PDL language processor, we will start with a discus-
sion of the syntactic analysis phase, consisting of lexical analysis (or lexing) and parsing. This
division is conventional and appears in virtually all language processors.
Lexing is the process of dividing up the input (a character string) into significant
syntactic units, called tokens. (Think of the entire picture specification as a sentence, the
characters of the input as the letters, and the tokens as the words; lexing groups the letters
into words, and parsing determines the syntactic structure of the sentence.) The function
Lex is given a string and produces a list of symbols and numbers.
In[7]:= example "square 20 containing c w oval 9, 18 ";
In[8]:= Lex example
Out[8]= square, lparen, 20, rparen, containing, center, slash, west,
lparen, oval, lparen, 9, comma, 18, rparen, rparen, eof
In[9]:= Map Head, %
Out[9]= Symbol, Symbol, Integer, Symbol, Symbol,
Symbol, Symbol, Symbol, Symbol, Symbol, Symbol,
Integer, Symbol, Integer, Symbol, Symbol, Symbol
In the lexed output, we have also replaced special characters like parentheses by
symbols and we have added a final symbol, eof (a traditional name meaning end of file).
Symbols are a little more convenient than strings for what we want to do. However,
their use requires that we introduce a new operator for comparing symbols that we have
11 Examples and applications 381
not needed until now, ===. The equality operator == works fine for numbers and strings,
and for lists of same, but not for symbols.
In[10]:= a, b a, b
Out[10]= True
In[11]:= a, b a, c
Out[11]= a, b a, c
Equal (==) can tell when two lists of symbols are identical, but not when they are
different. SameQ (===) compares symbols for identity.
In[12]:= a, b a, b
Out[12]= True
In[13]:= a, b a, c
Out[13]= False
All the code used by Lex is shown in the following Program Listing 11.1. The basic
process is: find the first sequence of characters that form a token, say t, recursively lex the
remaining characters, and join t to the result. Technicalities arise with the treatment of
numbers and the desire to ignore blanks.
Needs["IPM3`BaseConvert`"]
(* LEXICAL ANALYSIS *)
mainRules = {
{"(", y___} -> {lparen, y},
{")", y___} -> {rparen, y},
{",", y___} -> {comma, y},
{"/", y___} -> {slash, y},
{"c", "o", "n", "n", "e", "c", "t", "i", "n", "g", y___}
-> {connecting, y},
{"c", "o", "n", "t", "a", "i", "n", "i", "n", "g", y___}
-> {containing, y},
{"s", "q", "u", "a", "r", "e", y___} -> {square, y},
{"c", "i", "r", "c", "l", "e", y___} -> {circle, y},
{"o", "v", "a", "l", y___} -> {oval, y},
{"r", "e", "c", "t", "a", "n", "g", "l", "e", y___}
-> {rectangle, y},
{"c", "l", "e", "a", "r", y___} -> {clear, y},
382 An Introduction to Programming with Mathematica
{"n", "e", y___} -> {northeast, y},
{"s", "e", y___} -> {southeast, y},
{"s", "w", y___} -> {southwest, y},
{"n", "w", y___} -> {northwest, y},
{"n", y___} -> {north, y},
{"e", y___} -> {east, y},
{"s", y___} -> {south, y},
{"w", y___} -> {west, y},
{"c", y___} -> {center, y}
};
convertDigits[L_]:= Map[If[DigitQ[#], StringToInteger[#], #]&, L]
numberRule =
{{m_?NumberQ, n_?NumberQ, y___} -> {10m+n, y}};
removeBlanks = { {" ", y___} -> {y} };Lex[input_]:=
Module[{inp=FromCharacterCode/@ToCharacterCode[input]},
Lexaux[Join[convertDigits[inp],{eof}]//.removeBlanks]]
Lexaux[{eof}]:= {eof}
Lexaux[input_]:=
Module[{lexed = If[NumberQ[First[input]],
input //. numberRule, input /. mainRules]},
Join[{First[lexed]}, Lexaux[Rest[lexed] //. removeBlanks]]]
Program Listing 11.2: Code for lexing PDL
The first thing Lex does is explode the input string into a list of character codes.
As we saw in Section 7.5 in the chapter on recursion, we can do whatever we want with
that list; however, this would involve looking up a lot of character codes, so a simpler
approach is to convert each character code back to a string containing just that character.
So, in Lex, inp is a list containing each of the characters in input. convertDigits is
applied to change all digit characters to numbers (for example, the string "4" becomes the
number 4), using StringToInteger from the BaseConvert package. eof is added to
the end of the list. As final preparation before calling Lexaux, the transformation rule
removeBlanks is applied repeatedly (//.) to remove all leading blanks. Thus, the
argument to Lexaux is a list of one-character strings, numbers, and a final eof symbol,
with the first element nonblank. Lexaux repeatedly looks for characters that constitute a
token at the start of the list and replaces those characters by the token; it does this either
by a single use of a rule in mainRules or by repeated use of numberRule. It recursively
lexes the rest of the list and returns its result. We have already shown the result for our
running example.
11 Examples and applications 383
Parse takes the list of tokens and, if it is a legal picture specification, returns its
parse tree. The parser is the most interesting part of our language processor, as it shows a
strong link between the grammar specification (see the section above on PDL syntax) and
the program.
Our method here is called top-down, or recursive descent, parsing. The idea is to build
the parse tree by starting with a variable and letting the input string guide us in adding
nodes to the tree by telling us which production is applicable. For example, consider the
following list of tokens.
{square, lparen, 20, rparen, containing, center, slash, west,
lparen, oval, lparen, 9, comma, 18, rparen, rparen, eof}
Suppose we wish to create a parse tree for this string from the variable picture. The
only production from picture is production 1, so we could just add it to the tree without
even looking at the input. However, we would also like to report any syntactic errors as
soon as possible, so we will look at the first token in the input and see if it is legal at this
point. It so happens that every string derivable from picture must begin with one of the
words square, circle, oval, rectangle, or clear. If the first token is not one of
these, we can report an error; if it is, we add shape and associations to the tree. We continue
by trying to use the variable shape to match part of the list of tokens. Again, there is only
one production for shape (production 7), and, after checking, that square can be the first
token in a string derivable from shape, we add production 7 to the tree. The first part of the
right-hand side of production 7 is the variable color. We have a choice now, production 8
or 9, and we have to choose correctly. However, it is clear that square is not the first
token in a string derived using production 9, so it must be production 8 and we fill that in.
The next unfinished part of the tree is the node containing the variable primitive, which
has four productions.
A look at the input makes it immediately clear that only production 10 will work
here, so we fill it in. Continuing in this way, we eventually get the tree shown in Figure
11.3 and use up all the input. The top-down parsing process is illustrated in the following
series of parsing tree figures.
picture
Figure 11.5: Input: {square, lparen, 20, rparen,}
384 An Introduction to Programming with Mathematica
picture
shape associations
Figure 11.6: Input: {square, lparen, 20, rparen,}
picture
shape associations
color primitive size
Figure 11.7: Input: {square, lparen, 20, rparen,}
picture
shape associations
color primitive size
Figure 11.8: Input: {square, lparen, 20, rparen,}
picture
shape associations
color primitive size
square
Figure 11.9: Input: {lparen, 20, rparen,}
picture
shape associations
color primitive size
square integer size2
Figure 11.10: Input: {20, rparen,}
11 Examples and applications 385
picture
shape associations
color primitive size
square integer size2
20
Figure 11.11: Input: {rparen,}
Parse trees are represented as terms, using function symbols Prod1, Prod2, and so
on. Only significant parts of the tree are retained, so that, for example, Prod5 has only
three arguments, the two directions and the tree corresponding to the picture. Here is the
parse tree for our running example; compare it with the tree in Figure 11.3.
In[14]:= Parse Lex example
Out[14]= Prod1 Prod7 Prod8 , Prod10 , Prod14 20, Prod15 ,
Prod4 Prod6 center, west, Prod1 Prod7 Prod8 , Prod12 ,
Prod14 9, Prod16 18 , Prod2 , Prod2
The Parse function comes in three forms:
Parse[tokens_] returns the parse tree corresponding to the list of tokens. This is
the form we just used.
Parse[pns_, tokens_], where pns is a list of production numbers, derives a string
matching part of the tokens using one of the productions in pns. It returns a pair
containing the parse tree and the suffix of tokens not derived from the production.
Parse[pn_, tokens_], where pn is a production number, derives a prefix of tokens
from production pn and, like the previous form, returns a parse tree and a suffix of
tokens.
For the tokens in our example, we have:
In[15]:= Parse 7, Lex example
Out[15]= Prod7 Prod8 , Prod10 , Prod14 20, Prod15 ,
containing, center, slash, west, lparen,
oval, lparen, 9, comma, 18, rparen, rparen, eof
In other words, the first four tokens were derived using production 7; the parse tree
for production 7 and the remaining tokens are returned.
386 An Introduction to Programming with Mathematica
Parsing is quite simple. We are trying to generate a prefix of the input tokens from a
given variable. We call the second form of Parse, passing a list of all the productions for
that variable (the function prodsFor, shown in Figure 11.3, gives us the list), and it looks
at each one to see which might be usable given the first token (matches, also in Figure
11.3, tells whether a given production might apply for a given token). When it has found
the correct production, it calls the third form of Parse, which uses that production to
derive a prefix of the list of tokens.
In our example, Parse[1, Lex[ex]] first calls Parse[{7}, Lex[ex]], since
7 is the only production for shape, which in turn calls Parse[7, Lex[ex]], returning
the pair shown above. It then calls Parse[2,3,4, {containing, center,
slash,}], 2, 3, and 4 being all the productions from associations, and containing,
center, slash, being the tokens not matched by shape.
The first form of Parse, with one argument, is the one used by PDL. It starts the
parsing off by attempting to derive the list of tokens from picture; if successful, it discards
the {eof} and returns just the parse tree. The code for the three forms of Parse is given
in the program listing below.
(* PARSING *)
prodsFor[picture] := {1}
prodsFor[associations] := {2, 3, 4}
prodsFor[connection] := {5}
prodsFor[containment] := {6}
prodsFor[shape] := {7}
prodsFor[color] := {8, 9}
prodsFor[primitive] := {10, 11, 12, 13}
prodsFor[size] := {14}
prodsFor[size2] := {15, 16}
matches[1, t_] := MemberQ[{clear, square, circle, oval,
rectangle}, t]
matches[2, t_] := Not[MemberQ[{connecting, containing}, t]]
matches[3, t_] := MemberQ[{connecting}, t]
matches[4, t_] := MemberQ[{containing}, t]
matches[5, t_] := MemberQ[{connecting}, t]
matches[6, t_] := MemberQ[{containing}, t]
matches[7, t_] :=
MemberQ[{clear, square, circle, oval, rectangle}, t]
matches[8, t_] := MemberQ[{square, circle, oval, rectangle}, t]
Program Listing 11.3: Code for prodsFor, matches, and Parse
11 Examples and applications 387
matches[9, t_] := MemberQ[{clear}, t]
matches[10, t_] := MemberQ[{square}, t]matches[11, t_] :=
MemberQ[{circle}, t]
matches[12, t_] := MemberQ[{oval}, t]matches[13, t_] :=
MemberQ[{rectangle}, t]
matches[14, t_] := MemberQ[{lparen}, t]
matches[15, t_] := MemberQ[{rparen}, t]
matches[16, t_] := MemberQ[{comma}, t]
Parse[tokens_]:= First[Parse[prodsFor[picture], tokens]]
Parse[{}, x_]:=
(Print["Syntax error: remaining input is ",
Take[x, Min[Length[x], 10]], " ..."];
Abort[])
Parse[{pn_, pns___}, tokens_]:=
If[matches[pn, First[tokens]], (* if pn applies *)
Parse[pn, tokens], (* parse using it *)
Parse[{pns}, tokens]] (* else try other prod's *)
Parse[1, tokens_] :=
Module[{part1 = Parse[prodsFor[shape], tokens], part2},
part2 = Parse[prodsFor[associations], part1[[2]]];
{Prod1[part1[[1]], part2[[1]]], part2[[2]]}]
Parse[2, tokens_]:= {Prod2[], tokens}
Parse[3, tokens_] := Module[{
part1 = Parse[prodsFor[connection], tokens], part2},
part2 = Parse[prodsFor[associations], part1[[2]]];
{Prod3[part1[[1]], part2[[1]]], part2[[2]]}]
Parse[4, tokens_] :=
Module[{part1 = Parse[prodsFor[containment], tokens],
part2},
part2 = Parse[prodsFor[associations], part1[[2]]];
{Prod4[part1[[1]], part2[[1]]], part2[[2]]}]
Parse[5, tokens_] :=
Module[{part1 = Parse[prodsFor[picture], Drop[tokens, 5]]},
{Prod5[tokens[[2]], tokens[[4]], part1[[1]]],
Rest[part1[[2]]]}]
Program Listing 11.4: Code for Parse
388 An Introduction to Programming with Mathematica
Parse[6, tokens_]:=
Module[{part1 = Parse[prodsFor[picture], Drop[tokens, 5]]},
{Prod6[tokens[[2]], tokens[[4]], part1[[1]]],
Rest[part1[[2]]]}]
Parse[7, tokens_] :=
Module[{part1 = Parse[prodsFor[color], tokens], part2, part3},
part2 = Parse[prodsFor[primitive], part1[[2]]];
part3 = Parse[prodsFor[size], part2[[2]]];
{Prod7[part1[[1]], part2[[1]], part3[[1]]], part3[[2]]}]
Parse[8, tokens_]:= {Prod8[], tokens}
Parse[9, tokens_]:= {Prod9[], Rest[tokens]}
Parse[10, tokens_]:= {Prod10[], Rest[tokens]}
Parse[11, tokens_]:= {Prod11[], Rest[tokens]}
Parse[12, tokens_]:= {Prod12[], Rest[tokens]}
Parse[13, tokens_]:= {Prod13[], Rest[tokens]}
Parse[14, tokens_]:=
Module[{part1 = Parse[prodsFor[size2], Drop[tokens, 2]]},
{Prod14[tokens[[2]], part1[[1]]], part1[[2]]}]
Parse[15, tokens_]:= {Prod15[], Rest[tokens]}
Parse[16, tokens_]:= {Prod16[tokens[[2]]], Drop[tokens, 3]}
Program Listing 11.5: Code for Parse (cont.)
Computing shapes
With the parse tree in hand, the remaining processing is a fairly routine matter of tree
traversal, such as we used in Section 7.5. By computing the characteristics of each shape
its center, size, and compass points we can compute the characteristics of the shapes it
contains or is connected to. The coding has its occasional tricky moments, but is not
basically very difficult.
Recall that there are two functions, ComputeShapes and ConvertShapes, in this
part of the program. ComputeShapes does the tree traversal; ConvertShapes just
converts the list of shapes to a list of Mathematica graphics. We take them in order.
ComputeShapes traverses the parse tree and produces a list of shapes. The key
point here is exactly what we mean by shape. That is, how do we store the information
about shapes that we mentioned above (center, size, compass points)? The structure is
shown in the code at the top of Figure 11.3. A shape is represented by a seven-element list:
its center (a pair of numbers), the primitive shape (a symbol), the color (a production,
11 Examples and applications 389
either Prod8 for a normal shape or Prod9 for a clear one), the distance from the center to
the east compass point, the distance from the center to the north compass point, the angle
of the northeast compass point (in radians), and the distance from the center to the north-
east compass point. We have defined functions center, primitive, color, east,
north, neangle, and nedist to extract these components from a shape.
center[{c_, ___}] := c
primitive[{_, p_, ___}] := p
color[{_, _, c_, ___}] := c
east[{_, _, _, e_, ___}] := e
north[{_, _, _, _, n_, ___}] := n
neangle[{_, _, _, _, _, a_, ___}] := a
nedist[{_, _, _, _, _, _, d_}] := d
angle[s_, north] := Pi/2
angle[s_, south] := -Pi/2
angle[s_, east] := 0
angle[s_, west] := Pi
angle[s_, northeast] := neangle[s]
angle[s_, southeast] := -neangle[s]
angle[s_, southwest] := Pi+neangle[s]
angle[s_, northwest] := Pi-neangle[s]
dist[s_, north] := north[s]
dist[s_, south] := north[s]
dist[s_, east] := east[s]
dist[s_, west] := east[s]
dist[s_, northeast] := nedist[s]
dist[s_, southeast] := nedist[s]
dist[s_, southwest] := nedist[s]
dist[s_, northwest] := nedist[s]
pointOf[s_, d_, delta_] :=
center[s] + vector[angle[s, d], dist[s, d] + delta]
computeCenter[s_, p_, center] := p
computeCenter[s_, p_, d_] :=
p + vector[Pi+angle[s, d], dist[s, d]]
vector[theta_, r_] := {r Cos[theta], r Sin[theta]}
Program Listing 11.6: Code for dealing with shapes and points
We will need to compute points using angles and distances from the center of a
shape. It is convenient to define the following functions, given above in the program listing.
390 An Introduction to Programming with Mathematica
1. angle[shape, direction] computes the angle from the center of shape to the given
compass point. East is always 0, north always 2, and so on, but the intermediate
points depend upon the dimensions of the shape (at least for ovals and rectangles).
2. dist[shape,direction] computes the distance from the center of a shape to the
given compass point. Distances for intermediate points are all the same as the
northeast distance.
3. pointOf[shape,direction,delta] computes the compass point given by direction
for shape, adjusted by delta. A positive delta moves the point away from the center
of the shape, a negative delta towards it.
4. computeCenter[shape,point,direction], where shape does not yet have a center,
though it has all its other information, computes its center, given that the com-
pass point named by direction is to be at point. For example, if s is a square with
sides of length 10, computeCenter[s,{4, 2},north] will return {4,-3};
if the square is centered at {4,-3}, its north point will be at {4,2}.
The tree traversal is initiated by a call to the one-argument form of Compute
Shapes, which is called with a Prod1 tree. It calls the three-argument form of Compute
Shapes, which returns a list of shapes. N is applied to the list to evaluate all numerical
formulas and all the clear shapes (Prod9) are removed. For our running example (Figure
11.2), we see the result in this session:
In[16]:= ComputeShapes Parse Lex example
Out[16]= 0., 0. , square, Prod8 , 10., 10., 0.785398, 14.1421 ,
4.5, 0. , oval, Prod8 , 4.5, 9., 1.10715, 7.11512
The main shape, centered at 0, 0 , is a 20 20 square (the 10s being the distance
from the center to the side and the top). The 9 18 oval is centered at 4.5, 0 .
The three-argument form of ComputeShapes takes a tree given in the form
Prod1[shape, associations] and computes the shape of shape and all the shapes in
associations, returning a list. The second and third arguments are a point p and a direction d.
First, shape is drawn with direction d at point p. This is done by calling computeShape;
shapeInfo computes all the location-independent information, which is everything but
the center, and the latter is filled in by a call to computeCenter, discussed above. Then
the shapes in associations are drawn in positions computed with respect to shape. This is
accomplished by calling computeAssociatedShapes, passing associations as the first
argument and the shape computed for shape as the second. The associations parse tree
(Prod2, Prod3, or Prod4) is traversed, and the shapes it contains are computed with
respect to that second argument. The auxiliary function compute
11 Examples and applications 391
Point[shape,dir
1
,dir
2
,relation] computes the meeting point of shape with whatever
shape it contains or connects to, given that dir
1
of shape is to meet dir
2
of the contained
shape. The computation also depends upon whether the shape is contained or connected,
as given by relation.
separation = .1;
ComputeShapes[tree_]:=
Select[N[computeShapes[tree, {0, 0}, center]],
(color[#] =!= Prod9[])&]
computeShapes[Prod1[sh_, assoc_], p_, d_]:=
Module[{s = computeShape[sh, p, d], as},
as = computeAssociatedShapes[assoc, s];
Join[{s}, as]]
computeAssociatedShapes[Prod2[], _]:= {}
computeAssociatedShapes[Prod3[Prod5[d1_,d2_,pic_],assoc_],s_]:=
Module[{p = computePoint[s, d1, d2, connecting], ss},
ss = computeShapes[pic, p, d2];
Join[ss, computeAssociatedShapes[assoc, s]]]
computeAssociatedShapes[Prod4[Prod6[d1_,d2_,pic_],assoc_],s_]:=
Module[{p = computePoint[s, d1, d2, containing], ss},
ss = computeShapes[pic, p, d2];
Join[ss, computeAssociatedShapes[assoc, s]]]
computePoint[s_, d1_, d2_, relation_]:=
Which[d1===center, center[s],
d2===center, pointOf[s, d1, 0],
relation===connecting, pointOf[s, d1, separation],
True, pointOf[s, d1, -separation]]
computeShape[s_, p_, d_]:= Module[{si = shapeInfo[s]},
Join[{computeCenter[si, p, d]}, Rest[si]]]
shapeInfo[Prod7[color_, Prod10[], Prod14[i_, _]]]:=
{0, square, color, i/2, i/2, Pi/4, i/Sqrt[2]}
shapeInfo[Prod7[color_, Prod11[], Prod14[i_, _]]]:=
{0, circle, color, i/2, i/2, Pi/4, i/2}
ovalNE[a_, b_, theta_]:=
a b Sqrt[(1 + Tan[theta]^2)/(a^2 + b^2 Tan[theta]^2)]
392 An Introduction to Programming with Mathematica
shapeInfo[Prod7[color_, Prod12[], Prod14[l_, Prod16[h_]]]] :=
{0, oval, color, l/2, h/2, ArcTan[h/l],
ovalNE[h/2, l/2, ArcTan[h/l]]}
shapeInfo[Prod7[color_, Prod13[], Prod14[l_, Prod16[h_]]]] :=
{0, rectangle, color, l/2, h/2, ArcTan[h/l], Sqrt[h^2 + l^2]/2}
Program Listing 11.7: Computing shapes
Finally, the list of shapes is converted to a list of Mathematica graphics by mapping
convertShape over the list. The Mathematica graphics primitives are well matched to
our representation of shapes, making convertShape easy to write. Here is the final
output of our example.
In[17]:= ConvertShapes ComputeShapes Parse Lex example
Out[17]= Line 10., 10. , 10., 10. , 10., 10. , 10., 10. ,
10., 10. , Circle 4.5, 0. , 4.5, 9.
In[18]:= Show Graphics % ;
makeRectangle[p_, l_, h_] :=
Line[{p, p+{0,h}, p+{l,h}, p+{l,0}, p}]
convertShape[s_] /; MemberQ[{square, rectangle}, s[[2]]] :=
makeRectangle[pointOf[s, southwest, 0], 2 east[s], 2 north[s]]
convertShape[s_] /; MemberQ[{circle, rectangle}, s[[2]]] :=
Circle[center[s], north[s]]
convertShape[s_] /; s[[2]] === oval :=
Circle[center[s], {east[s], north[s]}]
ConvertShapes[ss_] := Map[convertShape, ss]
Program Listing 11.8: Converting shapes to Mathematica graphics objects
11 Examples and applications 393
12 Writing packages
Packages are text files that contain Mathematica commands. They are designed to
make it easy to distribute your programs to others, but they also provide a mechanism
for you to write programs that integrate with Mathematica in a seamless manner. In
this chapter we will discuss the organization and creation of packages including a
discussion of contexts, which are a mechanism for organizing new names and symbols
in your Mathematica sessions.
12.1 Introduction
When you begin a Mathematica session, the built-in functions are immediately available for
you to use. There are, however, many more functions that you can access that reside in
files supplied with Mathematica. In principle, the only difference between those files and
the ones you create is that those were written by professional programmers. There is
another difference: the definitions in those files are placed in special structures called
packages. Indeed, these files themselves are often called packages instead of files.
Packages are a name localizing construct, analogous to Module, but for entire files of
definitions. Their purpose is to allow the programmer to define a collection of functions
for export. These exported functions are for the users of the package to work with and are
often referred to as public functions. Other functions, those that are not for export, are
auxiliary, or private functions, and are not intended to be accessible to users.
In this chapter, you will learn how to write your own packages. Much of the chapter
is devoted to an explanation of a more primitive notion, that of contexts, which is a prerequi-
site to understanding packages. We then describe packages and give a simple example,
showing the standard and accepted style for writing them. We will also distinguish
between functions for export and auxiliary functions that users of your package need not
be concerned with.
12.2 Using packages
Mathematica packages have been written for a great variety of problem domains. Many are
provided with each version of Mathematica and are referred to as the Standard Packages.
Their documentation is available in the Help Browser. Below, we list some examples of
some of the standard Mathematica packages. Note that package names always end with a
back quote (`), and often have back quotes within them as well. We will discuss the
meaning of this back quote shortly.
Calculus`VectorAnalysis`: This package provides a variety of variables and
functions for doing calculus in various three-dimensional coordinate systems; for
example, SetCoordinates to set the coordinate system (Cartesian, polar, etc.);
CrossProduct to compute cross products; Curl to give the curl of a vector field.
Graphics`MultipleListPlot`: Provides functions for superimposing several
plots on the same graphic. MultipleListPlot is the main function in this
package. It plots lists of data as separate plots on the same axes. Also provided is
MakeSymbol which creates symbols to use in labeling the separate plots, plus a
number of functions for specifying symbols.
Loading packages
Once you know which package you want to use, you can load it in one of two ways. For
example, to load the package Calculus`VectorAnalysis`, you can use either Get or
Needs.
<<Calculus`VectorAnalysis` will read the file and evaluate each expression
and definition as if it had been typed in. Actually, the argument of << is a string, but
the quotation marks can be omitted. <<package` is shorthand for Get["package`"].
Needs["Calculus`VectorAnalysis`"] will read the package, just like <<,
but only if it has not already been read.
Here is an example of using the Calculus`VectorAnalysis` package.
In[1]:= Needs "Calculus`VectorAnalysis`"
396 An Introduction to Programming with Mathematica
Here is the usage message for one of the functions defined in the package.
In[2]:= ?CrossProduct
CrossProduct v1, v2 gives the cross product
sometimes called vector product of the two vectors v1,
v2 in three space in the default coordinate system.
CrossProduct v1, v2, coordsys gives the cross product
of v1 and v2 in the coordinate system coordsys. More
This computes the cross product of two symbolic vectors using the CrossProduct
function defined in Calculus`VectorAnalysis`.
In[3]:= CrossProduct x
1
, y
1
, z
1
, x
2
, y
2
, z
2
Out[3]= y
2
z
1
y
1
z
2
, x
2
z
1
x
1
z
2
, x
2
y
1
x
1
y
2
Finding out what is in a package
To use the Mathematica packages, you need to know what they provide. In fact, program-
mers find that even remembering what is in their own packages is not easy, if they have not
looked at them for a while. If you know the name of the package and you want to know
what it defines, first load it, using <<package` or Needs["package`"].
In[4]:= Needs "DiscreteMath`ComputationalGeometry`"
Now you can get a list of hyperlinks of the functions defined in this package as
follows.
In[5]:= ?DiscreteMath`ComputationalGeometry`*
DiscreteMath`ComputationalGeometry`
AllPoints NearestNeighbor
BoundedDiagram PlanarGraphPlot
ConvexHull Ray
DelaunayTriangulation TileAreas
DelaunayTriangulationQ TriangularSurfacePlot
DiagramPlot TrimPoints
Hull VoronoiDiagram
LabelPoints
Clicking any of the above links will display the usage message associated with that
function.
12 Writing packages 397
You can also display a list of the names defined in the package using Names.
In[6]:= Names "DiscreteMath`ComputationalGeometry` "
Out[6]= AllPoints, BoundedDiagram, ConvexHull, DelaunayTriangulation,
DelaunayTriangulationQ, DiagramPlot, Hull, LabelPoints,
NearestNeighbor, PlanarGraphPlot, Ray, TileAreas,
TriangularSurfacePlot, TrimPoints, VoronoiDiagram
Once you have loaded the package you can use ? to get the usage message for any of
those names.
In[7]:= ?DelaunayTriangulation
DelaunayTriangulation x1,y1 , x2,y2 ,..., xn,
yn yields the planar Delaunay triangulation
of the points. The triangulation is represented as
a vertex adjacency list, one entry for each unique
point in the original coordinate list indicating the
adjacent vertices in counterclockwise order. More
If, on the other hand, you forget the name of the package, you can easily browse
through the Help Browser which lists all packages, names, and usage messages of any
functions defined in these packages. Alternatively, you can find out where the directory of
packages is stored on your system, and browse through it in your file system.
Avoiding name collisions
Sometimes, you will read in a package that defines a function f whose name you have
already mentioned in your current session. It is very common, for example, to forget to
load a package before calling one of its functions. By simply mentioning the functions
name you create a symbol in the current context. Then, if you try to make a call to f,
Mathematica will assume you are talking about the f in the current context rather than the
one defined in the package.
For example, suppose we attempted to use a function RandomPermutation that
we mistakenly believed was a built-in function.
In[8]:= RandomPermutation 4
Out[8]= RandomPermutation 4
After a little searching in the Help Browser we discover that RandomPermutation
is not a built-in function, but is in fact, defined in the package DiscreteMath`Combina
torica`. So let us try to load the package.
398 An Introduction to Programming with Mathematica
In[9]:= DiscreteMath`Combinatorica`
RandomPermutation::shdw :
Symbol RandomPermutation appears in multiple contexts
DiscreteMath`Combinatorica`, Global` ; definitions
in context DiscreteMath`Combinatorica` may
shadow or be shadowed by other definitions. More
If you try to use the RandomPermutation function defined in the Discrete
Math`Combinatorica` package, you will not be able to do so in the usual way as its
definition is shadowed by the RandomPermutation function that was placed in the
Global` context when we first tried to use it.
In[10]:= RandomPermutation 4
Out[10]= RandomPermutation 4
You can still use the RandomPermutation function from the Combinatorica
package but you have to explicitly use its full context.
In[11]:= DiscreteMath`Combinatorica`RandomPermutation 3
Out[11]= 2, 1, 3
If, however, you want to be able to call DiscreteMath`Combinatorica`Ran
domPermutation by its short name, and forget the RandomPermutation you defined
in the Global` context, use the function Remove.
In[12]:= Remove RandomPermutation
This will make it seem that you had never mentioned the name Global`Random
Permutation at all as it completely removes the symbol RandomPermutation from
the Global` context. Now you can use the short name for the RandomPermutation
function from the DiscreteMath`Combinatorica` package.
In[13]:= RandomPermutation 3
Out[13]= 3, 1, 2
Note that evaluating Clear[RandomPermutation] is not enough; that would
clear values associated with any assignments attached to RandomPermutation, but it
would not un-mention the symbol itself; in other words, Clear[symbol] clears out the
right-hand side of any definition associated with symbol, but it does not remove symbol from
the context within which it was first created.
There is a way to minimize this problem, if you have certain packages that you often
use.
12 Writing packages 399
DeclarePackage["package`", {"name
1
", "name
2
", }]
DeclarePackage tells Mathematica that whenever you use one of the names name
1
,
name
2
,, it should load package (if it has not already been loaded). It is a good practice to
make a file containing a DeclarePackage for each package you frequently use, listing all
the names of functions you use from that package. For example, if that file is called
mypackage.m then, whenever you start a Mathematica session, enter <<mypackage.m
as your first input. Alternatively, you could put mypackage.m in one of the init.m files
and Mathematica will automatically load it whenever you start a session. There are several
locations where kernel init.m files can be found:
In[14]:= Map ToFileName,
$BaseDirectory, "Autoload", "_", "Kernel", "init.m" ,
$UserBaseDirectory, "Autoload", "_", "Kernel", "init.m" ,
$InstallationDirectory,
"Configuration", "Kernel", "init.m" ,
$InstallationDirectory, "AddOns",
"Autoload", "_", "Kernel", "init.m"
Out[14]= C:\Documents and Settings\All Users\Application
Data\Mathematica\Autoload\ \Kernel\init.m\,
C:\Documents and Settings\Paul Wellin\Application
Data\Mathematica\Autoload\ \Kernel\init.m\,
C:\Program Files\Wolfram Research\Mathematica
\5.1\Configuration\Kernel\init.m\,
C:\Program Files\Wolfram Research\Mathematica
\5.1\AddOns\Autoload\ \Kernel\init.m\
The first two locations given above are the preferred directories to place your
init.m files. The last two are dependent upon the version of Mathematica and hence will
need to be updated or moved when you upgrade to a newer version of Mathematica.
Lastly, you can also put your init.m in a Kernel directory in a package directory in
any of the Applications directories. For example, if you have a directory named MathApps
that lives inside one of the Applications directories, then put a Kernel directory inside
MathApps and an init.m inside that Kernel directory. Your packages will live inside
MathApps. So loading a package (<<MathApps`mypackage`) will automatically load
the init.m inside the MathApps/Kernel directory.
400 An Introduction to Programming with Mathematica
12.3 Contexts
Every symbol you use in a computation in Mathematica has a full name consisting of the
symbol preceded by the context in which the name was first mentioned. The context is a
means for organizing symbols. You can think of the context like a namespace different
symbols are in different contexts just like different files on your computer live in different
directories.
When you first start your session, the current context is Global` (again note the back
quote), and any symbol symbol you mention now has full name Global`symbol. A symbol
can be given with its full name or in its regular, short form.
Here is a function created in the Global` context.
In[1]:= f x_ : x 1
In[2]:= Context f
Out[2]= Global`
We can use the function with its full name.
In[3]:= Global`f 3
Out[3]= 4
But, of course, it is much more convenient to use the regular, short form.
In[4]:= f 3
Out[4]= 4
Mathematica first searches the current context for definitions associated with any
symbols; by default, this is the Global` context. To see a list of the contexts that Mathe-
matica uses to search for symbols, use $ContextPath.
In[5]:= $ContextPath
Out[5]= Global`, System`
As we saw above, symbols you define when your session begins have context Glo
bal`. Built-in functions have context System`.
In[6]:= Map Context, Integrate, Plot, , List
Out[6]= System`, System`, System`, System`
12 Writing packages 401
You can tell Mathematica to use a different context for any new symbols you mention
by using the function Begin.
In[7]:= Begin "ContextA`"
Out[7]= ContextA`
In[8]:= g x_ : x 2
We can use the full name for g:
In[9]:= ContextA`g 3
Out[9]= 5
Or, since we are currently in the ContextA` context, we can use the short name.
In[10]:= g 3
Out[10]= 5
Here is the current context.
In[11]:= $Context
Out[11]= ContextA`
In this new context, the name g is an abbreviation for ContextA`g.
In[12]:= Map g, 5, 7, 9
Out[12]= 7, 9, 11
Note that we can still refer to f, even though it was not defined in this context.
In[13]:= Map Global`f, 5, 7, 9
Out[13]= 6, 8, 10
In[14]:= Map f, 5, 7, 9
Out[14]= 6, 8, 10
After exiting the context using the End function, we may define a different g, having
context Global`.
In[15]:= End
Out[15]= ContextA`
In[16]:= g x_ : x 3
In[17]:= g 3
Out[17]= 6
402 An Introduction to Programming with Mathematica
We now have two definitions of g, or, rather, one definition of Global`g and one
of ContextA`g. Since our current context is Global`, when we just say g we get
Global`g; but we can still refer to ContextA`g by its full name.
In[18]:= g 3
Out[18]= 6
In[19]:= ContextA`g 3
Out[19]= 5
The question arises: when you enter a symbol symbol, how does Mathematica decide
which version of symbol to use? And how can you tell which one it has chosen?
To answer the second question first: the function Context gives the context of a
symbol.
In[20]:= Context g
Out[20]= Global`
In[21]:= Context Map
Out[21]= System`
In[22]:= Context ContextA`g
Out[22]= ContextA`
You can also use ?.
In[23]:= ?g
Global`g
g x_ : x 3
How, then, does Mathematica decide which definition to use? It maintains two
variables, $Context and $ContextPath. $Context contains a context (that is, a string
giving the name of a context), which is the current context, and $ContextPath contains
a list of contexts. Mathematica looks in $Context first, then in the contexts in $Context
Path in the order in which they appear there; if it does not find the symbol at all, then it
creates it in context $Context. Of course, none of this applies if you give the symbols
full name.
In[24]:= $Context
Out[24]= Global`
12 Writing packages 403
In[25]:= $ContextPath
Out[25]= Global`, System`
In[26]:= Begin "ContextA`"
Out[26]= ContextA`
In[27]:= $Context
Out[27]= ContextA`
In[28]:= $ContextPath
Out[28]= Global`, System`
In[29]:= End
Out[29]= ContextA`
In[30]:= $Context, $ContextPath
Out[30]= Global`, Global`, System`
So the effect of entering a new context using Begin is simply to change the value of
$Context; End[] changes it back. In either case, $ContextPath is not changed.
One final point about contexts: contexts can be nested within contexts. That is, you
can have context names like A`B`C`. To enter contexts like this, do the following.
In[31]:= Begin["A`"] (* enter context A` *)
Out[31]= A`
In[32]:= Begin["`B`"] (* enter context A`B` *)
Out[32]= A`B`
In[33]:= Begin["`C`"] (* enter context A`B`C` *)
Out[33]= A`B`C`
In[34]:= End[] (* back in context A`B` *)
Out[34]= A`B`C`
In[35]:= End[] (* back in context A` *)
Out[35]= A`B`
404 An Introduction to Programming with Mathematica
In[36]:= End[] (* back in context Global` *)
Out[36]= A`
Note the back quote before the context name in the second and third Begin. This is
used to indicate that the new context should be a sub-context of the current context. We
could have also indicated this as follows:
In[37]:= Begin "A`"
Out[37]= A`
In[38]:= Begin "A`B`"
Out[38]= A`B`
In[39]:= Begin "A`B`C`"
Out[39]= A`B`C`
Nested contexts are a way of managing the multiplicity of contexts. You will have
noticed how the names of the standard packages we discussed earlier look just like nested
contexts. In fact, package names are contexts. Mathematica organizes the standard packages
into about ten major contexts (for example, Calculus` and Graphics`), each with
about ten nested contexts; it is just a way of keeping things organized. Most readers will
recognize this as the idea behind hierarchical file systems. In fact, when you load a package
using Needs or <<, Mathematica translates the package name directly into a path name in
the hierarchical file system on your computer.
For example, you can load the package mypackage.m that lives in a directory
MathApps as follows:
Windows <<MathApps\mypackage.m
Unix/Linux/OS X <<MathApps/mypackage.m
Macintosh Classic <<MathApps:mypackage.m
But since Mathematica provides a system-independent means of loading packages,
you can simply use Get with the following syntax and Mathematica will automatically
translate this into a path name appropriate for your computer.
MathApps`mypackage`
12 Writing packages 405
Summary
Any name mentioned in a Mathematica session has a full name, containing a context
and the short name.
When using a name, you may give its full name. If you choose not to (as is custom-
ary), Mathematica will decide what the full name is; that is, what the context of the
name is.
Here is how Mathematica decides on the context:
First, it looks in the context given by the variable $Context.
Next, it looks in all the contexts given in the variable $ContextPath, in the
order in which they appear there.
If those searches do not succeed, Mathematica assumes this is the first mention of
the name, and so gives it the context $Context.
Begin["context`"] and End[] alter the value of $Context (but do not affect
$ContextPath). Specifically, Begin["context`"] sets $Context to context`, and
End[] restores it to its prior value before the Begin.
As of now, these functions are the only ways we know to alter the contents of these
two variables. In the next section, we will see two other functions that change them in a
subtly, but crucially different way.
12.4 The elements of packages
Packages allow you to create an organized collection of function definitions and values,
while avoiding collisions with any other definitions of those names. For example, if you
load a package that defines functions f and g, and the definition of g contains a call to f,
then g should always work that is, call the f defined in the package even if you have
defined f separately in your session (in the Global` context). Furthermore, packages can
define their own auxiliary (or private) functions and constants that the user, or client, of
the package will not ordinarily see at all.
406 An Introduction to Programming with Mathematica
All this is achieved using contexts, with two new functions:
BeginPackage["package`"] sets $Context to package`, and $ContextPath to
{package`, System`}.
EndPackage[] resets both variables to their values prior to the evaluation of
BeginPackage[], and then prepends package` to $ContextPath.
Thus, if you are in a Mathematica session, with current context Global`, and you
read in a file containing:
BeginPackage["P`"]
f[x_] :=
g[y_] :=
EndPackage[]
then after it is read, the functions f and g, with full names P`f and P`g, will be defined,
and the context P` will be in $ContextPath. If you do not have any other definitions of
f, you can refer to it as just f; if you do, then use P`f; and similarly for the function g.
The precise definition of BeginPackage[package`] is important as it changes
$ContextPath to {package`,System`}. Thus, all the names defined in the package
will have context package`. In our example above, the f and g in the package can be
referred to as P`f and P`g, regardless of any other definitions you may have given for
them.
It is important to realize, too, that Mathematica determines the full name of any name
when it reads it in. Thus, if g calls f, then the occurrence of f in the body of g becomes
P`f when the package is loaded. g will always call this f, even if there is a different f
defined in the context in which the call to g is made.
The BeginPackage function can be given multiple arguments. The second and
subsequent arguments are the names of other packages that this one uses. They are treated
as if they were arguments to the Needs function; that is, they are loaded if they have not
already been. Furthermore, they are included in $ContextPath during the loading of this
package, so its functions can refer to their functions by their short names.
Summary
BeginPackage["package`"] sets $Context to package`, and $ContextPath
to {package`, System`}, so that any names subsequently mentioned, other than
the names of built-in functions and constants, are defined in context package`.
12 Writing packages 407
EndPackage[] resets $Context and $ContextPath to their prior values,
except that package` is added to the front of $ContextPath.
12.5 Writing your own packages
The RandomWalks package
In this section, we list the full RandomWalks.m package, elements of which were devel-
oped in earlier chapters. We will add several important user interface elements, such as
expressions for options and usage statements. The full package is included in the IPM3
archive as indicated in the Preface.
BeginPackage
First, we set the value of Context`, which causes $ContextPath to be set to {IPM3`
RandomWalks`, System`}.
In[1]:= BeginPackage "IPM3`RandomWalks`"
Out[1]= IPM3`RandomWalks`
Importing other packages
You could import a package by using an optional argument to BeginPackage. In that
case, you would have BeginPackage["IPM3`RandomWalks`",{Graphics`Arg
Colors`,Graphics`Polyhedra`"}] above. The argument against this approach is
that the two packages Graphics`ArgColors` and Graphics`Polyhedra` will be
left on the search path after the RandomWalks` package is read in. It is considered poor
programming style to alter the users environment by simply reading in a package at least
you should try to alter it as little as possible. There is another method of loading a package
within a package, and that is to call Needs after the call to BeginPackage. Using this
mechanism, the Graphics`ArgColors` context will not remain on the context path
after the RandomWalks package is read in.
In[2]:= Needs "Graphics`ArgColors`"
In[3]:= Needs "Graphics`Polyhedra`"
408 An Introduction to Programming with Mathematica
Usage statements
Defining usage messages for the functions in your packages creates symbols for the func-
tions in the current context. Each of the functions for which you define a usage message
will then be exported for public use; that is, those functions are visible and usable immedi-
ately after loading the package. This is in distinction to any functions that are defined in
your package for which you do not have usage messages (or, more precisely, for which you
have not explicitly exported by mentioning that symbol before the Begin statement).
Those functions will be private, unavailable for the user of your package to access.
Making your functions behave much like the built-in functions will make it easier for
users of your packages, since they will expect usage messages and general functionality
similar to that of Mathematicas functions. It is also a good way for you to document your
programs. We would go so far as to suggest that you consider writing your usage messages
before you write the function definitions in Mathematica. This will help you to clearly
understand what it is you want your functions to do.
In[4]:= RandomWalk::"usage"
"RandomWalk n generates an n step walk in two dimensions.
The default behavior gives a lattice walk with steps
in one of the four compass directions. The option
LatticeWalk takes values True or False. The value
of the option Dimensions can be any of 1, 2, or 3.";
In[5]:= LatticeWalk::"usage"
"LatticeWalk val is an option to RandomWalk
that determines whether the random walk will
be a lattice walk or an off lattice walk.
Possible values are True and False.";
In[6]:= ShowWalk::"usage"
"ShowWalk walk displays a one, two, or three dimensional
random walk connecting each site with a line. Graphics
options can be passed to ShowWalk. E.g., ShowWalk walk,
Background GrayLevel 0 to produce a black background.";
In[7]:= AnimateWalk::"usage"
"AnimateWalk walk, opts creates an animation
of a two dimensional random walk. A red ball
will be seen to move to the current position in
the walk to aid in visualizing the animation.";
12 Writing packages 409
Warning messages
In[8]:= RandomWalk::rwn "Argument `1` is not a positive integer.";
Options
In[9]:= Options RandomWalk LatticeWalk True, Dimension 2
Out[9]= LatticeWalk True, Dimension 2
Begin private context
The Begin command changes the current context without affecting the context path. By
starting the argument `Private` with a context mark `, we change to a subcontext of
the current context. This new subcontext is IPM3`RandomWalks`Private`.
In[10]:= Begin "`Private`"
Out[10]= IPM3`RandomWalks`Private`
The function definitions
In[11]:= walk1D n_ : NestList # 1
Random Integer
&, 0, n
In[12]:= walk1DOffLattice n_ :
FoldList Plus, 0, Table Random Real, 1, 1 , n
In[13]:= walk2D n_ :
Module NSEW 0, 1 , 1, 0 , 0, 1 , 1, 0 ,
FoldList Plus, 0, 0 ,
NSEW Table Random Integer, 1, 4 , n
In[14]:= walk2DOffLattice n_ :
FoldList Plus, 0, 0 ,
Map Cos # , Sin # &, Table Random Real, 0, 2 , n
In[15]:= walk3D n_ : Module NSEW3 2 Vertices Cube , FoldList
Plus, 0, 0, 0 , NSEW3 Table Random Integer, 1, 8 , n
In[16]:= walk3DOffLattice n_ :
FoldList Plus, 0, 0, 0 , Map Cos # , Sin # ,
#
2
&,
Table Random Real, 2 , 2 , n
410 An Introduction to Programming with Mathematica
In[17]:= RandomWalk n_, opts___?OptionQ : Module dim, latticeQ ,
If Not IntegerQ n && n 0 , Message RandomWalk::rwn, n ,
latticeQ, dim LatticeWalk, Dimension .
Flatten opts, Options RandomWalk ;
Which
dim 1, If latticeQ, walk1D n , walk1DOffLattice n ,
dim 2, If latticeQ, walk2D n , walk2DOffLattice n ,
dim 3, If latticeQ, walk3D n , walk3DOffLattice n
In[18]:= ShowWalk coords_, opts___ :
Which
Length Dimensions coords 1,
ListPlot coords, opts, PlotJoined True ,
Dimensions coords 2 2,
Show Graphics Line coords , opts,
AspectRatio Automatic , Dimensions coords 2 3,
Show Graphics3D Line coords , opts,
AspectRatio Automatic
In[19]:= AnimateWalk coords_, opts___ :
Scan Show Graphics RGBColor 1, 0, 0 , PointSize 0.025 ,
Point coords #1 , Line Take coords, #1 ,
opts, AspectRatio Automatic, PlotRange
Min #1 0.2, Max #1 0.2 & Transpose coords &,
Range 2, Length coords
End private context
The End[] command closes the Begin[] and puts us back in the context Random
Walks`. Any symbols that were defined in the subcontext IPM3`RandomWalks`Pri
vate` can no longer be accessed.
In[20]:= End
Out[20]= IPM3`RandomWalks`Private`
EndPackage
The EndPackage[] command puts us back in the context we were in prior to the
BeginPackage[] command.
In[21]:= EndPackage
12 Writing packages 411
Examples
Starting with a new session, and making sure that the RandomWalks package is in a
directory/folder where Mathematica can find it, this loads the package.
In[22]:= Quit
In[1]:= IPM3`RandomWalks`
Here is the usage message for the RandomWalk function.
In[2]:= ?RandomWalk
RandomWalk n generates an n step walk in two
dimensions. The default behavior gives a lattice walk
with steps in one of the four compass directions. The
option LatticeWalk takes values True or False. The
value of the option Dimensions can be any of 1, 2, or 3.
This gives a random walk of length 10 in two dimensions.
In[3]:= RandomWalk 10, Dimension 2
Out[3]= 0, 0 , 0, 1 , 0, 0 , 0, 1 , 0, 2 ,
1, 2 , 1, 1 , 1, 0 , 1, 1 , 0, 1 , 0, 2
This shows a 250-step off-lattice random walk using the default of two dimensions.
In[4]:= ShowWalk RandomWalk 250, LatticeWalk False ;
A 500-step two-dimensional random walk with some graphics options.
In[5]:= ShowWalk RandomWalk 500 , Frame True ;
20 15 10 5 0 5
10
5
0
5
10
412 An Introduction to Programming with Mathematica
A 100 step off-lattice random walk in three dimensions.
In[6]:= walk3
ShowWalk RandomWalk 10
3
, Dimension 3, LatticeWalk False ;
Using a transformation rule, it is straightforward to change the coordinates of each
line to a gray point.
In[7]:= Show walk3 . Line x_ : GrayLevel .5 , Map Point, x ;
Finally, we should check that RandomWalk does the right thing when passed a bad
argument.
In[8]:= RandomWalk 5
RandomWalk::rwn : Argument 5 is not a positive integer.
Although we have omitted them here, several additional functions are available in the
package IPM3`RandomWalks` for performing numerical analysis on random walks.
12 Writing packages 413
Exercises
1. This series of exercises will walk you through the creation of a package Collatz.m,
a package of functions for performing various operations related to the Collatz
problem that we investigated earlier (Exercise 5 of Section 5.3, Exercises 6 and 7 of
Section 6.2, and Exercise 3 of Section 7.6). Recall that the Collatz function, for any
integer n, returns 3 n 1 for odd n, and
n
2
for even n. The (as yet unproven) Collatz
Conjecture is the statement that, for any initial positive integer n, the iterates of the
Collatz function always reach the cycle 4, 2, 1, . Start by creating an auxiliary
function collatz[n] that returns 3 n 1 for n odd and n 2 for n even.
2. Create the function CollatzSequence[n] that returns a list of the iterates of the
auxiliary function collatz[n] from the previous exercise. Here is some sample
output of the CollatzSequence function.
In[1]:= CollatzSequence 7
Out[1]= 7, 22, 11, 34, 17, 52, 26, 13, 40, 20, 10, 5, 16, 8, 4, 2, 1
In[2]:= CollatzSequence 111
Out[2]= 111, 334, 167, 502, 251, 754, 377, 1132, 566, 283, 850, 425,
1276, 638, 319, 958, 479, 1438, 719, 2158, 1079, 3238,
1619, 4858, 2429, 7288, 3644, 1822, 911, 2734, 1367, 4102,
2051, 6154, 3077, 9232, 4616, 2308, 1154, 577, 1732, 866,
433, 1300, 650, 325, 976, 488, 244, 122, 61, 184, 92, 46,
23, 70, 35, 106, 53, 160, 80, 40, 20, 10, 5, 16, 8, 4, 2, 1
3. Create a usage message for CollatzSequence and warning messages for each of
the following situations.
a. notint: the argument to CollatzSequence is not a positive integer
b. argx: CollatzSequence was called with the wrong number of arguments
4. Modify the definition of CollatzSequence that you created in Exercise 2 above
so that it does some error trapping and issues the appropriate warning message that
you created in Exercise 3.
414 An Introduction to Programming with Mathematica
5. Finally, put all the pieces together and write a package CollatzSequence.m that
includes the appropriate BeginPackage and Begin statements, usage messages,
warning messages, and function definitions. Put your package in a directory where
Mathematica can find it on its search path and then test it to see that it returns correct
output such as the examples below.
In[1]:= IPM3`Collatz`
In[2]:= ?CollatzSequence
CollatzSequence n computes the sequence of
Collatz iterates starting with initial value n. The
sequence terminates as soon as it reaches the value 1.
In[3]:= CollatzSequence 37
Out[3]= 37, 112, 56, 28, 14, 7, 22, 11, 34,
17, 52, 26, 13, 40, 20, 10, 5, 16, 8, 4, 2, 1
Here are various cases in which CollatzSequence is given bad input.
In[4]:= CollatzSequence 5
CollatzSequence::notint : First argument, 5,
to CollatzSequence must be a positive integer.
In[5]:= CollatzSequence 4, 6
CollatzSequence::argx :
CollatzSequence called with 2 arguments;
1 argument is expected. More
Out[5]= CollatzSequence 4, 6
12 Writing packages 415
Appendix A How expressions are
evaluated
Evaluation of expressions
Evaluation takes place whenever an expression is entered. Here is the general procedure
followed by Mathematica when evaluating an expression (with a few exceptions):
1. If the expression is a number or a string, it is left unchanged.
In[1]:= 4.58425
Out[1]= 4.58425
2. If the expression is a symbol, it is rewritten if there is an applicable rewrite rule in
the global rule base; otherwise, it is unchanged.
In[2]:= expr
Out[2]= expr
In[3]:= mysymbol
Out[3]= mysymbol
3. If the expression is not a number, string or symbol, its parts are evaluated in a
specific order:
The head of the expression is evaluated.
The arguments of the expression are evaluated in order, except when the head
is a symbol with a Hold attribute. In this case, some of its arguments are left
in their unevaluated forms.
4. After the head and arguments of an expression are each completely evaluated, the
expression consisting of the evaluated head and arguments is rewritten (after
making any necessary changes to the arguments based on the Attributes of
the head) if there is an applicable rewrite rule in the global rule base.
5. After carrying out the previous steps, the resulting expression is evaluated in the
same way and then the result of that evaluation is evaluated, and so on until there
are no more applicable rewrite rules.
The term rewriting process done in steps 2 and 4 above can be described as follows:
pattern match parts of an expression and the left-hand side of a rewrite rule
substitute the values which match labeled blanks in the pattern into the right-hand
side of the rewrite rule and evaluate it
replace the matched part of the expression with the evaluated result
Both built-in and user-defined rewrite rules are available for use in evaluation. When
more than one rewrite rule is found to match an expression, the rule used for term rewrit-
ing is selected based on the following priority:
user-defined rules are used before built-in rules
more specific rules are used before more general rules
one rule is more specific than another if its left-hand side matches fewer expressions;
for example, the rule f[0]:= is more specific than f[_]:=. This is discussed
further in Section 5.3.
The evaluation process can be illustrated with a simple case. We first enter a simple
rewrite rule into the global rule base.
In[4]:= square x_ : x
2
If we now evaluate the following expression, the number 9 is returned as the result.
In[5]:= square 3
Out[5]= 9
We can step through the details of the evaluation process that took place above.
1. The head, square, was evaluated first. The global rule base was searched for a
rewrite rule whose left-hand side was the symbol square. No matching rewrite
rule was found and so the symbol was left unchanged.
2. The argument 3 was evaluated. Since 3 is a number, it was left unchanged.
3. The expression square[3] was evaluated. The global rule base was searched for
a rewrite rule whose left-hand side pattern matched square[3]. The pattern
square[3] was found to match square[x_] and so the value of 3 was substi-
418 An Introduction to Programming with Mathematica
tuted for x in the right-hand side of the rewrite rule, Power[x,2], to give
Power[3,2].
4. Power[3,2] was then evaluated (by the same general procedure) to give 9.
5. The value 9 was evaluated. Since 9 is a number, it was left unchanged.
6. Since there were no more rules to use, the final value 9 was returned.
These steps can be seen in detail by using Trace with the TraceOriginal option
set to True.
In[6]:= Trace square 3 , TraceOriginal True
Out[6]= square 3 , square , 3 ,
square 3 , 3
2
, Power , 3 , 2 , 3
2
, 9
Appendix A How expressions are evaluated 419
Appendix B Debugging
Whenever you write programs, much of your time will be spent in debugging figuring out
why your program does not work. In this appendix, we offer a few tips on debugging, and
also give some examples of common programming errors.
Tracing evaluation
In any programming language, the programmer will, at some point, be faced with an
unexpected, and perhaps, mysterious result. You might be expecting one output, but an
entirely different one is generated. Or your program may not run to completion and only
give error or warning messages that are difficult to decipher. In such situations, you may
find it helpful to take a peek at Mathematicas evaluation process. This is most easily done
with Trace and related functions. For example, using Trace on a simple arithmetic
operation, you can see that Mathematicas evaluator works from the inside out, following
the order of operations for arithmetic.
In[1]:= Trace 2 3 4 5 6
Out[1]= 5 6, 11 , 4 11, 44 , 3 44, 47 , 2 47, 94
Similarly, tracing the evaluation of an If statement shows that only the first argument to
the If function is evaluated initially; the If function itself returns no value, hence the
Null at the end of the trace.
In[2]:= Trace If 4 9, Print "true" , Print "false"
false
Out[2]= 4 9, False , If False, Print true , Print false ,
Print false , MakeBoxes false, StandardForm , "false" , Null
Trace and TracePrint can be especially useful when you know how to use their second
argument. If the second argument is just a symbol, then only those parts of the trace that
use rewrite rules for that symbol are shown. If it is a pattern, only those lines of the trace
that match the pattern will be printed; an example of this was seen in Section 7.1. If the
second argument is a transformation rule, then, when the pattern matches a line of the
trace, the rule is applied before printing it.
For example, first we trace the evaluation of the Fibonacci function showing all
expressions used in the evaluation.
In[3]:= fib 0 : 0
fib 1 : 1
fib n_ : fib n 2 fib n 1
In[6]:= Trace fib 3
Out[6]= fib 3 , fib 3 2 fib 3 1 , 3 2, 1 , fib 1 , 1 ,
3 1, 2 , fib 2 , fib 2 2 fib 2 1 , 2 2, 0 , fib 0 , 0 ,
2 1, 1 , fib 1 , 1 , 0 1, 1 , 1 1, 2
Most of this, you will agree, is not very interesting. We can confine it to only those
parts that involve the applications of a fib rule by giving fib as the second argument to
Trace.
In[7]:= Trace fib 3 , fib
Out[7]= fib 3 , fib 3 2 fib 3 1 , fib 1 , 1 ,
fib 2 , fib 2 2 fib 2 1 , fib 0 , 0 , fib 1 , 1
Perhaps more useful here would be the pattern fib[_], which includes all lines in
the original trace of the formfib[expr].
In[8]:= TracePrint fib 3 , fib _
fib 3
fib 3 2
fib 1
fib 3 1
fib 2
fib 2 2
fib 0
fib 2 1
fib 1
Out[8]= 2
421
An Introduction to Programming with Mathematica
Using a transformation rule, we can show just the arguments of the various calls that
are either fib applied to an integer, or the right-hand side of the recursive rule.
In[9]:= TracePrint fib 3 , fib n_Integer n
3
1
2
0
1
Out[9]= 2
Printing variables
The classic debugging method, used in all programming languages, is to insert Print
statements in the body of a program to show where evaluation is occurring and what the
values of variables are at that point. Keep in mind that if expr is any expression, the com-
pound expression (Print[]; expr) has the same value as expr, so it is easy to insert
Print statements without changing how the program works.
The most common use of Print is to show the values of a functions arguments. A
rule f[x_] := expr can be changed to f[x_] := (Print[x]; expr) and it will print
the value of the argument in each call.
In[10]:= F n_ : Print n ; F n 2 F n 1 ; n 1
In[11]:= F 4
4
2
3
2
Out[11]= 2 F 0 3 F 1
Reap and Sow
Another way of viewing intermediate results in a computation is to use Reap and Sow.
The arguments to Sow will be collected by the nearest Reap.
Appendix B Debugging 422
For example, recall the simple procedural program we created in Chapter 5 for
implementing Newtons method for root finding.
In[12]:= f x_ : x
2
50
In[13]:= a 50;
Do a N a
f a
f a
, 7
At the end of the Do loop, the approximation to the root is in the symbol a.
In[15]:= a
Out[15]= 7.07107
If we Sow a, and then Reap all the values that a took on during the loop, we can see
the intermediate values.
In[16]:= a 50;
Reap
Do Sow a N a
f a
f a
, 7
2
Out[17]= 25.5, 13.7304, 8.68597, 7.22119, 7.07263, 7.07107, 7.07107
Common errors
Many of the errors you will see when programming are obvious. Here is one of the most
common ones.
In[18]:= Part x, y, z , 4
Part::partw : Part 4 of x, y, z does not exist. More
Out[18]= x, y, z 4
Here, you are attempting to extract a part of an expression that does not have that
part; that is, trying to extract the fourth element of a list with only three elements.
Another thing you will often see is an entire expression returned instead of a value
sometimes the exact same expression you entered.
In[19]:= y n : Table i, i, n, n
In[20]:= y 10
Out[20]= y 10
423
An Introduction to Programming with Mathematica
What is the problem? The n in the argument list for y is missing the blank. Mathemat-
ica sees the left-hand side as a pattern that matches the expression y[n] and nothing else
in particular, not y[10]. Of course, when there are no rules to apply to an expression,
Mathematica is done it does not even know there is an error!
Another very common case where this occurs is when you fail to supply enough
arguments to a function.
In[21]:= Clear f, x, y, r
In[22]:= f x_, r___ , y_ : If x 0, y, f r , f r
In[23]:= f , _ :
In[24]:= f 5, 4, 17 , 1
Out[24]= 1, f 4, 17
Similarly, this error occurs when you supply too few arguments.
In[25]:= Clear g
In[26]:= g x_, r___ , y_ : If x 0, y, g r, y , g r, y
In[27]:= g , _ :
In[28]:= g 5, 4, 17, 12, 21 , 1
Out[28]= 1, g 4, 17, 12, 21, 1
In the first example, the recursive call to f had just one argument, and there were no
rules for this case. In the second, we forgot to put the r in list braces in the recursive call,
so g was called with all the elements of r as arguments, giving it too many arguments.
Another very common error is to get your program in a loop where it seems to go on
forever. If this happens when you are working with recursive definitions, the chances are
that your function is continually making recursive calls and not finishing them. In this case,
you will reach Mathematicas limit on the number of recursive calls it allows, which is
stored in the variable $RecursionLimit.
In[29]:= h x_ : h x 1 h x 1
In[30]:= h 0 : 0
Appendix B Debugging 424
In[31]:= h 1
$RecursionLimit::reclim :
Recursion depth of 256 exceeded. More
General::stop :
Further output of $RecursionLimit::reclim will
be suppressed during this calculation. More
Out[31]= $Aborted
Another possibility is to get the same message, but for $IterationLimit, as we
saw in Section 7.4.
In either case, Mathematica may not stop the computation, but instead continue to
give this message. If this occurs, you will have to terminate the program from the key-
board, as described in Section 1.2. There are times when you may want to increase the
recursion limit, which you can do by assigning a larger integer to $RecursionLimit,
but usually if you exceed it you are in a loop.
When solving problems using iteration, you may go into a loop without doing
recursive calls, in which case the program will just go on forever without printing any error
messages, or may print an error message indicating that $IterationLimit is exceeded.
One option is to terminate the program from the keyboard. You can increase the value of
$IterationLimit, but only do that if you are sure there is no error.
425 An Introduction to Programming with Mathematica
References
Abbott, P. Finding roots in an interval. The Mathematica Journal, 7(2)108112, 1998.
Bentley, J. L. and Friedman, J. H. Algorithms for reporting and counting geometric intersections. Technical
Report C-28, IEEE Transactions on Computing, 1979.
Burden, R. L. and Faires, J. D. Numerical Analysis. Pacific Grove, CA: Brooks/Cole Publishing Co., 7th edition,
2000.
Casti, J. Reality Rules I, Picturing the World in Mathematics The Fundamentals. New York: John Wiley & Sons,
Inc., 1992.
Casti, J. Reality Rules II, Picturing the World in Mathematics The Frontier. New York: John Wiley & Sons, Inc.,
1992.
Crandall, R. E. Personal communication, March 1993.
Crandall, R. E. Projects in Scientific Computation. Santa Clara: TELOS/Springer-Verlag, 1994.
Crandall, R. E. Topics in Advanced Scientific Computation. Santa Clara: TELOS/Springer-Verlag, 1996.
Crandall, R. E. and Pomerance, C. Prime Numbers: A Computational Perspective. New York: Springer-Verlag,
2001
de Berg, M., van Kreveld, M., Overmars, and M., Schwarzkopf, O. Computational Geometry, Algorithms and
Applications. Heidelberg: Springer-Verlag, 2nd edition, 2000.
Gardner, M. Fractal Music, Hypercards and More. New York: W. H. Freeman and Company, 1992.
Garey, M. and Johnson, D. Computers and Intractability: A Guide to the Theory of NP-Completeness. San Francisco:
W. H. Freeman, 1979.
Gaylord, R. and Wellin, P. Computer Simulations with Mathematica, Explorations in Complex Physical and
Biological Systems. Santa Clara: TELOS/Springer-Verlag, 1995.
Glynn, J. and Gray, T. The Beginners Guide to Mathematica, Version 4. Cambridge University Press, 1999.
Golin, M. and Sedgewick, R. Analysis of a simple yet efficient convex hull algorithm. In 4th Annual Symposium
on Computational Geometry. ACM, 1988.
Golub, G. H. and Van Loan, C. F. Matrix Computations. Baltimore: Johns Hopkins University Press, 3rd
edition, 1996.
Graham, R. An efficient algorithm for determining the convex hull of a finite planar set. Information Processing
Letters, 1, 1972.
Graham, R., Knuth, D. E., and Patashnik, O. Concrete Mathematics. Reading: Addison-Wesley Publishing
Company, 2nd edition, 1994.
Gray, A. Modern Differential Geometry of Curves and Surfaces with Mathematica. Boca Raton: CRC Press, 2nd
edition, 1997.
Gray, T. and Glynn, J. Exploring Mathematics with Mathematica. Redwood City: Addison-Wesley Publishing
Company, 1992.
Hayes, A. Sums of cubes of digits, driven to abstraction. Mathematica in Education 1(4) 311, 1992.
Hibbard, A. and Levasseur, K. Exploring Abstract Algebra with Mathematica. New York: Springer-Verlag, 1999.
Honsberger, R. Mathematical Gems II. The Dolciani Mathematical Expositions, Number Two. Providence:
The Mathematical Association of America, 1976.
Jarvis, R. A. On the identification of the convex hull of a finite set of points in the plane. Information Processing
Letters, 2, 1973.
Knuth, D. E. The Art of Computer Programming, Volume 1, Fundamental Algorithms. Reading: Addison-Wesley
Publishing Company, 3rd edition, 1997.
Knuth, D. E. The Art of Computer Programming, Volume 2, Seminumerical Algorithms. Reading: Addison-Wesley
Publishing Company, 3rd edition, 1997.
Korhfage, R. R. Discrete Computational Structures. Orlando: Academic Press, Inc., 2nd edition, 1984.
Lagarias, J. The 3x 1 problem. The American Mathematical Monthly, 92 323, 1985.
Lawler, E., Lenstra, J. K., Rinnooy Kan, A. H. G., and Shmoys, D. B. The Traveling Salesman Problem: A
Guided Tour of Combinatorial Optimization. New York: John Wiley & Sons, Inc., 1985.
Lin, S. Computer solutions of the traveling salesman problem. Bell System Technical Journal, 44 22452269,
1965.
McMahon, T. A. and Bonner, J. T. On Size and Life. New York: Scientific American Books, Inc., 1983.
Maeder, R. The design of the Mathematica programming language. Dr. Dobbs Journal, 17(4) 86, 1992.
Maeder, R. The Mathematica Programmer. Cambridge, MA: Academic Press, Inc., 1994.
Maeder, R. Programming in Mathematica. Reading, MA: Addison-Wesley Publishing Company, 3rd edition,
1997.
Maeder, R. Computer Science with Mathematica: Theory and Practice for Science, Mathematics, and Engineering.
Cambridge University Press, 2000.
Mandelbrot, B. The Fractal Geometry of Nature. New York: W. H. Freeman and Company, 1988.
428 An Introduction to Programming with Mathematica
The Mathematica Journal. Wolfram Media, Champaign.
Mathews, M. V. The Technology of Computer Music. Cambridge, MA: MIT Press, 1969.
Miyaji, C. and Abbott, P. MathLink: Network Programming with Mathematica. Cambridge University Press,
2001.
Nijenhuis, A. and Wilf, H. Combinatorial Algorithms. New York: Academic Press, Inc., 2nd edition, 1978.
ORourke, J. Computational Geometry in C. Cambridge University Press, 2nd edition, 1998.
Pemmaraju, S. and Skiena, S. Computational Discrete Mathematics: Combinatorics and Graph Theory with Mathemat-
ica. Cambridge University Press, 2003.
Pierce, J. R. The Science of Musical Sound. New York: W. H. Freeman & Company, 1983.
Platzman, L. K. and Bartholdi, J. J. Spacefilling curves and the planar traveling salesman problem. Journal
Assoc. for Computing Machinery, 36 719737, 1989.
Poundstone, W. The Recursive Universe: Cosmic Complexity and the Limits of Scientific Knowledge. Oxford: Oxford
University Press, 1985.
Preparata, F. P. and Shamos, M. I. Computational Geometry: An Introduction. New York: Springer-Verlag, 1985.
Reingold, E. M. and Tilford, J. S. Tidier drawings of trees. IEEE Trans. Software Eng, SE7 223228, 1981.
Rosenkrantz, D. J. and Stearns, R. E. An analysis of several heuristics for the traveling salesman problem.
SIAM Journal of Computing, 6 563581, 1977.
Rossing, T. D. The Science of Sound. Reading: Addison-Wesley Publishing Company, 2nd edition, 1990.
Rust, B. and Burrus, W. R. Mathematical Programming and the Numerical Solution of Linear Equations. New
York: American Elsevier Publishing Co., 1972.
Sedgewick, R. Algorithms. Reading: Addison-Wesley Publishing Company, 1988.
Shamos, M. I. and Hoey, D. Closest-point problems. In 16th Annual Symposium on Foundations of Computer
Science. IEEE, 1975.
Shepard, R. The analysis of proximities: multidimensional scaling with an unknown distance factor. Psychomet-
rics, 27 125140, 1962.
Skeel, R. D. and Keiper, J. B. Elementary Numerical Computing with Mathematica. New York: McGraw-Hill, Inc.,
1993.
Thomsen, D. E. Making music fractally. Science News, 117 187, 1980.
Trott, M. The Mathematica Guidebook for Programming. New York: Springer-Verlag, 2004.
Trott, M. The Mathematica Guidebook for Graphics. New York: Springer-Verlag, 2004.
Voss, R. F. and Clarke, J. 1 f noise in music and speech. Nature, 258 317318, 1975.
References 429
Voss, R. F. and Clarke, J. 1 f noise in music and speech. The Journal of the Acoustical Society of America, 63
258263, 1978.
Wagon, S. Mathematica in Action. Santa Clara: TELOS/Springer-Verlag, 2nd edition, 1999.
Weiss, G. H. Random walks and their applications. American Scientist, 71 6571, 1983.
Wickham-Jones, T. Computer Graphics with Mathematica. Santa Clara: TELOS/Springer-Verlag, 1994.
Wolfram, S. The Mathematica Book. Wolfram Media, Inc., 5th edition, 2003.
430 An Introduction to Programming with Mathematica
Solutions to exercises
2 The Mathematica language
2.1 Expressions
1. The expression a(b+c) is given in full form as Times[a,Plus[b,c]].
2. This is simply
a
b c
as can be seen by evaluating the full form expression.
In[1]:= Times a, Power Plus b, c , 1
Out[1]=
a
b c
3. Looking at the internal representation of this expression with FullForm helps to unwind the
part specification.
In[2]:= FullForm x^2 y z w
Out[2]//FullForm=
Times Power w, 1 , Plus Power x, 2 , y , z
In[3]:= x^2 y z w 2, 1, 2
Out[3]= 2
4. There are three terms in the expression, with the term b x being the second.
In[4]:= expr a x
2
b x c;
In[5]:= FullForm expr
Out[5]//FullForm=
Plus c, Times b, x , Times a, Power x, 2
The b is the first element of Times[b,x], so the part specification is 2,1.
In[6]:= expr 2
Out[6]= b x
In[7]:= expr 2, 1
Out[7]= b
2.2 Definitions
1. This exercise focuses on the difference between immediate and delayed assignments.
a. This will generate a list of n random numbers.
In[1]:= randLis1 n_ : Table Random , n
In[2]:= ?randLis1
Global`randLis1
randLis1 n_ : Table Random , n
In[3]:= randLis1 3
Out[3]= 0.0405431, 0.043554, 0.699358
b. Since the definition for x is an immediate assignment, its value does not change in the body
of randLis2. But each time randLis2 is called, a new value is assigned to x.
In[4]:= randLis2 n_ : x Random ; Table x, n
In[5]:= ?randLis2
Global`randLis2
randLis2 n_ : x Random ; Table x, n
In[6]:= randLis2 3
Out[6]= 0.651026, 0.651026, 0.651026
c. Because the definition for x is a delayed assignment, the definition for randLis3 is
functionally equivalent to randLis1.
In[7]:= randLis3 n_ : x : Random ; Table x, n
In[8]:= ?randLis3
Global`randLis3
randLis3 n_ : x : Random ; Table x, n
In[9]:= randLis3 3
Out[9]= 0.304574, 0.184163, 0.744351
d. Recall that in an immediate assignment, the right-hand side of the definition is evaluated
first. But in this case, n does not have a value, so Table is not able to evaluate properly.
432 An Introduction to Programming with Mathematica
In[10]:= randLis4 n_ Table Random , n
Table::iterb : Iterator n does not have appropriate bounds. More
Out[10]= Table Random , n
In[11]:= ?randLis4
Global`randLis4
randLis4 n_ Table Random , n
2.3 Predicates and Boolean operations
1. There are several ways to define this function, using either the relational operator for less than,
or with the absolute value function.
In[1]:= f x_ : 1 x 1
In[2]:= f x_ : Abs x 1
In[3]:= f 4
Out[3]= False
In[4]:= f 0.35
Out[4]= True
2. A number n can be considered a natural number if it is an integer and greater than or equal to
zero.
In[5]:= Positive 0
Out[5]= False
In[6]:= NaturalQ n_ : IntegerQ n && n 0
In[7]:= NaturalQ 0
Out[7]= True
In[8]:= NaturalQ 4
Out[8]= False
3. The empty set is a subset of every set. So first we need a definition to cover this case.
In[9]:= SubsetQ , lis2_ : True
Solutions to exercises 433
The intersection of lis1 and lis2 will be identical to lis1 whenever lis1 is a subset of
lis2.
In[10]:= SubsetQ lis1_, lis2_ : Intersection lis1, lis2 lis1
In[11]:= A a, b, c ;
B a, b, c, d, e ;
In[13]:= SubsetQ A, B
Out[13]= True
We can also give a definition in terms of the subset character which can be entered by typing
-sub- or by using one of the palettes.
In[14]:= lis1_ lis2_ : Intersection lis1, lis2 lis1
In[15]:= A B
Out[15]= True
3 Lists
3.2 Creating and measuring lists
1. You can take every other element in the iterator list, or encode that in the function 2j.
In[1]:= Table j, i, 0, 8, 2 , j, 0, i, 2
Out[1]= 0 , 0, 2 , 0, 2, 4 , 0, 2, 4, 6 , 0, 2, 4, 6, 8
In[2]:= Table 2 j, i, 0, 4 , j, 0, i
Out[2]= 0 , 0, 2 , 0, 2, 4 , 0, 2, 4, 6 , 0, 2, 4, 6, 8
2. This is probably the simplest way to generate random 1s, 0s, and 1s.
In[3]:= Table Random Integer, 1, 1 , 10
Out[3]= 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
3. Here are three ways to generate the list.
In[4]:= Table 2 Random Integer 1, 10
Out[4]= 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
434 An Introduction to Programming with Mathematica
In[5]:= Table 1
Random Integer
, 10
Out[5]= 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
The following solution will become clearer in the next section after we have discussed the
Part function in some detail.
In[6]:= 1, 1 Table Random Integer, 1, 2 , 10
Out[6]= 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
4. These lists can be generated with Table, using two iterators in the second example.
In[7]:= Table f i , i, 5
Out[7]= f 1 , f 2 , f 3 , f 4 , f 5
In[8]:= Table f i, j , i, 3 , j, 4
Out[8]= f 1, 1 , f 1, 2 , f 1, 3 , f 1, 4 ,
f 2, 1 , f 2, 2 , f 2, 3 , f 2, 4 , f 3, 1 , f 3, 2 , f 3, 3 , f 3, 4
5. From the top level, there are two lists, each consisting of two sublists, each sublist consisting of
two elements.
In[9]:= Dimensions 1, a , 4, d , 2, b , 3, c
Out[9]= 2, 2, 2
3.3 Manipulating lists
1. The Position function tells us that the 9s are located in the second sublist, first position, and
in the fourth sublist, third position.
In[1]:= Position 2, 1, 10 , 9, 5, 7 , 2, 10, 4 , 10, 1, 9 , 6, 1, 6 , 9
Out[1]= 2, 1 , 4, 3
2. This is a straightforward use of the Transpose function.
In[2]:= Transpose x1, y1 , x2, y2 , x3, y3 , x4, y4 , x5, y5
Out[2]= x1, x2, x3, x4, x5 , y1, y2, y3, y4, y5
3. Here is one way to do it. First create a list representing the directions.
In[3]:= NSEW 0, 1 , 0, 1 , 1, 0 , 1, 0 ;
Solutions to exercises 435
In[4]:= Table NSEW Random Integer, 1, 4 , 10
Out[4]= 1, 0 , 0, 1 , 0, 1 , 0, 1 ,
1, 0 , 1, 0 , 0, 1 , 1, 0 , 0, 1 , 1, 0
4. We first drop the first element in the list, then create a nested list of every other element in
the remaining list, and finally unnest the resulting list.
In[5]:= Rest a, b, c, d, e, f, g
Out[5]= b, c, d, e, f, g
In[6]:= Partition %, 1, 2
Out[6]= b , d , f
In[7]:= Flatten %
Out[7]= b, d, f
5.
In[8]:= a, b, c, d 3, 2, 4, 1
Out[8]= c, b, d, a
6.
In[9]:= Transpose 3, 2, 4, 1 , a, b, c, d
Out[9]= 3, a , 2, b , 4, c , 1, d
In[10]:= Sort %
Out[10]= 1, d , 2, b , 3, a , 4, c
In[11]:= Transpose %
Out[11]= 1, 2, 3, 4 , d, b, a, c
In[12]:= % 2
Out[12]= d, b, a, c
3.4 Working with several lists
1. Join expects lists as arguments.
In[1]:= Join z , x, y
Out[1]= z, x, y
436 An Introduction to Programming with Mathematica
2. The trick here is partitioning the joined list so that you get every other element.
In[2]:= expr Join 1, 2, 3, 4 , a, b, c, d
Out[2]= 1, 2, 3, 4, a, b, c, d
In[3]:= Rest expr
Out[3]= 2, 3, 4, a, b, c, d
In[4]:= Partition %, 1, 2
Out[4]= 2 , 4 , b , d
In[5]:= Flatten %
Out[5]= 2, 4, b, d
This can also be done using the Take function.
In[6]:= Take expr, 2, Length expr , 2
Out[6]= 2, 4, b, d
3. This is another way of asking for all those elements that are in the union but not the intersec-
tion of the two sets.
In[7]:= A a, b, c, d ;
B a, b, e, f ;
In[9]:= Complement A B, A B
Out[9]= c, d, e, f
In[10]:= Complement Union A, B , Intersection A, B
Out[10]= c, d, e, f
3.5 Strings and characters
1. Here is a test string we will use for this exercise.
In[1]:= str "this is a test string"
Out[1]= this is a test string
This extracts the first character fromstr.
In[2]:= StringTake str, 1
Out[2]= t
Solutions to exercises 437
Here is its character code.
In[3]:= ToCharacterCode %
Out[3]= 116
For each lowercase letter of the English alphabet, subtracting 32 gives the corresponding
uppercase character.
In[4]:= % 32
Out[4]= 84
Convert back to a character.
In[5]:= FromCharacterCode %
Out[5]= T
Take the original string minus its first character.
In[6]:= StringDrop str, 1
Out[6]= his is a test string
Finally, join the previous string with the capital T.
In[7]:= StringJoin %%, %
Out[7]= This is a test string
2. We first need to extract the character codes from this string.
In[8]:= numstr "73"
Out[8]= 73
In[9]:= ToCharacterCode numstr 1
Out[9]= 55
In[10]:= 10 % 48
Out[10]= 70
In[11]:= ToCharacterCode numstr 2
Out[11]= 51
In[12]:= % 48
Out[12]= 3
438 An Introduction to Programming with Mathematica
In[13]:= % %%%
Out[13]= 73
Here it is all put together in one line.
In[14]:= 10 ToCharacterCode numstr 1 48 ToCharacterCode numstr 2 48
Out[14]= 73
There is a built-in function for this task, ToExpression. See the next exercise for details.
3.
In[15]:= numb ToCharacterCode "73"
Out[15]= 55, 51
In[16]:= numb 48
Out[16]= 7, 3
In[17]:= 8 Part %, 1 Part %, 2
Out[17]= 59
Here is another approach that converts the single characters into regular expressions and then
operates on those directly.
In[18]:= ToExpression Characters "73"
Out[18]= 7, 3
In[19]:= 8 First % Last %
Out[19]= 59
4. One approach converts the string to character codes.
In[20]:= ToCharacterCode "10495"
Out[20]= 49, 48, 52, 57, 53
In[21]:= % 48
Out[21]= 1, 0, 4, 9, 5
In[22]:= Reverse Table 10
j
, j, 0, 4
Out[22]= 10000, 1000, 100, 10, 1
Solutions to exercises 439
In[23]:= %.%%
Out[23]= 10495
A direct approach uses ToExpression.
In[24]:= ToExpression "10495"
Out[24]= 10495
5. First, consider the character code of a string.
In[25]:= ToCharacterCode "best"
Out[25]= 98, 101, 115, 116
Then we need only know if this list of codes is in order.
In[26]:= OrderedQ %
Out[26]= True
So here is our Boolean function OrderedWordQ.
In[27]:= OrderedWordQ w_String : OrderedQ ToCharacterCode w
Now we will find all the words in the dictionary file that comes with Mathematica that are
ordered in this way. First we generate a platform-independent path to the dictionary file.
In[28]:= wordfile ToFileName $InstallationDirectory, "Documentation",
"English", "Demos", "DataFiles" , "dictionary.dat"
Out[28]= C:\Program Files\Wolfram Research\Mathematica\5
.1\Documentation\English\Demos\DataFiles\dictionary.dat
Then we read the file using ReadList, specifying the type of data we are reading in as a
Word.
In[29]:= words ReadList wordfile, Word ;
Finally, we select those elements from the list words that pass the OrderedWordQ test.
In[30]:= Select words, OrderedWordQ Shallow
Out[30]//Shallow=
a, AAA, AAAS, abbe, abbey, abbot, Abbott, abc, Abe, Abel, 565
6. Here is the function that checks if a string is a palindrome.
In[31]:= PalindromeQ str_String : StringReverse str str
440 An Introduction to Programming with Mathematica
In[32]:= PalindromeQ "mood"
Out[32]= False
In[33]:= PalindromeQ "PoP"
Out[33]= True
In[34]:= PalindromeQ num_Integer : PalindromeQ ToString num
In[35]:= PalindromeQ 12522521
Out[35]= True
Create a path to the file dictionary.dat.
In[36]:= dictfile ToFileName $BaseDirectory,
"Applications", "IPM3", "DataFiles" , "dictionary.dat"
Out[36]= C:\Documents and Settings\All Users\Application Data\
Mathematica\Applications\IPM3\DataFiles\dictionary.dat
Import the file.
In[37]:= words Import dictfile, "Words" ;
In[38]:= Select words, PalindromeQ
Out[38]= a, AAA, ABA, ala, AMA, ana, b, bib, bob, bub, c, CDC, civic, d, dad, deed,
did, DOD, dud, e, eke, ere, eve, ewe, eye, f, g, gag, gig, gog, h, huh, i,
ii, iii, j, k, l, level, m, madam, minim, mum, n, non, noon, nun, o, p, pap,
PDP, peep, pep, pip, poop, pop, pup, q, r, radar, refer, rever, rotor, s,
sis, s's, t, tat, teet, tenet, tit, TNT, toot, tot, u, v, w, wow, x, y, z
4 Functional programming
4.2 Functions for manipulating expressions
1. Here is a sample set of pairs of numbers.
In[1]:= data 1, 2 , 2, 3 , 3, 4 , 4, 5 , 5, 6 ;
The pairSum function can be written simply as:
In[2]:= addPair x_, y_ : x y
Solutions to exercises 441
Finally we map pairSum across data.
In[3]:= Map addPair, data
Out[3]= 3, 5, 7, 9, 11
2. Here is a sample set of pairs of numbers.
In[4]:= data 1, 2 , 2, 3 , 3, 4 , 4, 5 , 5, 6 ;
Since Apply normally works at level 0, we need to give it a third argument to get it to apply
Plus at level 1.
In[5]:= Apply Plus, data, 1
Out[5]= 3, 5, 7, 9, 11
3. First you need to transpose the matrix and then reverse the pairs.
In[6]:= lis 1, 2, 3 , 4, 5, 6
Out[6]= 1, 2, 3 , 4, 5, 6
In[7]:= Transpose lis
Out[7]= 1, 4 , 2, 5 , 3, 6
In[8]:= Map Reverse, %
Out[8]= 4, 1 , 5, 2 , 6, 3
This can also be accomplished using Thread.
In[9]:= Map Reverse, Thread lis
Out[9]= 4, 1 , 5, 2 , 6, 3
4. This can be done either in two steps, or by using the Inner function.
In[10]:= Transpose 1, 2 , 3, 4 x, y
Out[10]= x, 3 x , 2 y, 4 y
In[11]:= Apply Plus, %
Out[11]= x 2 y, 3 x 4 y
In[12]:= Inner Times, 1, 2 , 3, 4 , x, y , Plus
Out[12]= x 2 y, 3 x 4 y
442 An Introduction to Programming with Mathematica
5. To get down to the second level of nested lists, you have to use a second argument to Apply.
In[13]:= facs FactorInteger 3628800
Out[13]= 2, 8 , 3, 4 , 5, 2 , 7, 1
In[14]:= Apply Power, facs, 2
Out[14]= 256, 81, 25, 7
One more use of Apply is needed to multiply these terms.
In[15]:= Apply Times, %
Out[15]= 3628800
Here is a function that puts this all together.
In[16]:= ExpandFactors lis_ : Apply Times, Apply Power, lis, 2
In[17]:= FactorInteger 295232799039604140847618609643520000000
Out[17]= 2, 32 , 3, 15 , 5, 7 , 7, 4 , 11, 3 ,
13, 2 , 17, 2 , 19, 1 , 23, 1 , 29, 1 , 31, 1
In[18]:= ExpandFactors %
Out[18]= 295232799039604140847618609643520000000
Another approach would be to use Transpose to separate the bases from their exponents,
then use MapThread to raise each base to the corresponding exponent.
In[19]:= Transpose facs
Out[19]= 2, 3, 5, 7 , 8, 4, 2, 1
In[20]:= MapThread Power, %
Out[20]= 256, 81, 25, 7
Finally, apply Times to the list.
In[21]:= Apply Times, %
Out[21]= 3628800
In[22]:= ExpandFactors2 lis_ : Apply Times, MapThread Power, Transpose lis
6. Here is a factorization we can use to work through this problem.
In[23]:= facs FactorInteger 10
Out[23]= 2, 8 , 3, 4 , 5, 2 , 7, 1
Solutions to exercises 443
First we extract the prime bases and their exponents.
In[24]:= bases Transpose facs 1
Out[24]= 2, 3, 5, 7
In[25]:= exponents Transpose facs 2
Out[25]= 8, 4, 2, 1
Here then is the inner product, threading Power over the lists and then multiplying the
resulting terms with Times.
In[26]:= Inner Power, bases, exponents, Times
Out[26]= 3628800
Here is a function that combines these steps.
In[27]:= ExpandFactors3 lis_ : Module facs Transpose lis ,
Inner Power, facs 1 , facs 2 , Times
In[28]:= ExpandFactors3 facs
Out[28]= 3628800
7. If we first look at a symbolic result, we should be able to see how to construct our function. For
three vectors and three variables, here is the divergence (think of d as the derivative operator).
In[29]:= Inner d, e1, e2, e3 , v1, v2, v3 , Plus
Out[29]= d e1, v1 d e2, v2 d e3, v3
So for arbitrary-length vectors and variables, we have:
In[30]:= div vecs_, vars_ : Inner D, vecs, vars, Plus
As a check, we can compute the divergence of the standard gravitational or electric force field,
which should be 0.
In[31]:= div x, y, z x
2
y
2
z
2
3 2
, x, y, z
Out[31]=
3 x
2
x
2
y
2
z
2
5 2
3 y
2
x
2
y
2
z
2
5 2
3 z
2
x
2
y
2
z
2
5 2
3
x
2
y
2
z
2
3 2
In[32]:= Simplify %
Out[32]= 0
Finally, we should note that this definition of divergence is a bit delicate as we are doing no
argument checking at this point. For example, it would be sensible to insure that the length of
the vector list is the same as the length of the variable list before starting the computation. The
444 An Introduction to Programming with Mathematica
reader should refer to Chapter 6 for a discussion of how to use pattern matching to deal with
this issue.
4.3 Iterating functions
1. First we generate the step directions.
In[1]:= Table 1
Random Integer
, 10
Out[1]= 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
Then, starting at 0, the fold operation generates the locations.
In[2]:= FoldList Plus, 0, %
Out[2]= 0, 1, 2, 3, 2, 1, 0, 1, 0, 1, 2
2. We can use the method of generating a list of step locations that was shown in an earlier
exercise.
In[3]:= 1, 0 , 1, 0 , 0, 1 , 0, 1 Table Random Integer, 1, 4 , 10
Out[3]= 0, 1 , 1, 0 , 0, 1 , 0, 1 ,
0, 1 , 0, 1 , 0, 1 , 0, 1 , 1, 0 , 1, 0
In[4]:= FoldList Plus, 0, 0 , %
Out[4]= 0, 0 , 0, 1 , 1, 1 , 1, 0 , 1, 1 ,
1, 0 , 1, 1 , 1, 2 , 1, 1 , 0, 1 , 1, 1
3. Starting with 1, we want to fold the Times functions across the first n integers.
In[5]:= fac n_ : Fold Times, 1, Range n
In[6]:= fac 10
Out[6]= 3628800
4.4 Programs as functions
1. The obvious way to do this is to take the list and simply pick out elements at random locations.
Note: the right-most location in the list is given by Length[lis], using the built-in Part and
Random functions.
In[1]:= chooseWithReplacement lis_, n_ :
lis Table Random Integer, 1, Length lis , n
In[2]:= chooseWithReplacement a, b, c, d, e, f, g, h , 3
Out[2]= f, e, c
Solutions to exercises 445
2. Here is our user-defined stringInsert.
In[3]:= stringInsert str1_, str2_, pos_ :
FromCharacterCode Join Take ToCharacterCode str1 , pos 1 ,
ToCharacterCode str2 , Drop ToCharacterCode str1 , pos 1
In[4]:= stringInsert "Joy world", "to the ", 5
Out[4]= Joy to the world
In[5]:= stringDrop str_, pos_ :
FromCharacterCode Drop ToCharacterCode str , pos
3. There are many ways of defining this function. Here we take advantage of the fact that if p and
q are each lists of two numbers, then p-q will subtract element-wise.
In[6]:= distance pt1_, pt2_ : Apply Plus, pt1 pt2
2
In[7]:= distance 2, 5 , 6, 8
Out[7]= 5
4. We assume that lis1 is longer than lis2 and pair off the corresponding elements in the lists
and then tack on the leftover elements fromlis1.
In[8]:= interLeave2 lis1_, lis2_ :
Flatten Join Transpose lis2, Take lis1, Length lis2 ,
Take lis1, Length lis2 Length lis1
In[9]:= interLeave2 a, b, c, d , 1, 2, 3
Out[9]= 1, a, 2, b, 3, c, d
5. After creating the card deck, we cut it in half and interleave the two halves.
In[10]:= cardDeck
Flatten Outer List, , , , , Join Range 2, 10 , J, Q, K, A , 1 ;
In[11]:= Flatten Transpose Partition cardDeck, 26 , 1
Out[11]= , 2 , , 2 , , 3 , , 3 , , 4 , , 4 , , 5 , , 5 , , 6 ,
, 6 , , 7 , , 7 , , 8 , , 8 , , 9 , , 9 , , 10 ,
, 10 , , J , , J , , Q , , Q , , K , , K , , A , , A ,
, 2 , , 2 , , 3 , , 3 , , 4 , , 4 , , 5 , , 5 , , 6 ,
, 6 , , 7 , , 7 , , 8 , , 8 , , 9 , , 9 , , 10 ,
, 10 , , J , , J , , Q , , Q , , K , , K , , A , , A
446 An Introduction to Programming with Mathematica
6. First, here is how we might write our own StringJoin.
In[12]:= FromCharacterCode
Join ToCharacterCode "To be, " , ToCharacterCode "or not to be"
Out[12]= To be, or not to be
And here is a how we might implement a StringReverse.
In[13]:= FromCharacterCode Reverse ToCharacterCode %
Out[13]= eb ot ton ro ,eb oT
4.5 Auxiliary functions
1. In the first definition, we only use one auxiliary function inside the Module.
In[1]:= latticeWalk1 n_ : Module steps ,
steps m_ : 1, 0 , 1, 0 , 0, 1 , 0, 1
Table Random Integer, 1, 4 , m ; FoldList Plus, 0, 0 , steps n
In[2]:= latticeWalk1 10
Out[2]= 0, 0 , 0, 1 , 0, 0 , 1, 0 , 1, 1 ,
0, 1 , 1, 1 , 2, 1 , 1, 1 , 0, 1 , 0, 0
Here we use two auxiliary functions, making the code a bit easier to read.
In[3]:= latticeWalk2 n_ :
Module choices, steps , choices 1, 0 , 1, 0 , 0, 1 , 0, 1 ;
steps m_ : choices Table Random Integer, 1, 4 , m ;
FoldList Plus, 0, 0 , steps n
In[4]:= latticeWalk2 10
Out[4]= 0, 0 , 0, 1 , 1, 1 , 2, 1 , 3, 1 ,
4, 1 , 5, 1 , 4, 1 , 5, 1 , 5, 2 , 4, 2
2. The following function creates a local function perfectQ using the Module construct. It
then checks every other number between n and m by using a third argument to the Range
function.
In[5]:= PerfectSearch n_, m_ : Module perfectQ ,
perfectQ j_ : Apply Plus, Divisors j 2 j;
Select Range n, m, 2 , perfectQ
In[6]:= PerfectSearch 2, 1000
Out[6]= 6, 28, 496
Solutions to exercises 447
This function does not guard against the user supplying bad inputs. For example, if the user
starts with an odd number, then this version of PerfectSearch will check every other odd
number, and, since it is known that there are no odd numbers below at least 10
300
, none is
reported.
In[7]:= PerfectSearch[1, 1000]
Out[7]=
You can fix this situation by using the (as yet unproved) assumption that there are no odd
perfect numbers. This next version first checks that the first argument is an even number.
In[8]:= Clear PerfectSearch
In[9]:= PerfectSearch n_?EvenQ, m_ : Module perfectQ ,
perfectQ j_ : Apply Plus, Divisors j 2 j;
Select Range n, m, 2 , perfectQ
Now, the function only works if the first argument is even.
In[10]:= PerfectSearch 2, 1000
Out[10]= 6, 28, 496
In[11]:= PerfectSearch 1, 1000
Out[11]= PerfectSearch 1, 1000
3. This only requires a slight change to the code from the PerfectSearch function from the
previous exercise.
In[12]:= PerfectSearch n_, m_, 3 : Module perfectQ ,
perfectQ j_ : Apply Plus, Divisors j 3 j;
Select Range n, m , perfectQ
It appears as if there are only three 3-perfect numbers below 10
6
.
In[13]:= PerfectSearch 1, 10
6
, 3
Out[13]= 120, 672, 523776
4. Again, this function only requires a slight modification from that for the PerfectSearch
function above.
In[14]:= PerfectSearch n_, m_, 4 : Module perfectQ ,
perfectQ j_ : Apply Plus, Divisors j 4 j;
Select Range n, m , perfectQ
The following computation can be quite time consuming and requires a fair amount of
memory to run to completion. If your computers resources are limited, you should split up the
search intervals into smaller units.
448 An Introduction to Programming with Mathematica
In[15]:= PerfectSearch 1, 2200000, 4 Timing
Out[15]= 54.769 Second, 30240, 32760, 2178540
5. This function requires a third argument.
In[16]:= Clear PerfectSearch ;
PerfectSearch n_, m_, k_ : Module perfectQ ,
perfectQ j_ : Apply Plus, Divisors j k j;
Select Range n, m , perfectQ
In[18]:= PerfectSearch 1, 100, 2
Out[18]= 6, 28
6. This function will require two auxiliary functions, the function and a predicate to determine
whether a number is super-perfect.
In[19]:= SuperPerfectSearch a_, b_ : Module sigma, superQ ,
sigma n_ : Apply Plus, Divisors n ;
superQ n_ : Nest sigma, n, 2 2 n;
Select Range a, b , superQ
Here, then, are all super-perfect numbers less than 100,000.
In[20]:= SuperPerfectSearch 1, 100000
Out[20]= 2, 4, 16, 64, 4096, 65536
7. Many implementations are possible for convertToDate. We show here a version that uses
string manipulation. First we extract the digits from the 8-digit number.
In[21]:= d IntegerDigits 20030515
Out[21]= 2, 0, 0, 3, 0, 5, 1, 5
The first four digits give us the year.
In[22]:= d Range 4
Out[22]= 2, 0, 0, 3
Here is a function that takes a list of digits, converts them to strings, concatenates them into
one string, and then converts that into a number.
In[23]:= convert str_ : ToExpression StringJoin Map ToString, str
In[24]:= convert d Range 4
Out[24]= 2003
Solutions to exercises 449
In[25]:= Head %
Out[25]= Integer
Using convert, here are the auxiliary functions to extract the year, month, and day as
numbers.
In[26]:= year str_ : convert str Range 4
In[27]:= year d
Out[27]= 2003
In[28]:= month str_ : convert str 5, 6
In[29]:= month d
Out[29]= 5
In[30]:= day str_ : convert str 7, 8
In[31]:= day d
Out[31]= 15
And here are all the pieces put together in the function convertToDate.
In[32]:= convertToDate n_ : Module d, convert, year, month, day ,
d IntegerDigits n ;
convert st_ : ToExpression StringJoin Map ToString, st ;
year st_ : convert st Range 4 ;
month st_ : convert st 5, 6 ;
day st_ : convert st 7, 8 ;
year d , month d , day d
In[33]:= convertToDate 20030515
Out[33]= 2003, 5, 15
4.6 Pure functions
1. This function adds the squares of the elements in lis.
In[1]:= elementsSquared lis_ : Apply Plus, lis
2
In[2]:= elementsSquared 3, 29, 2, 17
Out[2]= 1143
450 An Introduction to Programming with Mathematica
Using a pure function, this becomes:
In[3]:= Function lis, Apply Plus, lis
2
3, 29, 2, 17
Out[3]= 1143
or simply,
In[4]:= Apply Plus, #
2
& 3, 29, 2, 17
Out[4]= 1143
2. Here is the function that sums the digits of any integer.
In[5]:= sumdigits x_Integer : Apply Plus, IntegerDigits x
In[6]:= sumdigits 629
Out[6]= 17
Using a pure function, this becomes:
In[7]:= Function x, Apply Plus, IntegerDigits x 629
Out[7]= 17
In[8]:= Apply Plus, IntegerDigits # & 629
Out[8]= 17
3. First, here is the distance function.
In[9]:= distance pt1_, pt2_ : Apply Plus, pt1 pt2
2
Here are some sample points.
In[10]:= points Table Random , 5 , 2
Out[10]= 0.408123, 0.110529 , 0.640705, 0.227085 ,
0.605818, 0.074615 , 0.868053, 0.302804 , 0.381267, 0.66605
Just as a check, this computes the distance between the first and second points in our list.
In[11]:= distance points 1 , points 2
Out[11]= 0.260153
Now we need the distance between every pair of points. So we first create the set of pairs.
In[12]:= pairs Distribute points, points , List ;
Solutions to exercises 451
Then we apply the distance function and take the Max.
In[13]:= Max Apply distance, pairs, 1
Out[13]= 0.632628
This puts it all together using a pure function in place of the distance function. Since the
diameter function operates on lists of pairs of numbers, we need to specify them in our pure
function by means of #1 and #2.
In[14]:= diameter lis_ :
Max Apply Sqrt Apply Plus, #1 #2 ^2 &,
Distribute lis, lis , List , 1
In[15]:= diameter points
Out[15]= 0.632628
As a final note, this function is not as efficient as it could be since it computes the distance
from every point to itself, as well as computing both the distance from point a to point b and
from point b to point a, for every pair of points a and b. In other words, for n points, we are
computing n
2
distances when we only need to compute
n
2
distances, highly sub-optimal. We
leave the optimization of this function as an exercise to the reader.
4. Using pure functions, removeRand becomes:
In[16]:= Function lis, Delete lis, Random Integer, 1, Length lis
a, b, c, d, e
Out[16]= a, b, c, e
In[17]:= Delete #1, Random Integer, 1, Length # & a, b, c, d, e
Out[17]= a, c, d, e
5. Here is the deal function written using a pure function in place of removeRand.
In[18]:= deal n_ : Module cardDeck , cardDeck
Flatten Outer List, , , , , Join Range 2, 10 , J, Q, K, A , 1 ;
Complement cardDeck,
Nest Delete #1, Random Integer, 1, Length #1 &, cardDeck, n
In[19]:= deal 5
Out[19]= , A , , 2 , , 9 , , 2 , , 5
6. This function is ideally written as an iteration.
In[20]:= RepUnit n_ : Nest 10 # 1 &, 1, n 1
452 An Introduction to Programming with Mathematica
In[21]:= RepUnit 7
Out[21]= 1111111
In[22]:= Map RepUnit # &, Range 12
Out[22]= 1, 11, 111, 1111, 11111, 111111, 1111111, 11111111,
111111111, 1111111111, 11111111111, 111111111111
7. Notice that it is not necessary to use the Module function here because the only expressions
on the right-hand side of the function definition are pure functions, built-in functions, and the
names of the arguments of the function.
In[23]:= chooseWithoutReplacement lis_, n_ : Complement lis,
Nest Delete #1, Random Integer, 1, Length #1 &, lis, n
In[24]:= chooseWithoutReplacement a, b, c, d, e , 4
Out[24]= a, c, d, e
8. Using the list of the step increments in the north, south, east, and west directions, this ten-step
walk starts at the origin.
In[25]:= NestList #1 1, 0 , 1, 0 , 0, 1 , 0, 1 Random Integer, 1, 4 &,
0, 0 , 10
Out[25]= 0, 0 , 1, 0 , 1, 1 , 1, 0 , 0, 0 ,
1, 0 , 2, 0 , 3, 0 , 4, 0 , 4, 1 , 4, 2
9. Here is the path to the dictionary file.
In[26]:= dictfile ToFileName $InstallationDirectory, "Documentation",
"English", "Demos", "DataFiles" , "dictionary.dat"
Out[26]= C:\Program Files\Wolfram Research\Mathematica\5
.1\Documentation\English\Demos\DataFiles\dictionary.dat
This reads in the file using ReadList specifying the type of data we are reading in as a Word.
In[27]:= words ReadList dictfile, Word ;
Here are three words from the dictionary.
In[28]:= words 5, 55, 555
Out[28]= Aaron, abolish, alder
First we need to create a function that takes a string as an argument and returns True if its
first character is char. As a first step, here is a pure function that checks if the first character
of the argument being passed to it ("abolish") starts with the letter "a".
Solutions to exercises 453
In[29]:= StringTake #, 1 "a" & "abolish"
Out[29]= True
Now we can use this pure function as the test to select all those words in lis that pass this
particular test.
In[30]:= WordsStartingWith lis_, char_ :
Select lis, StringTake #, 1 char &
Finally we can check all the words in the dictionary file that start with the letter "z" say.
In[31]:= WordsStartingWith words, "z"
Out[31]= z, zag, zagging, zap, zazen, zeal, zealot, zealous, zebra, zenith, zero,
zeroes, zeroth, zest, zesty, zeta, zig, zigging, zigzag, zigzagging,
zilch, zinc, zing, zip, zircon, zirconium, zloty, zodiac, zodiacal,
zombie, zone, zoo, zoology, zoom, zounds, z's, zucchini, zygote
This can also be accomplished using the new (in Version 5.1) StringMatchQ together with a
wildcard character.
In[32]:= Select words, StringMatchQ #, "z " &
Out[32]= z, zag, zagging, zap, zazen, zeal, zealot, zealous, zebra, zenith, zero,
zeroes, zeroth, zest, zesty, zeta, zig, zigging, zigzag, zigzagging,
zilch, zinc, zing, zip, zircon, zirconium, zloty, zodiac, zodiacal,
zombie, zone, zoo, zoology, zoom, zounds, z's, zucchini, zygote
Or you can get all those words that start with either z or Z by using the IgnoreCase
option to StringMatchQ.
In[33]:= Select words, StringMatchQ #, "z ", IgnoreCase True &
Out[33]= z, Zachary, zag, zagging, Zagreb, Zaire, Zambia, Zan, Zanzibar,
zap, zazen, zeal, Zealand, zealot, zealous, zebra, Zeiss,
Zellerbach, Zen, zenith, zero, zeroes, zeroth, zest, zesty,
zeta, Zeus, Ziegler, zig, zigging, zigzag, zigzagging, zilch,
Zimmerman, zinc, zing, Zion, zip, zircon, zirconium, zloty, zodiac,
zodiacal, Zoe, Zomba, zombie, zone, zoo, zoology, zoom, Zorn,
Zoroaster, Zoroastrian, zounds, z's, zucchini, Zurich, zygote
Or, using the new (in Version 5.1) Pick function:
In[34]:= Pick words, StringMatchQ words, "z" __
Out[34]= zag, zagging, zap, zazen, zeal, zealot, zealous, zebra, zenith, zero,
zeroes, zeroth, zest, zesty, zeta, zig, zigging, zigzag, zigzagging,
zilch, zinc, zing, zip, zircon, zirconium, zloty, zodiac, zodiacal,
zombie, zone, zoo, zoology, zoom, zounds, z's, zucchini, zygote
454 An Introduction to Programming with Mathematica
10. Several modifications to the solution to Exercise 9 are needed. First, we must choose only
those words with string length greater than or equal to the string length of the second argu-
ment to WordsStartingWith. Secondly, from this modified list, we choose those words
whose first several characters match the string we are working with.
In[35]:= Clear WordsStartingWith
In[36]:= WordsStartingWith lis_, str_ : Module lis2 ,
lis2 Select lis, StringLength # StringLength str & ;
Select lis2, StringTake #, StringLength str str &
In[37]:= WordsStartingWith words, "zoo"
Out[37]= zoo, zoology, zoom
Or, using StringMatchQ from Version 5.1, you have to join the string with the wildcard
character using ~~.
In[38]:= Clear WordsStartingWith
In[39]:= WordsStartingWith lis_List, str_String :
Select lis, StringMatchQ #, str " " &
In[40]:= WordsStartingWith words, "zoo"
Out[40]= zoo, zoology, zoom
Or, using the new (in Version 5.1) Pick function (note the need for the triple-blank here):
In[41]:= Pick words, StringMatchQ words, "zoo" ___
Out[41]= zoo, zoology, zoom
11. Using Fold, this pure function requires two arguments. The key is to start with initial value 0.
In[42]:= Horner list_List, base_ : Fold base #1 #2 &, 0, list ;
In[43]:= Horner a, b, c, d, e , x
Out[43]= e x d x c x b a x
In[44]:= Expand %
Out[44]= e d x c x
2
b x
3
a x
4
Solutions to exercises 455
4.7 One-liners
1. If we map the Mod function with base 2 over a list, it will return 1 for every odd element and 0
for every even element.
In[1]:= Map Mod #, 2 & , 1, 1, 0, 2, 1
Out[1]= 1, 1, 0, 0, 1
Taking two lists, if we add them element-wise, we then need to select those that pass the mod
test above.
In[2]:= l1 1, 0, 0, 1, 1 ;
l2 0, 1, 0, 1, 0 ;
In[4]:= lis l1 l2
Out[4]= 1, 1, 0, 2, 1
In[5]:= Select lis, Mod #, 2 1 &
Out[5]= 1, 1, 1
And finally, we need to know how many elements are in this last list.
In[6]:= Length %
Out[6]= 3
In[7]:= HammingDistance3 lis1_, lis2_ :
Length Select lis1 lis2, Mod #, 2 1 &
Actually this could have been done more cleanly by using the predicate OddQ.
In[8]:= HammingDistance4 lis1_, lis2_ :
Length Select lis1 lis2, OddQ
2. Using Total, which simply gives the sum of the elements in a list, Hamming distance can be
computed as follows:
In[9]:= HammingDistance5 lis1_, lis2_ : Total Mod lis1 lis2, 2
In[10]:= HammingDistance5 l1, l2
Out[10]= 3
Some timing tests show that the implementation with Total is quite a bit more efficient than
the previous versions.
In[11]:= data1 Table Random Integer , 10
6
;
456 An Introduction to Programming with Mathematica
In[12]:= data2 Table Random Integer , 10
6
;
In[13]:= Timing HammingDistance5 data1, data2
Out[13]= 0.06 Second, 499016
In[14]:= Timing HammingDistance4 data1, data2
Out[14]= 0.691 Second, 499016
In[15]:= Timing HammingDistance3 data1, data2
Out[15]= 2.514 Second, 499016
3. a.
In[16]:= frequencies lis_ : Module pair ,
pair x_ : x, Count lis, x ;
Map pair, Union lis
In[17]:= frequencies a, a, b, b, b, a, c, c
Out[17]= a, 3 , b, 3 , c, 2
b.
In[18]:= split1 lis_, parts_ : Module lis1, lis2 ,
lis1 y_, z_ : Take lis, y, z ;
lis2 x_ : Inner lis1, Drop x, 1 1, Rest x , List ;
lis2 FoldList Plus, 0, parts
In[19]:= split1 Range 10 , 2, 5, 0, 3
Out[19]= 1, 2 , 3, 4, 5, 6, 7 , , 8, 9, 10
In[20]:= split2 lis_, parts_ : Module lis1 ,
lis1 x_ : Take lis, x 1, 0 ;
Map lis1, Partition FoldList Plus, 0, parts , 2, 1
In[21]:= split2 Range 10 , 2, 5, 0, 3
Out[21]= 1, 2 , 3, 4, 5, 6, 7 , , 8, 9, 10
c.
In[22]:= lotto1 lis_, n_ : Module lis1, lis2, lis3 , lis1 x_ :
Flatten Rest MapThread Complement, RotateRight x , x , 1 ;
lis2 y_ : Delete y, Random Integer, 1, Length y ;
lis3 z_ : NestList lis2, z, n ;
lis1 lis3 lis
Solutions to exercises 457
In[23]:= lotto1 Range 10 , 5
Out[23]= 2, 5, 4, 6, 8
In[24]:= lotto2 lis_, n_ : Take Transpose
Sort Transpose Table Random , Length lis , lis 2 , n
In[25]:= lotto2 Range 10 , 5
Out[25]= 2, 5, 1, 8, 7
4.
In[26]:= Timing lotto1 Range 50000 , 3 ; , Timing lotto2 Range 50000 , 3 ;
Out[26]= 0.09 Second, Null , 0.421 Second, Null
In[27]:= Timing lotto1 Range 50000 , 60 ; , Timing lotto2 Range 50000 , 60 ;
Out[27]= 1.362 Second, Null , 0.42 Second, Null
5. Here are the list of coins.
In[28]:= coins p, p, q, n, d, d, p, q, q, p
Out[28]= p, p, q, n, d, d, p, q, q, p
In[29]:= pocketChange2 x_ :
Dot Map Count x, # & , p, n, d, q , 1, 5, 10, 25
In[30]:= pocketChange2 coins
Out[30]= 104
In[31]:= pocketChange3 x_ :
Inner Times, Map Count x, # & , p, n, d, q , 1, 5, 10, 25 , Plus
In[32]:= pocketChange3 coins
Out[32]= 104
6.
In[33]:= makeChange x_ : Module coins 25, 10, 5, 1 ,
Quotient FoldList Mod, x, Drop coins, 1 , coins
In[34]:= makeChange 119
Out[34]= 4, 1, 1, 4
458 An Introduction to Programming with Mathematica
7.
In[35]:= offLattice n_ :
Map Sin # , Cos # & , Table Random Real, 0, 2 , n
In[36]:= offLattice n_ : Module step ,
step x_ : Sin x , Cos x ;
Map step, Table Random Real, 0, 2 , n
In[37]:= offLattice 3
Out[37]= 0.194181, 0.980966 , 0.956556, 0.291548 , 0.431374, 0.902173
8. First, notice what FromDigits does.
In[38]:= ?FromDigits
FromDigits list constructs an integer
from the list of its decimal digits. FromDigits
list, b takes the digits to be given in base b. More
We use With to create a local constant d, as this expression never changes throughout the
body of the function.
In[39]:= convertToDate2 num_ : With d IntegerDigits num ,
FromDigits Take d, 4 ,
FromDigits Take d, 5, 6 ,
FromDigits Take d, 7, 8
In[40]:= convertToDate2 20030515
Out[40]= 2003, 5, 15
5 Procedural programming
5.2 Loops and iteration
1. Using a compound expression inside the Do function, this computes the next approximations
of both square roots each time through the loop.
In[1]:= next fun_, x_ : N x
fun x
fun x
Solutions to exercises 459
In[2]:= a 50;
b 60;
Do
a next #
2
50 &, a ;
b next #
2
60 &, b ,
10
In[5]:= a, b
Out[5]= 7.07107, 7.74597
2. Notice that to compute the square root of a number r, we need to iterate the following
expression.
In[6]:= fun x_ : x
2
r;
Simplify x
fun x
fun' x
Out[7]=
r x
2
2 x
This can be written as a pure function, with a second argument giving the initial guess. Here
we iterate ten times.
In[8]:= nestSqrt r_, init_ : Nest
r #
2
2 #
&, N init , 10
In[9]:= nestSqrt 50, 10
Out[9]= 7.07107
3. We need to place the two expressions that were in the body of the Do into a list. Try copying
the body of the Do exactly as above and see what happens.
In[10]:= next fun_, x_ : N x
fun x
fun x
In[11]:= a 50;
b 60;
Table
a next #
2
50 &, a ,
b next #
2
60 &, b ,
10
Out[13]= 25.5, 30.5 , 13.7304, 16.2336 , 8.68597, 9.96482 , 7.22119, 7.993 ,
7.07263, 7.74978 , 7.07107, 7.74597 , 7.07107, 7.74597 ,
7.07107, 7.74597 , 7.07107, 7.74597 , 7.07107, 7.74597
460 An Introduction to Programming with Mathematica
To mimic the solution to this problem obtained with the Do loop, we need to extract the last
set of values obtained.
In[14]:= Last %
Out[14]= 7.07107, 7.74597
4. Note that this version of the Fibonacci function is much more efficient than the simple
recursive version, and is closer to the version that uses dynamic programming.
In[15]:= fib n_ : Module prev 0, this 1, next ,
Do next prev this;
prev this;
this next,
n ;
prev
In[16]:= Table fib i , i, 1, 10
Out[16]= 1, 1, 2, 3, 5, 8, 13, 21, 34, 55
Actually, this code can be simplified a bit by using parallel assignments.
In[17]:= fib2 n_ : Module f1 0, f2 1 ,
Do f1, f2 f2, f1 f2 ,
n 1 ;
f2
In[18]:= Table fib2 i , i, 1, 10
Out[18]= 1, 1, 2, 3, 5, 8, 13, 21, 34, 55
Both of these implementations are quite fast and avoid the deep recursion of the classical
definition.
In[19]:= Timing fib 100000 ; , Timing fib2 100000 ;
Out[19]= 0.781 Second, Null , 0.631 Second, Null
5. We compute the derivative df inside the Module and then use that throughout the body of
the function.
In[20]:= Clear findRoot
In[21]:= findRoot fun_, init_, _ :
Module xi init, funxi fun init , df fun' , While Abs funxi ,
xi N xi
funxi
df xi
;
funxi fun xi ;
xi
Solutions to exercises 461
In[22]:= findRoot f, 50, 0.0001
Out[22]= 50
6. The variable b is the current approximation, and the variable a is the previous approximation.
In[23]:= findRoot fun_, init_, _ : Module a init, b fun init ,
While Abs b a ,
a b;
b N b
fun b
fun b
;
b
In[24]:= f x_ : x
2
50
In[25]:= findRoot f, 10, .001
Out[25]= 7.07107
7. This solution is based on the solution to Exercise 5 above.
In[26]:= findRootList fun_, init_, _ : Module a init, b, solns init ,
b N a
fun a
fun a
;
While Abs b a ,
a b;
b N b
fun b
fun b
;
solns Join solns, a ;
Join solns, b
In[27]:= f x_ : x
2
50
In[28]:= findRootList f, 50, 10
6
Out[28]= 50, 25.5, 13.7304, 8.68597, 7.22119, 7.07263, 7.07107, 7.07107
8. We go back to a previous version of findRoot and add multiple initial values.
In[29]:= findRootList fun_, inits_, _ : Module a inits ,
While Min Abs Map fun, a ,
a Map N #
fun #
fun #
&, a ;
Select a, Min Abs Map fun, a Abs fun # &
In[30]:= findRootList #
2
50 &, 25, 50, 75, 100 , .001
Out[30]= 7.07107
462 An Introduction to Programming with Mathematica
9.
In[31]:= bisect f_, a_, b_, _ : Module
low Min a, b , high Max a, b , mid N
a b
2
, fofMid N f
a b
2
,
While Abs fofMid ,
If fofMid 0, low mid, high mid ;
mid N
low high
2
;
fofMid N f mid ;
mid
In[32]:= f x_ : x
2
2
bisect f, 0, 2, .001
Out[33]= 1.41406
10. Here is a direct implementation of the Euclidean algorithm.
In[34]:= gcd m_, n_ : Module a m, b n, tmpa ,
While b 0,
tmpa a;
a b;
b Mod tmpa, b ;
a
In[35]:= m 12782;
n 5531207;
gcd m, n
Out[37]= 11
We can avoid the need for the temporary variable tmpa by performing a parallel assignment as
in the following function. This results in a much cleaner implementation.
In[38]:= gcd m_, n_ : Module a m, b n ,
While b 0, a, b b, Mod a, b ;
a
In[39]:= m 12782;
n 5531207;
gcd m, n
Out[41]= 11
11.
a. Create a list rvec of 0s, then use a Do loop to set rvec[[i]] to vec[[n i]], where n is
the length of vec.
Solutions to exercises 463
In[42]:= Clear reverse, a, b, c, d, e
In[43]:= reverse vec_ : Module vecA Table 0, Length vec ,
Do vecA i vec Length vec i 1 ,
i, 1, Length vec ;
vecA
In[44]:= reverse a, b, c, d, e
Out[44]= e, d, c, b, a
In[45]:= reverseStruc vec_ : Module vecA Table 0, len Length vec ,
Table vecA i vec len i 1 , i, len
In[46]:= reverseStruc a, b, c, d, e
Out[46]= e, d, c, b, a
b. The key to this problem is to use the Mod operator to compute the target address for any
item fromvec. That is, the element vec[i] must move to, roughly speaking, position n i
mod Length[vec]. The roughly speaking is due to the fact that the Mod operator returns
values in the range 0, , Length vec 1, whereas vectors are indexed by values
1, , Length vec . This causes a little trickiness in this problem.
In[47]:= rotateRight vec_, n_ : Module vecA Table 0, Length vec ,
Do vecA 1 Mod n i 1, Length vec vec i , i, 1, Length vec ;
vecA
In[48]:= rotateRight a, b, c, d, e , 2
Out[48]= d, e, a, b, c
In[49]:= rotateRightStruc vec_, n_ :
Module vecA Table 0, len Length vec ,
Table vecA 1 Mod n i 1, len vec i , i, len ;
vecA
In[50]:= rotateRightStruc a, b, c, d, e , 3
Out[50]= c, d, e, a, b
c. Iterate over the rows of mat, setting row i to the result of calling rotateRight.
In[51]:= rotateRows mat_ : Module matA Table 0, len Length mat ,
Do matA i rotateRight mat i , i ,
i, 1, len ;
matA
464 An Introduction to Programming with Mathematica
In[52]:= rotateRows a, b, c , d, e, f , g, h, k
Out[52]= c, a, b , e, f, d , g, h, k
d.
In[53]:= rotateRowsByS mat_, S_ : Module matA Table 0, Length mat ,
Do matA i rotateRight mat i , S i ,
i, 1, Length mat ;
matA
In[54]:= rotateRowsByS a, b, c , d, e, f , g, h, k , 1, 2, 3
Out[54]= c, a, b , e, f, d , g, h, k
e. Create a list lisC of correct length, then iterate over lisA and lisB, moving lisA[[i]]
to lisC whenever lisB[[i]] is True. The position in lisC that receives this value is
not necessarily i; we use the variable last to keep track of the next position in lisC that
will receive a value fromlisA.
In[55]:= compress lisA_, lisB_ :
Module lisC Table 0, Count lisB, True , last 1 ,
Do If lisB i , lisC last lisA i ;
last last 1,
Null ,
i, 1, Length lisB ;
lisC
In[56]:= compress a, b, c, d, e , True, True, False, False, True
Out[56]= a, b, e
5.3 Flow control
1. Here are the conditional definitions.
In[1]:= signum1 x_ ; x 0 : 1
signum1 x_ ; x 0 : 1
signum1 0 : 0
In[4]:= Map signum1, 2, 0, 1
Out[4]= 1, 0, 1
Here is the signum function defined using If.
In[5]:= signum2 x_ : If x 0, 1, If x 0, 0, 1
Solutions to exercises 465
In[6]:= Map signum2, 2, 0, 1
Out[6]= 1, 0, 1
Here is the signum function defined using Which.
In[7]:= signum3 x_ : Which x 0, 1, x 0, 0, True, 1
In[8]:= Map signum3, 2, 0, 1
Out[8]= 1, 0, 1
Finally, here is the signum function defined using Piecewise.
In[9]:= Piecewise 1, x 0 , 1, x 0 , 0, x 0
Out[9]=
1 x 0
1 x 0
2.
In[10]:= signum1 x_ ; x 0 : 1
signum1 x_ ; x 0 : 1
signum1 0 : 0
signum1 0.0 : 0
In[14]:= Map signum1, 2, 0, 2
Out[14]= 1, 0, 1
In[15]:= signum2 x_ : If x 0, 1, If x 0, 1, 0
In[16]:= Map signum2, 2, 0, 2
Out[16]= 1, 0, 1
In[17]:= signum3 x_ : Which x 0, 1, x 0, 1, True, 0
In[18]:= Map signum3, 2, 0, 2
Out[18]= 1, 0, 1
3.
In[19]:= applyChar[{"+", nums__}] := Apply[Plus, {nums}]
applyChar[{"-", nums__}] := Apply[Minus, {nums}]
applyChar[{"*", nums__}] := Apply[Times, {nums}]
applyChar[{"/", nums__}] := Apply[Divide, {nums}]
applyChar[_] := Print["Bad argument to applyChar"];
466 An Introduction to Programming with Mathematica
4.
a.
In[24]:= doublePos lis_ : Map If # 0, 2 #, # &, lis
b.
In[25]:= remove3Repetitions lis_ : Fold
If Length #1 2 && #2 #1 1 #1 2 , #1, Join #1, #2 &, , lis
c.
In[26]:= positiveSum L_ : Fold If #1 #2 0, 0, #1 #2 &, 0, L
5. First we define the auxiliary function using conditional statements.
In[27]:= collatz n_ :
n
2
; EvenQ n
In[28]:= collatz n_ : 3 n 1 ; OddQ n
Then iterate Collatz, starting with n, and continue while n is not equal to 1.
In[29]:= CollatzSequence n_ : NestWhileList collatz, n, # 1 &
In[30]:= CollatzSequence 13
Out[30]= 13, 40, 20, 10, 5, 16, 8, 4, 2, 1
5.4 Examples
1. Here is the gcd function implemented using an If structure.
In[1]:= Clear gcd
In[2]:= gcd m_Integer, n_Integer : If m 0, gcd Mod n, m , m , gcd m, n n
In[3]:= m 12782;
n 5531207;
gcd m, n
Out[5]= 11
Solutions to exercises 467
2. This is a direct implementation using Piecewise.
In[6]:= Piecewise 0, x 0 && y 0 , 1, y 0 , 2, x 0 ,
1, x 0 && y 0 , 2, x 0 && y 0 , 3, x 0 && y 0 , 4
Out[6]=
0 x 0 && y 0
1 y 0
2 x 0
1 x 0 && y 0
2 x 0 && y 0
3 x 0 && y 0
4 True
In[7]:= pointLocPW x_, y_ :
Piecewise 0, x 0 && y 0 , 1, y 0 , 2, x 0 ,
1, x 0 && y 0 , 2, x 0 && y 0 , 3, x 0 && y 0 , 4
In[8]:= Map pointLocPW, 0, 0 , 4, 0 , 0, 1.3 ,
2, 4 , 2, 4 , 2, 4 , 2, 4 , 2, 0 , 3, 4
Out[8]= 0, 1, 2, 1, 2, 3, 4, 1, 4
3.
In[9]:= pointLoc 0, 0 : 0
pointLoc x_, 0 : 1
pointLoc 0, y_ : 2
pointLoc x_, y_ : If x 0, 2, 1 ; y 0
pointLoc x_, y_ : If x 0, 3, 4
pointLoc x_, y_, z_ : If x 0, 2, 1 ; y 0 && z 0
pointLoc x_, y_, z_ : If x 0, 3, 4 ; y 0 && z 0
pointLoc x_, y_, z_ : If x 0, 6, 5 ; y 0 && z 0
pointLoc x_, y_, z_ : If x 0, 7, 8 ; y 0 && z 0
In[18]:= Map pointLoc, 2, 0 , 3, 4
Out[18]= 1, 4
6 Rule-based programming
6.2 Patterns
1. Using the FullForm of the expression, we can find many pattern matches.
In[1]:= FullForm x
3
y z
Out[1]//FullForm=
Plus Power x, 3 , Times y, z
468 An Introduction to Programming with Mathematica
In[2]:= MatchQ x
3
y z, _Plus
Out[2]= True
In[3]:= MatchQ x
3
y z, _Power _Times
Out[3]= True
There are many more possible matches, including the trivial one.
In[4]:= MatchQ x
3
y z, _
Out[4]= True
2. First look at the FullForm of this expression.
In[5]:= FullForm 5, erina, "give me a break"
Out[5]//FullForm=
List 5, erina, "give me a break"
In[6]:= MatchQ 5, erina, "give me a break" , _List
Out[6]= True
In[7]:= MatchQ 5, erina, "give me a break" , _Integer, _Symbol, _String
Out[7]= True
3. Again, the FullForm should help to guide you.
In[8]:= FullForm 4, a, b , "g"
Out[8]//FullForm=
List 4, List a, b , "g"
In[9]:= MatchQ 4, a, b , "g" , x_List ; Length x 3
Out[9]= True
In[10]:= MatchQ 4, a, b , "g" , _List? Length #1 3 &
Out[10]= True
In[11]:= MatchQ 4, a, b , "g" , _, y_, _ ; y 0 List
Out[11]= True
In[12]:= MatchQ 4, a, b , "g" , x_, y_, z_ ; AtomQ z
Out[12]= True
Solutions to exercises 469
In[13]:= MatchQ 4, a, b , "g" , x_, _, _ ; EvenQ x
Out[13]= True
4. Here is the original solution as from Chapter 5, but, in this case, we check that both m and n
have head Integer.
In[14]:= gcd m_Integer, n_Integer : Module a m, b n ,
While b 0, a, b b, Mod a, b ;
a
In[15]:= gcd 39874, 2868878
Out[15]= 2
5. Here is the function FindSubsequence as given in the text.
In[16]:= FindSubsequence lis_List, subseq_List : Module p ,
p Partition lis, Length subseq , 1 ;
Position p, Flatten ___, subseq, ___
This creates another rule associated with FindSubsequence that simply takes each integer
argument, converts them to lists of integer digits, and then passes that off to the rule above.
In[17]:= FindSubsequence n_Integer, subseq_Integer :
Module nlist IntegerDigits n , sublist IntegerDigits subseq ,
FindSubsequence nlist, sublist
We create the list of the first 100,000 digits of .
In[18]:= pi FromDigits RealDigits N Pi, 10
5
3 1 ;
This show that the subsequence 1415 occurs seven times at the following locations in the digit
expansion of .
In[19]:= FindSubsequence pi, 1415
Out[19]= 1 , 6955 , 29136 , 45234 , 79687 , 85880 , 88009
6. The Collatz function has a direct implementation based on its definition.
In[20]:= Collatz n_?OddQ : 3 n 1
In[21]:= Collatz n_?EvenQ :
n
2
In[22]:= Collatz 4.3
Out[22]= Collatz 4.3
470 An Introduction to Programming with Mathematica
Here we iterate the Collatz function 111 times starting with an initial value of 27.
In[23]:= NestList Collatz, 27, 111
Out[23]= 27, 82, 41, 124, 62, 31, 94, 47, 142, 71, 214, 107, 322, 161, 484, 242,
121, 364, 182, 91, 274, 137, 412, 206, 103, 310, 155, 466, 233, 700,
350, 175, 526, 263, 790, 395, 1186, 593, 1780, 890, 445, 1336, 668,
334, 167, 502, 251, 754, 377, 1132, 566, 283, 850, 425, 1276, 638,
319, 958, 479, 1438, 719, 2158, 1079, 3238, 1619, 4858, 2429, 7288,
3644, 1822, 911, 2734, 1367, 4102, 2051, 6154, 3077, 9232, 4616, 2308,
1154, 577, 1732, 866, 433, 1300, 650, 325, 976, 488, 244, 122, 61,
184, 92, 46, 23, 70, 35, 106, 53, 160, 80, 40, 20, 10, 5, 16, 8, 4, 2, 1
7. Here again is the Collatz function, but this time using a condition on the right-hand side of
the definition.
In[24]:= Clear Collatz
In[25]:= Collatz n_ : 3 n 1 ; OddQ n && Positive n
In[26]:= Collatz n_ :
n
2
; EvenQ n && Positive n
In[27]:= Collatz 4.3
Out[27]= Collatz 4.3
In[28]:= Collatz 3
Out[28]= Collatz 3
In[29]:= NestList Collatz, 27, 111
Out[29]= 27, 82, 41, 124, 62, 31, 94, 47, 142, 71, 214, 107, 322, 161, 484, 242,
121, 364, 182, 91, 274, 137, 412, 206, 103, 310, 155, 466, 233, 700,
350, 175, 526, 263, 790, 395, 1186, 593, 1780, 890, 445, 1336, 668,
334, 167, 502, 251, 754, 377, 1132, 566, 283, 850, 425, 1276, 638,
319, 958, 479, 1438, 719, 2158, 1079, 3238, 1619, 4858, 2429, 7288,
3644, 1822, 911, 2734, 1367, 4102, 2051, 6154, 3077, 9232, 4616, 2308,
1154, 577, 1732, 866, 433, 1300, 650, 325, 976, 488, 244, 122, 61,
184, 92, 46, 23, 70, 35, 106, 53, 160, 80, 40, 20, 10, 5, 16, 8, 4, 2, 1
8. Using alternatives, this gives the definition for real, integer, or rational arguments.
In[30]:= abs x_Real x_Integer x_Rational : If x 0, x, x
Here is the definition for complex arguments.
In[31]:= abs x_Complex : Re x
2
Im x
2
It is probably a good idea to also add a definition for symbolic arguments.
In[32]:= abs x_Symbol : Abs x
Solutions to exercises 471
In[33]:= Map abs, 3, 3 4 I,
4
5
, a
Out[33]= 3, 5,
4
5
, Abs a
9. We first have to consider the base cases. Given a list with no elements, swapTwo should
return the empty list. And, given a list with one element, swapping should give that one
element back.
In[34]:= swapTwo :
swapTwo x_ : x
Now, we use the triple-blank to indicate that r could be a sequence of 0 or more elements.
In[36]:= swapTwo x_, y_, r___ : y, x, r
In[37]:= Map swapTwo, , a , a, b, c, d
Out[37]= , a , b, a, c, d
Notice in this second definition for swapTwo that the second clause covers both the situation
where the argument is the empty list and when it contains only one element.
In[38]:= swapTwo2 x_, y_, r___ : y, x, r
swapTwo2 x_ : x
In[40]:= Map swapTwo2, , a , a, b, c, d
Out[40]= , a , b, a, c, d
10. This one requires the triple blank.
In[41]:= f x1_Integer, ___ , 1 : x1 1
f x_Integer, y_ : x y
11. Here are two sample lists.
In[43]:= l1 1, 0, 0, 1, 1 ;
l2 0, 1, 0, 1, 0 ;
First we pair them up.
In[45]:= ll Transpose l1, l2
Out[45]= 1, 0 , 0, 1 , 0, 0 , 1, 1 , 1, 0
Here is the conditional pattern that matches any pair where the two elements are not identical.
In[46]:= Cases ll, p_, q_ ; p q
Out[46]= 1, 0 , 0, 1 , 1, 0
472 An Introduction to Programming with Mathematica
The Hamming distance is the number of such non-identical pairs.
In[47]:= Length %
Out[47]= 3
Finally, here is a function that puts this all together.
In[48]:= HammingDistance lis1_List, lis2_List :
Length Cases Transpose lis1, lis2 , p_, q_ ; p q
In[49]:= HammingDistance l1, l2
Out[49]= 3
The running times of this version of HammingDistance are comparable with those from
Chapter 4, where we used bit operators.
In[50]:= HammingDistance2 lis1_, lis2_ : Apply Plus,
Apply BitXor, Transpose lis1, lis2 , 1
In[51]:= data1 Table Random Integer , 10
6
;
In[52]:= data2 Table Random Integer , 10
6
;
In[53]:= Timing HammingDistance data1, data2
Out[53]= 2.905 Second, 500168
In[54]:= Timing HammingDistance2 data1, data2
Out[54]= 1.592 Second, 500168
6.3 Transformation rules
1. The pattern matched function is slower because it repeatedly applies transformation rules.
In[1]:= maxima x_ : Union Rest FoldList Max, , x
In[2]:= maximaR x_List : x . a___, b_, c___, d_, e___ ; d b a, b, c, e
Solutions to exercises 473
In[3]:= Trace maxima 3, 5, 2, 6, 1, 8, 4, 9, 7
Out[3]= maxima 3, 5, 2, 6, 1, 8, 4, 9, 7 ,
Union Rest FoldList Max, , 3, 5, 2, 6, 1, 8, 4, 9, 7 ,
, , , , FoldList Max, , 3, 5, 2, 6, 1, 8, 4, 9, 7 ,
Max , 3 , Max 3, , 3 , Max 3, 5 , 5 ,
Max 5, 2 , Max 2, 5 , 5 , Max 5, 6 , 6 , Max 6, 1 , Max 1, 6 , 6 ,
Max 6, 8 , 8 , Max 8, 4 , Max 4, 8 , 8 , Max 8, 9 , 9 ,
Max 9, 7 , Max 7, 9 , 9 , , 3, 5, 5, 6, 6, 8, 8, 9, 9 ,
Rest , 3, 5, 5, 6, 6, 8, 8, 9, 9 , 3, 5, 5, 6, 6, 8, 8, 9, 9 ,
Union 3, 5, 5, 6, 6, 8, 8, 9, 9 , 3, 5, 6, 8, 9
In[4]:= Trace maximaR 3, 5, 2, 6, 1, 8, 4, 9, 7
Out[4]= maximaR 3, 5, 2, 6, 1, 8, 4, 9, 7 , 3, 5, 2, 6, 1, 8, 4, 9, 7 .
a___, b_, c___, d_, e___ ; d b a, b, c, e ,
a___, b_, c___, d_, e___ ; d b a, b, c, e ,
a___, b_, c___, d_, e___ ; d b a, b, c, e ,
3, 5, 2, 6, 1, 8, 4, 9, 7 .
a___, b_, c___, d_, e___ ; d b a, b, c, e , 5 3, False ,
2 3, True , 5 3, False , 6 3, False , 6 5, False ,
1 3, True , 5 3, False , 6 3, False , 6 5, False ,
8 3, False , 8 5, False , 8 6, False , 4 3, False , 4 5, True ,
5 3, False , 6 3, False , 6 5, False , 8 3, False ,
8 5, False , 8 6, False , 9 3, False , 9 5, False ,
9 6, False , 9 8, False , 7 3, False , 7 5, False ,
7 6, False , 7 8, True , 5 3, False , 6 3, False , 6 5, False ,
8 3, False , 8 5, False , 8 6, False , 9 3, False ,
9 5, False , 9 6, False , 9 8, False , 3, 5, 6, 8, 9
2. The evaluation sequence can be seen directly from the Trace of this compound expression.
In[5]:= Trace y 11; a 9; y 3 . y a
Out[5]= y 11; a 9; y 3 . y a, y 11, 11 , a 9, 9 , y, 11 , 11 3, 14 ,
y, 11 , a, 9 , 11 9, 11 9 , 14 . 11 9, 14 , 14
3. First make sure that a and y have no values associated with them.
In[6]:= Clear a, y
In[7]:= Hold y 11 ;
a 9;
y 3 . y a
Out[9]= 12
474 An Introduction to Programming with Mathematica
4. You need to maintain the left-hand side of the transformation rule unevaluated for purposes of
pattern matching and the right-hand side of the rule unevaluated until the rule is used.
In[10]:= Trace g x_ x . z___ Times z
Out[10]= z___, z___ , Times z , z , z___ z, z___ z , x . z___ z, x ,
g x_ x, x
In[11]:= Clear a, g
In[12]:= g x_ : x . Literal z___ Times z
In[13]:= g a b c
Out[13]= a b c
5. The transformation rule unnests lists within a list.
In[14]:= unNest lis_ : Map # . x__List x & , lis
In[15]:= unNest a, a, a , a , b, b, b , b, b , a, a
Out[15]= a, a, a , a , b, b, b , b, b , a, a
6.
In[16]:= sumList lis_ : First lis . x_, y___ x y
In[17]:= sumList 1, 5, 8, 3, 9, 3
Out[17]= 29
7. The triple blank is required both before and after the variables x and y.
In[18]:= cartesianProduct lis1_, lis2_ :
ReplaceList lis1, lis2 , ___, x_, ___ , ___, y_, ___ x, y
We should also have a rule for the base case.
In[19]:= cartesianProduct :
In[20]:= Clear x, y, z, a, b, c
In[21]:= cartesianProduct a, b, c , x, y, z
Out[21]= a, x , a, y , a, z , b, x , b, y , b, z , c, x , c, y , c, z
In[22]:= cartesianProduct
Out[22]=
Solutions to exercises 475
8. Note that RasterArray and Raster both display an array of values from bottom to top,
hence the need to reverse the argument lis.
In[23]:= CAGraphics lis_List : Module colors ,
colors 1 Hue .2 , 0 Hue .8 ;
Graphics RasterArray Reverse lis . colors
Here is a larger example of rule 30.
In[1]:= ca30 CellularAutomaton 30, 1 , 0 , 400 ;
In[25]:= Show CAGraphics ca30 ;
Note that this can also be accomplished much more cleanly using ArrayPlot (new in
Version 5.1).
In[26]:= ArrayPlot ca30, ColorRules 1 Hue .2 , 0 Hue .8 ;
6.4 Examples
1. This is a simple modification of the code given in the text.
In[1]:= alphabet Map FromCharacterCode, Range 97, 122
Out[1]= a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t, u, v, w, x, y, z
In[2]:= coderules Thread alphabet RotateRight alphabet, 5
Out[2]= a v, b w, c x, d y, e z, f a, g b, h c,
i d, j e, k f, l g, m h, n i, o j, p k, q l,
r m, s n, t o, u p, v q, w r, x s, y t, z u
476 An Introduction to Programming with Mathematica
In[3]:= decoderules Thread alphabet RotateLeft alphabet, 5
Out[3]= a f, b g, c h, d i, e j, f k, g l, h m,
i n, j o, k p, l q, m r, n s, o t, p u, q v,
r w, s x, t y, u z, v a, w b, x c, y d, z e
In[4]:= code str_String : Apply StringJoin, Characters str . coderules
In[5]:= decode str_String : Apply StringJoin, Characters str . decoderules
In[6]:= code "squeamish ossifrage"
Out[6]= nlpzvhdnc jnndamvbz
In[7]:= decode %
Out[7]= squeamish ossifrage
3. This version of matrixPlot requires a list of rules as the second argument.
In[8]:= matrixPlot mat_List, rules_ :
Show Graphics RasterArray Reverse mat . rules ,
AspectRatio Automatic
In[9]:= dat Table Random Integer , 50 , 50 ;
In[10]:= matrixPlot dat, 0 GrayLevel .2 , 1 GrayLevel .6
Out[10]= Graphics
You can plot any rectangular array of values with matrixPlot so long as you specify the rules
for coloring the various elements. For example, the following example generates 100 steps in
the evolution of the rule 30 cellular automaton, starting with a single 1 cell and surrounded by
0s.
In[11]:= ca30 CellularAutomaton 30, 1 , 0 , 100 ;
Solutions to exercises 477
In[12]:= matrixPlot ca30, 1 Hue .2 , 0 Hue .6 ;
To get matrixPlot to produce similar output to the new ArrayPlot, you need to make a
few changes to the Frame and FrameTicks options.
In[13]:= matrixPlot mat_List, rules_ :
Show Graphics RasterArray Reverse mat . rules ,
AspectRatio Automatic, Frame True, FrameTicks False
In[14]:= matrixPlot ca30, 1 Hue .2 , 0 Hue .6 ;
In[15]:= ArrayPlot ca30, ColorRules 1 Hue .2 , 0 Hue .6 ;
4. Here is the plot of the sine function.
In[13]:= splot Plot Sin x , x, 2 , 2 ;
6 4 2 2 4 6
1
0.5
0.5
1
This replacement rule interchanges each ordered pair of numbers.
478 An Introduction to Programming with Mathematica
In[14]:= Show splot . x_?NumberQ, y_?NumberQ y, x ;
1 0.5 0.5 1
6
4
2
2
4
6
Although this particular example may have worked without the argument checking (_?Num
berQ), it is a good idea to include it so that pairs of arbitrary expressions are not pattern
matched here. We only want to interchange pairs of numbers, not pairs of options or other
expressions that might be present in the underlying expression representing the graphic.
6. Using the standard rotation matrix, each point is taken to its image under the rotation transfor-
mation. Notice that this function first checks that its first argument is in fact a graphics object
via pattern matching.
In[15]:= rotatePlot p_Graphics, theta_ : Show p . x_?NumberQ, y_?NumberQ
x, y . Cos theta , Sin theta , Sin theta , Cos theta
In[16]:= plot1 Plot Sin x , x, 0, 2 ;
1 2 3 4 5 6
1
0.5
0.5
1
In[17]:= rotatePlot[plot1, Pi];
6 5 4 3 2 1
1
0.5
0.5
1
Solutions to exercises 479
7 Recursion
7.1 Fibonacci numbers
1.
a. This is a straightforward recursion, multiplying the previous two values to get the next.
In[1]:= a 1 : 2
a 2 : 3
a i_ : a i 1 a i 2
In[4]:= Table a i , i, 1, 8
Out[4]= 2, 3, 6, 18, 108, 1944, 209952, 408146688
b. The sequence is obtained by taking the difference of the previous two values.
In[5]:= b 1 : 0
b 2 : 1
b i_ : b i 2 b i 1
In[8]:= Table b i , i, 1, 9
Out[8]= 0, 1, 1, 2, 3, 5, 8, 13, 21
c. Here we add the previous three values.
In[9]:= c 1 : 0
c 2 : 1
c 3 : 2
c i_ : c i 3 c i 2 c i 1
In[13]:= Table c i , i, 1, 9
Out[13]= 0, 1, 2, 3, 6, 11, 20, 37, 68
2. It is important to get the two base cases right here.
In[14]:= FA 1 : 0
FA 2 : 0
FA i_ : FA i 2 FA i 1 1
In[17]:= Map FA, Range 8
Out[17]= 0, 0, 1, 2, 4, 7, 12, 20
480 An Introduction to Programming with Mathematica
7.2 List functions
1.
In[1]:= sumOddElements : 0
sumOddElements x_, y___ :
x sumOddElements y ; IntegerQ x && OddQ x
sumOddElements x_, y___ : sumOddElements y
In[4]:= sumOddElements 2, 3, 5, 6, 7, 9, 12, 13
Out[4]= 37
2.
In[5]:= sumEveryOtherElement : 0
sumEveryOtherElement x_ : x
sumEveryOtherElement x_, y_, r___ : x sumEveryOtherElement r
In[8]:= sumEveryOtherElement 1, 2, 3, 4, 5, 6, 7, 8, 9
Out[8]= 25
3.
In[9]:= addTriples , , :
addTriples x1_, y1___ , x2_, y2___ , x3_, y3___ :
Join x1 x2 x3 , addTriples y1 , y2 , y3
In[11]:= addTriples w
1
, x
1
, y
1
, z
1
, w
2
, x
2
, y
2
, z
2
, w
3
, x
3
, y
3
, z
3
Out[11]= w
1
w
2
w
3
, x
1
x
2
x
3
, y
1
y
2
y
3
, z
1
z
2
z
3
4.
In[12]:= multAllPairs :
multAllPairs _ :
multAllPairs x_, y_, r___ : Join x y , multAllPairs y, r
In[15]:= multAllPairs 3, 9, 17, 2, 6, 60
Out[15]= 27, 153, 34, 12, 360
5.
In[16]:= maxPairs , :
maxPairs x_, r___ , y_, s___ : Join Max x, y , maxPairs r , s
In[18]:= maxPairs 1, 2, 4 , 2, 7, 2
Out[18]= 2, 7, 4
Solutions to exercises 481
6.
In[19]:= interleave , :
interleave x_, r___ , y_, s___ : Join x, y , interleave r , s
In[21]:= interleave a, b, c , x, y, z
Out[21]= a, x, b, y, c, z
7.3 Thinking recursively: examples
1.
In[1]:= prefixMatch L_, :
prefixMatch , M_ :
prefixMatch x_, r___ , x_, s___ : Join x , prefixMatch r , s
prefixMatch x_, r___ , y_, s___ :
2.
In[5]:= runEncode2 :
runEncode2 x_ : x
runEncode2 x_, r__ : runEncode2 r .
y_, k_ , s___ If x y, x, k 1 , s , x, y, k , s ,
y_, s___ If x y, x, 2 , s , x, y, s
3. Perhaps the most straightforward way to do this is to write an auxiliary function that takes the
output fromrunEncode and produces output such as Split would generate.
In[8]:= runEncode[{}] := {}
runEncode[{x_}] := {{x, 1}}
In[10]:= runEncode x_, res___ : Module R runEncode res , p ,
p First R ;
If x First p ,
Join x, p 2 1 , Rest R ,
Join x, 1 , R
Then our split (named to mimic the built-in Split) simply operates on the output of
runEncode.
In[11]:= sp lis_ : Map Table # 1 , # 2 &, lis
In[12]:= sp 3, 2 , 4, 1 , 2, 5
Out[12]= 3, 3 , 4 , 2, 2, 2, 2, 2
In[13]:= split lis_ : sp runEncode lis
482 An Introduction to Programming with Mathematica
In[14]:= split 9, 9, 9, 9, 9, 4, 3, 3, 3, 3, 5, 5, 5, 5, 5, 5
Out[14]= 9, 9, 9, 9, 9 , 4 , 3, 3, 3, 3 , 5, 5, 5, 5, 5, 5
4.
In[15]:= runEncode :
runEncode x_, r___ : runEncode x, 1, r
runEncode x_, k_, : x, k
runEncode x_, k_, x_, r___ : runEncode x, k 1, r
runEncode x_, k_, y_, r___ : Join x, k , runEncode y, 1, r
5.
In[20]:= maxima :
maxima x_, r___ : maxima x, r
In[22]:= maxima x_, : x
maxima x_, y_, r___ : maxima x, r ; x y
maxima x_, y_, r___ : Join x , maxima y, r
6.
In[25]:= runDecode :
runDecode x_, k_ , r___ : Join Table x, k , runDecode r
7. We will need two sets of rules for the subsets function.
In[27]:= Clear subsets ;
subsets lis_, 0 :
subsets , k_ :
In[30]:= subsets lis_, k_ : Module ksubs subsets Rest lis , k 1 ,
Join Map Join First lis , # & , ksubs , subsets Rest lis , k
In[31]:= subsets Range 5 , 2
Out[31]= 1, 2 , 1, 3 , 1, 4 , 1, 5 , 2, 3 , 2, 4 , 2, 5 , 3, 4 , 3, 5 , 4, 5
The second form simply calls the first.
In[32]:= subsets lis_, k_ : Flatten Map subsets lis, # &, Range 0, k , 1
The second form simply calls the first. This gives all subsets up to length 3.
In[33]:= subsets Range 5 , 3
Out[33]= , 1 , 2 , 3 , 4 , 5 , 1, 2 , 1, 3 , 1, 4 , 1, 5 , 2, 3 ,
2, 4 , 2, 5 , 3, 4 , 3, 5 , 4, 5 , 1, 2, 3 , 1, 2, 4 , 1, 2, 5 ,
1, 3, 4 , 1, 3, 5 , 1, 4, 5 , 2, 3, 4 , 2, 3, 5 , 2, 4, 5 , 3, 4, 5
Solutions to exercises 483
A comparison with the built-in Subsets functions.
In[34]:= Subsets Range 5 , 3
Out[34]= , 1 , 2 , 3 , 4 , 5 , 1, 2 , 1, 3 , 1, 4 , 1, 5 , 2, 3 ,
2, 4 , 2, 5 , 3, 4 , 3, 5 , 4, 5 , 1, 2, 3 , 1, 2, 4 , 1, 2, 5 ,
1, 3, 4 , 1, 3, 5 , 1, 4, 5 , 2, 3, 4 , 2, 3, 5 , 2, 4, 5 , 3, 4, 5
The recursion in this definition of subsets can get quite deep.
In[35]:= Timing subsets Range 1000 , 2 ;
$RecursionLimit::reclim : Recursion depth of 256 exceeded. More
General::stop : Further output of $RecursionLimit::reclim will
be suppressed during this calculation. More
You can temporarily increase the value of $RecursionLimit to let this computation run to
the end.
In[36]:= Timing
Block $RecursionLimit ,
subsets Range 1000 , 2 ;
Out[36]= 44.855 Second, Null
But we can see pretty clearly just how inefficient our recursive approach to this problem is for
large computations by comparing with the built-in Subsets function which is more than two
orders of magnitude faster for sets of this size.
In[37]:= Timing Subsets Range 1000 , 2 ;
Out[37]= 0.14 Second, Null
7.4 Recursion and symbolic computations
1.
In[1]:= ddx c_?NumericQ : 0
ddx x : 1
ddx u_ v_ : ddx u ddx v
ddx u_ v_ : ddx u ddx v
ddx u_ v_ : u ddx v v ddx u
ddx
u_
v_
:
v ddx u u ddx v
v
2
ddx u_
c_?NumericQ
: c u
c 1
ddx u
484 An Introduction to Programming with Mathematica
In[8]:= ddx Sin u_ : Cos u ddx u
ddx Cos u_ : Sin u ddx u
ddx Tan u_ :
1
Cos u
2
ddx u
In[11]:= ddx Sin 2 x Cos 3 x
Out[11]= 2 Cos 2 x 3 Sin 3 x
In[12]:= ddx Tan 3 x
5
Out[12]= 15 x
4
Sec 3 x
5
2
2.
In[13]:= Clear ddx
In[14]:= ddx c_?NumericQ : 0
ddx x : 1
ddx u_ v_ : ddx u ddx v
ddx u_ v_ : ddx u ddx v
ddx u_ v_ : u ddx v v ddx u
ddx
u_
v_
:
v ddx u u ddx v
v
2
ddx u_
c_?NumericQ
: c u
c 1
ddx u
In[21]:= ddx u_ : 0 ; nox u
In[22]:= nox c_?NumericQ : True
nox x : False
nox y_ : True ; Head y Symbol&& y x
nox u_ v_ : nox u && nox v
nox u_ v_ : nox u && nox v
nox u_ v_ : u nox v && v nox u
nox
u_
v_
: nox u && nox v
nox u_
c_?NumericQ
: nox u
3.
In[30]:= Clear ddx ;
ddx c_?NumericQ, y_ : 0
ddx x_, x_ : 1
ddx y_, x_ : 0 ; FreeQ y, x
ddx u_ v_, x_ : ddx u, x ddx v, x
ddx u_ v_, x_ : ddx u, x ddx v, x
ddx u_ v_, x_ : u ddx v, x v ddx u, x
ddx
u_
v_
, x_ :
v ddx u, x u ddx v, x
v
2
ddx u_
c_?NumericQ
, x_ : c u
c 1
ddx u, x
Solutions to exercises 485
In[39]:= ddx
3 2
,
Out[39]= 2 3
2
In[40]:= ddx
1
3
,
Out[40]=
1 2
3
1
3
2
7.5 Classical examples
1. The solution to this problem also appears in Section 8.5. We will call our new function
solvep (for pivoting).
In[1]:= Clear solve
In[2]:= solvep S_ : Module
S1 pivot S , E1, a12toa1n, x2toxn , x2toxn solvep elimx1 S1 ;
E1 First S1 ;
a12toa1n Drop Rest E1 , 1 ;
Join
Last E1 a12toa1n.x2toxn
First E1
, x2toxn ;
In[3]:= solvep a11_, b1_ :
b1
a11
In[4]:= elimx1 S_ : Map subtractE1 S 1 , # &, Rest S
In[5]:= subtractE1 E1_, Ei_ : Rest Ei
Ei 1
E1 1
Rest E1
In[6]:= pivot Q_ : Module p, ST1, pivotrow , ST1 Transpose Q 1 ;
p Position ST1, x_ ; x 0 ;
If p ,
Print "Matrix is singular" ; Q,
pivotrow p 1 1 ; Join Q pivotrow , Delete Q, pivotrow
In[7]:= solve A_, B_ : solvep Transpose Join Transpose A , B
Here are some test examples.
In[8]:= mat Table Random , 4 , 4
Out[8]= 0.554127, 0.426593, 0.861278, 0.492521 ,
0.572684, 0.477244, 0.690375, 0.88366 ,
0.401935, 0.648486, 0.818292, 0.516009 ,
0.129603, 0.562562, 0.116779, 0.699194
486 An Introduction to Programming with Mathematica
In[9]:= b Table Random , 4
Out[9]= 0.564681, 0.489887, 0.542515, 0.264061
In[10]:= x solve mat, b
Out[10]= 1.59998, 0.96497, 0.502052, 0.611457
In[11]:= mat.x b
Out[11]= 1.11022 10
16
, 1.66533 10
16
, 1.11022 10
16
, 4.44089 10
16
In[12]:= Chop %
Out[12]= 0, 0, 0, 0
2. To compute solveUpper[A,B], first recursively compute solveUpper[A ,B ], where A
is the lower-right square submatrix of A, and B is the Rest of B. This solution gives the values
of x
2
, , x
n
. B[[1]] is equal to the dot product of the top row of A (that is, A[[1]]) and the
vector x
1
, , x
n
(that is, B[[1]]) is equal to A[[1]]*x
1
+ ... + A[[n]]*x
n
. It is easy to
compute x
1
from this formula.
In[13]:= solveUpper ann_ , bn_ :
bn
ann
In[14]:= solveUpper A1_, rA__ , b1_, rB__ :
Module subsoln solveUpper Rest rA , rB ,
Join
b1 Rest A1 .subsoln
First A1
, subsoln
It is easy to show that if you rotate a matrix by 90 degrees, and turn the vector B upside down,
the solution to the resulting system is the same as the solution to the original system, but
turned upside down.
In[15]:= rotateMatrix A_ : Reverse Map Reverse, A
In[16]:= solveLower A_, B_ : Reverse solveUpper rotateMatrix A , Reverse B
3.
In[17]:= LUdecomp1 S_ : Module mults multipliers S 1, 1 , Rest S , Module
Sprime elimx1 mults, Rest S , Module LU LUdecomp1 Sprime ,
expandL mults, LU 1 , expandU First S , LU 2
In[18]:= LUdecomp1 a11_ : 1 , a11
In[19]:= expandU S1_, U_ : Join S1 , Join 0 , #1 & U
In[20]:= expandL mults_, L_ : Transpose expandU Join 1 , mults , Transpose L
Solutions to exercises 487
In[21]:= elimx1 mults_, subS_ :
Table subS i 1 mults i subS 1 , i, 1, Length mults
In[22]:= multipliers S11_, restS_ : Map
#
S11
&, Transpose restS 1
In[23]:= LUdecomp2 S_ : Module soln LUdecomp1 S ,
soln 1 IdentityMatrix Length S soln 2
4.
In[24]:= sumNodes lab_ : lab
sumNodes lab_, lc_, rc_ : lab sumNodes lc sumNodes rc
5.
In[26]:= catNodes lab_ : lab
catNodes lab_, lc_, rc_ :
StringJoin lab, catNodes lc , catNodes rc
6.
In[28]:= balanced t_ : balancedHeight t 2
balancedHeight lab_ : 0, True
In[30]:= balancedHeight lab_, lc_, rc_ :
Module lbh, rbh , lbh balancedHeight lc ;
If lbh 2 , rbh balancedHeight rc ; If rbh 2 && Abs lbh 1 rbh 1 1,
Max lbh 1 , rbh 1 1, True , 0, False , 0, False
7.
In[31]:= listLevel 0, t_ : t 1
listLevel lab_ , n_ :
In[33]:= listLevel lab_, lc_, rc_ , n_ :
Join listLevel lc, n 1 , listLevel rc, n 1
8.
In[34]:= minInTree lab_ : lab
minInTree lab_, subtrees__ :
Sort Join lab , Map minInTree, subtrees 1
In[36]:= height lab_ : 0
height lab_, subtrees__ : 1 Apply Max, Map height, subtrees
In[38]:= printTree t_ : printTree t, 0
In[39]:= printTree lab_ , k_ : printIndented lab, 3 k
printTree lab_, subtrees__ , k_ :
printIndented lab, 3 k ; Map printTree #, k 1 &, subtrees ;
488 An Introduction to Programming with Mathematica
In[41]:= printIndented x_, spaces_ :
Print Apply StringJoin, Table " ", spaces , x
9. We have used a slightly different representation for the list of trees than the one shown in the
chapter. Instead of a nodes label containing a list of characters and a number, it contains a
string and a number. The only reason for this is that it makes the result come out looking like
the tree called Htree (shown in Figure 7.1). Note that the algorithm may give different results
depending upon how it is programmed, since there are arbitrary choices made at each step.
The result of applying our function constructHTree to the initial list of trees shown at the
end of the last section (which we have included here as testlist) is different fromHtree.
In our solution, we solve the problem of finding the two trees of smallest weight by keeping
the list of trees sorted by weight; then we simply always pick the first two.
In[42]:= HTreeSort trees_ : Sort trees, #1 1, 2 #2 1, 2 &
In[43]:= joinHTrees cl_, wt_ , kids___ : cl, kids
joinHTrees cl1_, wt1_ , kids1___ ,
cl2_, wt2_ , kids2___ , trees___ : joinHTrees
HTreeSort cl1 cl2, wt1 wt2 , cl1, kids1 , cl2, kids2 , trees
In[45]:= constructHTree t_ : joinHTrees HTreeSort t
In[46]:= htnode a_, b_ : a, b
In[47]:= testlist Join htnode " ", 6 , htnode "A", 3 ,
htnode "B", 1 , htnode "E", 5 , htnode "H", 2 ,
htnode "N", 2 , htnode "O", 2 , htnode "S", 3 , htnode "T", 3
Out[47]= , 6 , A, 3 , B, 1 , E, 5 ,
H, 2 , N, 2 , O, 2 , S, 3 , T, 3
10. To make the results here comparable to those in the book, we will use Htree from the book
as our sample tree. makeTreeTable[tree] produces a list of rules as described in the
problem. encodeString[str, rules] decodes the string according to those rules.
In[48]:= Htree " ABEHONST", " AT", " " , "AT", "T" , "A" ,
"BEHONS", "EON", "E" , "ON", "O" , "N" ,
"BHS", "BH", "H" , "B" , "S" ;
In[49]:= makeTreeTable prefix_, ch_ ch prefix ;
In[50]:= makeTreeTable prefix_, _, left_, right_ :
Join makeTreeTable Join prefix, 0 , left ,
makeTreeTable Join prefix, 1 , right
In[51]:= makeTreeTable tree_ : makeTreeTable , tree
Solutions to exercises 489
In[52]:= HtreeRules makeTreeTable Htree
Out[52]= 0, 0 , T 0, 1, 0 , A 0, 1, 1 , E 1, 0, 0 , O 1, 0, 1, 0 ,
N 1, 0, 1, 1 , H 1, 1, 0, 0 , B 1, 1, 0, 1 , S 1, 1, 1
In[53]:= encodeString str_, rules_ : Flatten Characters str . rules
In[54]:= encodeString str_ : encodeString str, HtreeRules
7.6 Dynamic programming
1. This implementation uses the identities given in the exercise together with some pattern
matching
In[1]:= F 1 : 1
F 2 : 1
In[3]:= F n_ ; EvenQ n : 2 F
n
2
1 F
n
2
F
n
2
2
F n_ ; OddQ n : F
n 1
2
1
2
F
n 1
2
2
In[5]:= Map F, Range 10
Out[5]= 1, 1, 2, 3, 5, 8, 13, 21, 34, 55
2.
In[6]:= FF 1 : 1
FF 2 : 1
In[8]:= FF n_?EvenQ : FF n 2 FF
n
2
1 FF
n
2
FF
n
2
2
FF n_?OddQ : FF n FF
n 1
2
1
2
FF
n 1
2
2
In[10]:= Map FF, Range 12
Out[10]= 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144
3.
In[11]:= Clear collatz
In[12]:= collatz n_, 0 : n
In[13]:= collatz n_, i_ :
collatz n, i
collatz n, i 1
2
; EvenQ collatz n, i 1
490 An Introduction to Programming with Mathematica
In[14]:= collatz n_, i_ :
collatz n, i 3 collatz n, i 1 1 ; OddQ collatz n, i 1
Here is the fifth iterate of the Collatz sequence for 27.
In[15]:= collatz 27, 5
Out[15]= 31
Here is the Collatz sequence for 27. You can see that it takes a long time for this sequence to
settle down to the cycle 4, 2, 1.
In[16]:= Table collatz 27, i , i, 0, 116
Out[16]= 27, 82, 41, 124, 62, 31, 94, 47, 142, 71, 214, 107, 322, 161, 484, 242, 121,
364, 182, 91, 274, 137, 412, 206, 103, 310, 155, 466, 233, 700, 350,
175, 526, 263, 790, 395, 1186, 593, 1780, 890, 445, 1336, 668, 334,
167, 502, 251, 754, 377, 1132, 566, 283, 850, 425, 1276, 638, 319, 958,
479, 1438, 719, 2158, 1079, 3238, 1619, 4858, 2429, 7288, 3644, 1822,
911, 2734, 1367, 4102, 2051, 6154, 3077, 9232, 4616, 2308, 1154, 577,
1732, 866, 433, 1300, 650, 325, 976, 488, 244, 122, 61, 184, 92, 46, 23,
70, 35, 106, 53, 160, 80, 40, 20, 10, 5, 16, 8, 4, 2, 1, 4, 2, 1, 4, 2
7.7 Higher-order functions and recursion
1. First, here is the definition for our user-defined fold.
In[1]:= fold f_, x_, : x
fold f_, x_, a_, r___ : fold f, f x, a , r
In[3]:= fold Plus, 0, a, b, c, d, e
Out[3]= a b c d e
In[4]:= foldList f_, x_, : x
foldList f_, x_, a_, r___ : Join x , foldList f, f x, a , r
In[6]:= foldList Times, 1, Range 6
Out[6]= 1, 1, 2, 6, 24, 120, 720
And here is nestList.
In[7]:= nestList f_, x_, 0 : x
nestList f_, x_, n_ : Join x , nestList f, f x , n 1
In[9]:= nestList Sin, , 3
Out[9]= , Sin , Sin Sin , Sin Sin Sin
Solutions to exercises 491
2. First, here is the recursive repeat function from this section.
In[10]:= repeat f_, lis_, pred_ : lis ; pred Drop lis, 1 , Last lis
In[11]:= repeat f_, lis_, pred_ : repeat f, Append lis, f Last lis , pred
Then the MemberQ function is used to test whether the currently computed point is a member
of the existing list of points.
In[12]:= landMineWalk :
repeat #1 0, 1 , 0, 1 , 1, 0 , 1, 0 Random Integer, 1, 4 &,
0, 0 , MemberQ #1, #2 &
Here is a test. On average, these walks will not be terribly long.
In[13]:= landMineWalk
Out[13]= 0, 0 , 1, 0 , 2, 0 , 1, 0
8 Numerics
8.2 Numbers
1. This function gives the polar form as a list consisting of the magnitude and the polar angle.
In[1]:= complexToPolar z_ : Abs z , Arg z
Here are the computations for the examples in the text.
In[2]:= complexToPolar 3 3
Out[2]= 3 2 ,
4
In[3]:= complexToPolar
3
Out[3]= 1,
3
2. This function uses a default value of 2 for the base. (Try replacing Fold with FoldList to
more clearly see what this function is doing.)
In[4]:= convert digits_List, base_ : 2 : Fold base #1 #2 &, 0, digits
Here are the digits for 9 in base 2:
In[5]:= IntegerDigits 9, 2
Out[5]= 1, 0, 0, 1
492 An Introduction to Programming with Mathematica
This converts them back to the base 10 representation.
In[6]:= convert %
Out[6]= 9
This does the same for the number 129 in base 16.
In[7]:= IntegerDigits 129, 16
Out[7]= 8, 1
In[8]:= convert %, 16
Out[8]= 129
This function is essentially an implementation of Horners method for fast polynomial
multiplication.
In[9]:= Clear a, b, c, d, e, x
In[10]:= convert a, b, c, d, e , x
Out[10]= e x d x c x b a x
In[11]:= Expand %
Out[11]= e d x c x
2
b x
3
a x
4
4. Here is the sumsOfCubes function.
In[12]:= sumsOfCubes n_Integer : Apply Plus, IntegerDigits n
3
Here is the function that performs the iteration.
In[13]:= sumsOfSums n_Integer, iter_ : NestList sumsOfCubes, n, iter
We see that the number 4 enters into a cycle.
In[14]:= sumsOfSums 4, 12
Out[14]= 4, 64, 280, 520, 133, 55, 250, 133, 55, 250, 133, 55, 250
In fact, it appears as if many initial values enter cycles.
In[15]:= sumsOfSums 32, 12
Out[15]= 32, 35, 152, 134, 92, 737, 713, 371, 371, 371, 371, 371, 371
In[16]:= sumsOfSums 7, 12
Out[16]= 7, 343, 118, 514, 190, 730, 370, 370, 370, 370, 370, 370, 370
Solutions to exercises 493
In[17]:= sumsOfSums 372, 12
Out[17]= 372, 378, 882, 1032, 36, 243, 99, 1458, 702, 351, 153, 153, 153
6. The function sumsOfPowers is a straightforward generalization of the previous cases.
In[18]:= sumsOfPowers n_, p_ : Apply Plus, IntegerDigits n
p
In[19]:= sumsOfPowers 123, 5
Out[19]= 276
8. Using the number 100 as an example, let us first put it in base 2.
In[20]:= BaseForm 100, 2
Out[20]//BaseForm=
1100100
2
Here is the list of its digits.
In[21]:= IntegerDigits 100, 2
Out[21]= 1, 1, 0, 0, 1, 0, 0
This performs a binary shift of one unit (actually, the 1 in RotateLeft is not needed here as
this is the default value to shift by).
In[22]:= l RotateLeft IntegerDigits 100, 2 , 1
Out[22]= 1, 0, 0, 1, 0, 0, 1
This converts back from base 2 to base 10 (using the convert function from Exercise 2).
In[23]:= convert l, 2
Out[23]= 73
Now we can put all of this code together to make the survivor function.
In[24]:= survivor n_ :
Module p , p RotateLeft IntegerDigits n, 2 ; Fold 2 #1 #2 &, 0, p
In[25]:= survivor 100
Out[25]= 73
You could of course do the same thing without the symbol p, but it is just a bit less readable.
In[26]:= survivor2 n_Integer :
Fold 2 #1 #2 &, 0, RotateLeft IntegerDigits n, 2
494 An Introduction to Programming with Mathematica
In[27]:= survivor2 100
Out[27]= 73
9. This function has a straightforward implementation. Each die can be viewed as a random
integer between 1 and 6.
In[28]:= rollEm : Random Integer, 1, 6 , Random Integer, 1, 6
In[29]:= rollEm
Out[29]= 3, 2
Here are five rolls in a row.
In[30]:= Table rollEm, 5
Out[30]= 6, 4 , 2, 3 , 5, 3 , 4, 3 , 4, 5
10. First generate a vector of 100 random real numbers on the interval 0 to 1.
In[31]:= data Table Random , 100 ;
You could rotate once to the left for each successive row.
In[32]:= ListDensityPlot NestList RotateLeft, data, Length data ;
0 20 40 60 80 100
0
20
40
60
80
100
Here are a few other things you can try.
In[33]:= ListDensityPlot NestList #
.75
&, data, Length data ;
0 20 40 60 80 100
0
20
40
60
80
100
Solutions to exercises 495
In[34]:= ListDensityPlot NestList
#1
1 #1
&, data, Length data , Mesh False ;
0 20 40 60 80 100
0
20
40
60
80
100
11. Here is the linear congruential generator.
In[35]:= linearCongruential[x_,m_,b_] := Mod[b x + 1, m]
With modulus 100 and multiplier 15, this generator quickly gets into a cycle.
In[36]:= NestList linearCongruential #, 100, 15 &, 5, 10
Out[36]= 5, 76, 41, 16, 41, 16, 41, 16, 41, 16, 41
With a larger modulus and multiplier, it appears as if this generator is doing better.
Here are the first 60 terms starting with a seed of 0.
In[37]:= data NestList linearCongruential #, 381, 15 &, 0, 60
Out[37]= 0, 1, 16, 241, 187, 139, 181, 49, 355, 373, 262, 121, 292, 190, 184, 94,
268, 211, 118, 247, 277, 346, 238, 142, 226, 343, 193, 229, 7, 106, 67,
244, 232, 52, 19, 286, 100, 358, 37, 175, 340, 148, 316, 169, 250, 322,
259, 76, 379, 352, 328, 349, 283, 55, 64, 199, 319, 214, 163, 160, 115
Sometimes it is hard to see if your generator is getting into a rut. Graphical analysis can help
by allowing you to see patterns over larger domains. Here is a ListPlot of this sequence
taken out to 5,000 terms.
In[38]:= ListPlot NestList linearCongruential #, 381, 15 &, 0.0, 5000 ;
1000 2000 3000 4000 5000
50
100
150
200
250
300
350
496 An Introduction to Programming with Mathematica
It appears as if certain numbers are repeating. Looking at the plot of the Fourier data shows
peaks at certain frequencies, indicating a periodic nature to the data.
In[39]:= ListPlot[Abs[Fourier[
NestList[linearCongruential[#,381,15]&,0.0,5000]]]];
1000 2000 3000 4000 5000
10
20
30
40
50
Using a much larger modulus and multiplier (chosen carefully), you can keep your generator
from getting in such short loops.
In[40]:= ListPlot
data NestList linearCongruential #, 2
16
, 27421 &, 0.0, 5000 ;
1000 2000 3000 4000 5000
10000
20000
30000
40000
50000
60000
In[41]:= ListPlot Abs Fourier data ;
1000 2000 3000 4000 5000
10000
20000
30000
40000
50000
13. First we implement the chi-square test and then use it to run tests on some data in the next
exercise.
In[42]:= chiSquare lis_List : Module m Length lis , n Max lis ,
i 1
n
Count lis, i
m
n
2
m
n
14. Here are some data generated using the linear congruential generator with small modulus and
multiplier.
Solutions to exercises 497
In[43]:= data NestList linearCongruential #, 381, 15 &, 0, 1000 ;
In[44]:= chiSquare data
Out[44]=
5018521
1001
In[45]:= N %
Out[45]= 5013.51
Notice that the statistic is quite far from2 n of n. This is a particularly pathological
sequence. You can see a cycle of length 63 within the first 100 iterates.
In[46]:= NestList linearCongruential #1, 381, 15 &, 0, 100
Out[46]= 0, 1, 16, 241, 187, 139, 181, 49, 355, 373, 262, 121, 292, 190, 184,
94, 268, 211, 118, 247, 277, 346, 238, 142, 226, 343, 193, 229, 7,
106, 67, 244, 232, 52, 19, 286, 100, 358, 37, 175, 340, 148, 316, 169,
250, 322, 259, 76, 379, 352, 328, 349, 283, 55, 64, 199, 319, 214,
163, 160, 115, 202, 364, 127, 1, 16, 241, 187, 139, 181, 49, 355,
373, 262, 121, 292, 190, 184, 94, 268, 211, 118, 247, 277, 346, 238,
142, 226, 343, 193, 229, 7, 106, 67, 244, 232, 52, 19, 286, 100, 358
Here are those positions that contain the number 1.
In[47]:= Position %, 1
Out[47]= 2 , 65
8.3 Working with numbers
1. The number 1.23 has machine precision.
In[1]:= Precision 1.23
Out[1]= MachinePrecision
Asking Mathematica to generate 100 digits of precision from a number that only contains about
16 digits of precision would require it to produce 84 digits without any information about
where those digits should come from.
498 An Introduction to Programming with Mathematica
2. You could simply produce a table showing the number of digits of precision needed in the
input compared with the accuracy of the result.
In[2]:= Table x, Accuracy N 2 , x
200
2
200
, x, 100, 140, 5 TableForm
Out[2]//TableForm=
100 67.596
105 72.596
110 77.596
115 82.596
120 87.596
125 92.596
130 97.596
135 102.596
140 107.596
8.4 Working with arrays of numbers
1. Note the need for a delayed rule in this function.
In[1]:= RandomSparseArray n_Integer : SparseArray i_, i_ Random , n, n
In[2]:= Normal RandomSparseArray 5 MatrixForm
Out[2]//MatrixForm=
0.197227 0 0 0 0
0 0.509405 0 0 0
0 0 0.965962 0 0
0 0 0 0.873469 0
0 0 0 0 0.959528
2. Here is the definition of tridiagonalMatrix.
In[3]:= tridiagonalMatrix n_, p_, q_ :
SparseArray i_, i_ p, i_, j_ ; Abs i j 1 q , n, n
In[4]:= tridiagonalMatrix 5, ,
Out[4]= SparseArray 13 , 5, 5
In[5]:= Normal % MatrixForm
Out[5]//MatrixForm=
0 0 0
0 0
0 0
0 0
0 0 0
Solutions to exercises 499
3. First we create the packed array vector.
In[6]:= vec Table Random , 10
6
;
In[7]:= Developer`PackedArrayQ vec
Out[7]= True
Replacing the first element in vec with a 1 gives us an expression which is not packed.
In[8]:= newvec ReplacePart vec, 1, 1 ;
In[9]:= Developer`PackedArrayQ newvec
Out[9]= False
The size of the unpacked object is about two and a half times larger than the packed array.
In[10]:= Map ByteCount, vec, newvec
Out[10]= 8000056, 20000032
In[11]:= % 2 % 1 N
Out[11]= 2.49999
Sorting the packed object is about four or five times faster than sorting the unpacked object.
In[12]:= Timing Do Sort vec , 5
Out[12]= 4.406 Second, Null
In[13]:= Timing Do Sort newvec , 5
Out[13]= 18.777 Second, Null
Finding the minimum element is about one order of magnitude faster with the packed array.
In[14]:= Timing Min vec ;
Out[14]= 0.01 Second, Null
In[15]:= Timing Min newvec ;
Out[15]= 0.131 Second, Null
500 An Introduction to Programming with Mathematica
8.5 Numerical computations
1. We will overload newton to invoke the secant method when given a list of two numbers as
the second argument.
In[1]:= Options newton
MaxIterations $RecursionLimit,
PrecisionGoal Automatic,
WorkingPrecision Automatic
;
In[2]:= newton fun_, x1_?NumericQ, x2_?NumericQ , opts___?OptionQ :
Module maxIterations, precisionGoal,
workingPrecision, initx, df, next, result ,
maxIterations, precisionGoal, workingPrecision
MaxIterations, PrecisionGoal, WorkingPrecision . Flatten opts .
Options newton ;
If precisionGoal Automatic, precisionGoal
Min Precision x1 , Precision x2 ;
If workingPrecision Automatic,
workingPrecision precisionGoal 10 ;
initx SetPrecision x1, x2 , workingPrecision ;
df a_, b_ : fun b fun a b a ;
next a_, b_ : a, b
fun b
df a, b
;
result FixedPoint next, initx, maxIterations 2 ;
SetPrecision result, precisionGoal
In[3]:= f x_ : x
2
2
In[4]:= newton f, 1., 2.
Out[4]= 1.41421
In[5]:= newton f, 1.0`60, 2.0`50
Out[5]= 1.4142135623730950488016887242096980785696740946953
In[6]:= Precision %
Out[6]= 50.
5. Here is a three-dimensional vector.
In[7]:= vec 1, 3, 2 ;
This computes the l norm of the vector.
In[8]:= norm v_?VectorQ, l_: Infinity : Max Abs v
Solutions to exercises 501
In[9]:= norm vec
Out[9]= 3
You can compare this with the built-in Norm function.
In[10]:= Norm vec, Infinity
Out[10]= 3
Here is a 3 3 matrix.
In[11]:= mat 1, 2, 3 , 1, 0, 2 , 2, 3, 2
Out[11]= 1, 2, 3 , 1, 0, 2 , 2, 3, 2
Here, then, is the matrix norm.
In[12]:= norm m_?MatrixQ, l_: Infinity :
norm Apply Plus, Abs Transpose m , Infinity
In[13]:= norm mat
Out[13]= 7
Again, a comparison with the built-in Norm function.
In[14]:= Norm mat, Infinity
Out[14]= 7
Notice how we overloaded the definition of the function norm so that it would act differently
depending upon what type of argument it was given. This is a particularly powerful feature of
Mathematica. The expression _?MatrixQ on the left-hand side of the definition causes the
function norm to use the definition on the right-hand side only if the argument is in fact a
matrix (if it passes the MatrixQ test). If that argument is a vector (if it passes the VectorQ
test), then the previous definition is used.
6. Here is the function to compute the condition number of a matrix (using the l norm).
In[15]:= conditionNumber m_?MatrixQ :
norm m, Infinity norm Inverse m , Infinity
In[16]:= HilbertMatrix n_ : Table
1
i j 1
, i, n , j, n
In[17]:= conditionNumber HilbertMatrix 3
Out[17]= 748
502 An Introduction to Programming with Mathematica
Compare this with the condition number of a random matrix.
In[18]:= conditionNumber Table Random , 3 , 3
Out[18]= 18.7428
Here are the condition numbers of the first ten Hilbert matrices.
In[19]:= Map conditionNumber HilbertMatrix # &, Range 10
Out[19]= 1, 27, 748, 28375, 943656, 29070279,
1970389773
2
,
33872791095,
2199309082685
2
, 35357439251992
In[20]:= N %
Out[20]= 1., 27., 748., 28375., 943656., 2.90703 10
7
,
9.85195 10
8
, 3.38728 10
10
, 1.09965 10
12
, 3.53574 10
13
9 Graphics programming
9.1 Structure of graphics
1. The color wheel can be obtained by mapping the Hue directive over successive sectors of a
disk. Note that the argument to Hue must be scaled so that it falls within the range 0 to 1.
In[1]:= colorWheel n_ :
Show Graphics Map Hue
#
2 n
, Disk 0, 0 , 1, #, # n &,
Range 0, 2 n, n , AspectRatio Automatic
Here is a color wheel created from 256 separate sectors (hues).
In[2]:= colorWheel
256
;
Solutions to exercises 503
2. Here is the circle graphic primitive together with a text label.
In[3]:= circ Circle 0, 0 , 1 ;
In[4]:= ctext Text StyleForm "Circle",
FontFamily "Times", FontSlant "Italic", FontSize 12 ,
Cos
5
4
.25, Sin
5
4
;
This generates the graphics primitive for the triangle and its text label.
In[5]:= tri Line 1, 0 , 0, 1 , 1, 0 , 1, 0 ;
In[6]:= ttext Text StyleForm "Triangle", FontFamily "Times",
FontSlant "Italic", FontSize 12 , 0, 0 .05 ;
Here is the rectangle and label.
In[7]:= rect Line 1, 1 , 1, 1 , 2, 1 , 2, 1 , 1, 1 ;
In[8]:= rtext Text StyleForm "Rectangle", FontFamily "Times",
FontSlant "Italic", FontSize 12 , 1.5, 1 .05 ;
Finally, this displays each of these graphics elements all together.
In[9]:= Show Graphics circ, tri, rect, ctext, ttext, rtext ,
AspectRatio Automatic ;
Circle
Triangle
Rectangle
3. First, we need to create the cuboid graphic object. Cuboid takes a list of three numbers as the
coordinates of its lower-left corner. This maps the object across two such lists.
In[10]:= Map Cuboid # &, Table Random , 2 , 3
Out[10]= Cuboid 0.177395, 0.551966, 0.857107 ,
Cuboid 0.545712, 0.76829, 0.48344
Here is a list of six cuboids and the resulting graphic. Notice the large amount of overlap of the
cubes. You can reduce the large overlap by specifying minimumand maximum values of the
cuboid.
In[11]:= cubes Map Cuboid #1 &, Table Random , 6 , 3 ;
504 An Introduction to Programming with Mathematica
In[12]:= Show Graphics3D cubes ;
4. First we create the Point graphics primitives randomly placed in the unit square.
In[13]:= randomcoords : Point Random , Random ;
This creates the point sizes according to the specification given in the statement of the
problem.
In[14]:= randomsize : PointSize Random Real, .01, .1
This will assign a random color to each primitive.
In[15]:= randomcolor : Hue Random
Here then are 500 points. (You may find it instructive to look at just one of these points.)
In[16]:= pts Table randomcolor, randomsize, randomcoords , 500 ;
And here is the graphic.
In[17]:= Show Graphics pts, PlotRange All
Out[17]= Graphics
Solutions to exercises 505
5. The algebraic solution is given by the following steps. First solve the equations for x and y.
In[18]:= Clear x, y, r
In[19]:= soln Solve x 1
2
y 1
2
2, x 3
2
y 4
2
r
2
, x, y
Out[19]= x
1
50
58 4 r
2
3 529 54 r
2
r
4
,
y
1
50
131 3 r
2
4 529 54 r
2
r
4
,
x
1
50
58 4 r
2
3 529 54 r
2
r
4
,
y
1
50
131 3 r
2
4 529 54 r
2
r
4
Then find those values of r for which the x and y coordinates are identical.
In[20]:= Solve x . soln 1 x . soln 2 ,
y . soln 1 y . soln 2 , r
Out[20]= r 5 2 , r 5 2 , r 5 2 , r 5 2
Here then are those values of r that are positive.
In[21]:= Cases %, r _?Positive
Out[21]= r 5 2 , r 5 2
To display the solution, we will plot the first circle with solid lines and the two solutions with
dashed lines together in one graphic. Here is the first circle centered at (1, 1).
In[22]:= circ Circle 1, 1 , 2 ;
In[23]:= Show Graphics circ, Axes Automatic, AspectRatio Automatic ;
0.5 1 1.5 2
0.5
1
1.5
2
Notice that we have used the Axes and AspectRatio options because we want these
commands to apply to the entire graphic.
506 An Introduction to Programming with Mathematica
Here are the circles that represent the solution to the problem.
In[24]:= r1 5 2 ;
r2 5 2 ;
In[26]:= Show Graphics circ, Circle 3, 4 , r1 , Circle 3, 4 , r2 ,
Axes Automatic, AspectRatio Automatic ;
8 6 4 2 2
2
2
4
6
8
10
We wanted to display the solutions (two circles) using dashed lines. The graphics directive
Dashing[{x,y}] directs all subsequent lines to be plotted as dashed, alternating the dash x
units and the space y units. We use it as a graphics directive on the two circles c1 and c2. The
important point to note here is that each of the circles inherits only those directives in whose
scope they appear.
In[27]:= dashc1 Dashing .025, .025 , Circle 3, 4 , r1 ;
dashc2 Dashing .05, .05 , Circle 3, 4 , r2 ;
In[29]:= Show Graphics circ, dashc1, dashc2 ,
Axes Automatic, AspectRatio Automatic ;
8 6 4 2 2
2
2
4
6
8
10
6. This loads the package containing the definitions for the polyhedra.
In[30]:= Needs "Graphics`Polyhedra`"
Solutions to exercises 507
It is often helpful to get a list of the functions defined in a recently loaded package.
In[31]:= Names "Graphics`Polyhedra` "
Out[31]= Geodesate, GreatDodecahedron, GreatIcosahedron,
GreatStellatedDodecahedron, OpenTruncate, Polyhedra,
Polyhedron, SmallStellatedDodecahedron, Stellate, Truncate
First the polyhedra are turned into Graphics3D objects.
In[32]:= solids Map Graphics3D, Cube , Dodecahedron , GreatDodecahedron ,
GreatIcosahedron , GreatStellatedDodecahedron , Icosahedron ,
Octahedron , Tetrahedron , SmallStellatedDodecahedron ;
We then use Partition to split the list of nine solids into three sublists and then display the
nine polyhedra with GraphicsArray and Show.
In[33]:= Show GraphicsArray Partition solids, 3
Out[33]= GraphicsArray
7. Here is a plot of the sine function.
In[34]:= sinplot Plot Sin x , x, 0, 2
1 2 3 4 5 6
1
0.5
0.5
1
Out[34]= Graphics
This solution is essentially that given in Exploring Mathematics with Mathematica (Gray and
Glynn 1991). Extracting the points from which Mathematica constructs the plot is accom-
plished by the Nest statement. The Line primitive is then mapped across those points in
508 An Introduction to Programming with Mathematica
such a way as to create lines from the points on the graph to points on the x-axis with the same
x-coordinate.
In[35]:= Show[sinplot,
Graphics[
{Thickness[.001],
Map[Line[{{#[[1]], 0}, #}]&,
Nest[First, sinplot, 4]]}]]
1 2 3 4 5 6
1
0.5
0.5
1
Out[35]= Graphics
You could also construct this using pattern matching. Here are the coordinates.
In[36]:= coords Cases sinplot, p_?NumericQ, q_?NumericQ , Infinity Short
Out[36]//Short=
2.61799 10
7
, 2.61799 10
7
, 80 , 6.28319, 23
Here is what we use to create vertical lines from each coordinate.
In[37]:= lines Map Line # 1 , 0 , # &, coords Short
Out[37]//Short=
Line 2.61799 10
7
, 0 , 2.61799 10
7
, 23 , 80 , Line 1
Here then is the final graphic.
In[38]:= Show sinplot, Graphics
Map Line # 1 , 0 , # &,
Cases sinplot, p_?NumericQ, q_?NumericQ , Infinity
;
1 2 3 4 5 6
1
0.5
0.5
1
Solutions to exercises 509
9.2 Graphics programming
1. The function ComplexListPlot plots a list of complex numbers in the complex plane, with
the real part identified with the horizontal axis and the imaginary part identified with the
vertical axis. The appropriate options are extracted fromComplexListPlot using Filter
Options, given the name complexOpts, and then passed to ListPlot.
In[1]:= Utilities`FilterOptions`
In[2]:= ComplexListPlot points_, opts___ :
Module complexOpts FilterOptions ListPlot, opts ,
ListPlot Map Re #1 , Im #1 &, points , complexOpts, PlotStyle
RGBColor 1, 0, 0 , PointSize .025 , AxesLabel "Re", "Im"
This plots four complex numbers in the plane.
In[3]:= ComplexListPlot 1 I, 2 I, 1 2 I, 0, 1 ;
1 0.5 0.5 1 1.5 2
Re
2
1.5
1
0.5
0.5
1
Im
2. The function RootPlot, takes a polynomial, solves for its roots, and then uses ComplexList-
Plot from Exercise 1 to plot these roots in the complex plane.
In[4]:= RootPlot poly_, z_, opts___ :
ComplexListPlot z . NSolve poly 0, z , opts
In[5]:= RootPlot 1 z 2 z
2
3 z
3
5 z
4
8 z
5
13 z
6
, z, AspectRatio Automatic ;
0.6 0.4 0.2 0.2 0.4
Re
0.6
0.4
0.2
0.2
0.4
0.6
Im
510 An Introduction to Programming with Mathematica
3.
In[6]:= Clear RootPlot
In[7]:= Utilities`FilterOptions`
In[8]:= RootPlot fun_, x_, xmin_, xmax_ , opts___ : Module
z, fplot, pts, spts, roots, points, f Function x, Evaluate fun ,
fplot Plot f x , x, xmin, xmax , DisplayFunction Identity,
Evaluate FilterOptions Plot, opts ;
pts Cases fplot, Line z__ z, ;
spts Map First, Select Split pts, Sign Last #2 Sign Last #1 & ,
Length #1 2 & , 2 ;
roots Map FindRoot f x 0, x, # 1 , # 2 &, spts ;
points Map Point #, 0 &, x . roots ;
Show fplot, DisplayFunction $DisplayFunction,
Epilog RGBColor 0, 0, 1 , PointSize .02 , points ;
roots
In[9]:= RootPlot Sin x 2 Sin x , x, , 4 , PlotStyle Dashing .02, .02 ;
2.5 2.5 5 7.5 10 12.5
1
0.5
0.5
1
4. Here is the new code for DataPlot.
In[10]:= Clear DataPlot
In[11]:= Needs "Utilities`FilterOptions`"
In[12]:= Options DataPlot Options ListPlot ;
In[13]:= DataPlot::baddim "The data used by DataPlot must
be in the form of a one or two dimensional list.";
In[14]:= DataPlot data_, opts___ : Module pjQ, pts, lines ,
pjQ PlotJoined . Flatten opts, Options DataPlot ;
pts Which
VectorQ data , MapIndexed #2 1 , #1 &, data ,
Dimensions data 2 2, data,
True, Message DataPlot::baddim ; $Failed ;
If pts $Failed,
Show Graphics PointSize .02 ,
Point pts, If pjQ, lines Line pts , lines ,
FilterOptions Graphics, opts , Axes Automatic
Solutions to exercises 511
Here is some sample two-dimensional data.
In[15]:= data2D 0.043, 0.575 , 0.151, 0.120 ,
0.234, 0.001 , 0.283, 0.930 , 0.343, 0.569 , 0.416, 0.768 ,
0.465, 0.675 , 0.539, 0.528 , 0.786, 0.856 , 0.914, 0.794 ;
And here is some sample one-dimensional data.
In[16]:= data1D Table Random Integer, 1, 10 , 8
Out[16]= 2, 5, 7, 5, 9, 9, 10, 6
In[17]:= DataPlot data1D, PlotJoined True
2 4 6 8
4
6
8
10
Out[17]= Graphics
In[18]:= DataPlot data2D, PlotJoined True
0.2 0.4 0.6 0.8
0.2
0.4
0.6
0.8
Out[18]= Graphics
Here is some data that DataPlot is not designed to deal with.
In[19]:= DataPlot 1, 2, 3 , 2, 3, 4 , 4, 5, 6
DataPlot::baddim : The data used by DataPlot
must be in the form of a one or two dimensional list.
5. There are a number of things that could go wrong with the algorithm by just choosing a base
point randomly and then sorting according to the arctangent. The default branch cut for
ArcTan gives values between 2 and 2. (The reader is encouraged to think about why
this could occasionally cause the algorithm in the text to fail.) By choosing the base point so
that it lies at some extreme of the diameter of the set of points, the polar angle algorithm given
in the text will work consistently. If you choose the base point so that it is lowest and left-most,
then all the angles will be in the range (0, ].
512 An Introduction to Programming with Mathematica
In[20]:= simpleClosedPath1 lis_List : Module base, angle, sorted ,
base Last Sort lis, #2 2 #1 2 & ;
angle a_, b_ : Apply ArcTan, b a ;
sorted
Sort Complement lis, base , angle base, #1 angle base, #2 & ;
Join base , sorted, base
In[21]:= pts Table Random , 20 , 2 ;
In[22]:= PointPlot coords_List :
Show Graphics
Line coords ,
PointSize .02 , RGBColor 1, 0, 0 , Map Point, coords
In[23]:= PointPlot simpleClosedPath1 pts ;
7. A simple change to the programsimpleClosedPath given in Exercise 5 chooses the base
point with the largest y-coordinate.
In[24]:= simpleClosedPath3 lis_ :
Module base, angle, sorted , base Last Sort lis, #2 2 #1 2 & ;
angle a_, b_ : ArcTan b a ; sorted Sort Complement lis, base ,
angle base, #1 angle base, #2 & ; Join base , sorted, base
In[25]:= pts Table Random , 20 , 2 ;
In[26]:= PointPlot simpleClosedPath3 pts ;
8. The area of a triangle is one-half the base times the altitude. For arbitrary points, the altitude
requires a bit of computation that does not generalize.
Solutions to exercises 513
The magnitude of the cross product of two vectors gives the area of the parallelogram that
they determine. Since the vectors we are working with are in two-dimensional space, we
embed them in three-dimensional space in the plane z 0 so that we can compute the cross
product which, for the purposes of this problem, only makes sense in three dimensions.
In[27]:= "Calculus`VectorAnalysis`"
In[28]:= CrossProduct x
2
, y
2
x
1
, y
1
, x
3
, y
3
x
1
, y
1
. x_, y_ x, y, 0
Out[28]= 0, 0, x
2
y
1
x
3
y
1
x
1
y
2
x
3
y
2
x
1
y
3
x
2
y
3
Here are the coordinates for a triangle.
In[29]:= a 0, 0 ;
b 5, 0 ;
c 3, 2 ;
And here is the computation for the cross product.
In[32]:= CrossProduct b a, c a . x_, y_ x, y, 0
Out[32]= 0, 0, 10
So the given area is then just half the magnitude of the cross product.
In[33]:=
Apply Plus, %
2
Out[33]= 5
Here is a function that computes the area of any triangle using the cross product.
In[34]:= triangleArea v_List :
1
2
Apply Plus,
CrossProduct v 2 v 1 , v 3 v 1 . x_, y_ x, y, 0
This is done more simply using determinants and this method generalizes more easily to
higher dimensions.
In[35]:= triangleArea v1_, v2_, v3_ :
1
2
Det v1, v2, v3 . x_, y_ x, y, 1
In[36]:= triangleArea a, b, c
Out[36]= 5
9. The key observation is that in computing the area of a triangle using the determinant formula-
tion as in Exercise 9, the area will be a positive quantity if the points are given in counter-clock-
wise order, and will be negative if in clockwise order. So, for a given point p not on a line ab,
the area of abp will be positive (computed using determinants), if p is to the left of ab.
Similarly, for each of the lines in a polygon, relative to the given point p. So, to perform the
computation, we first partition the polygon into pairs of points, and then map the triangle area
514 An Introduction to Programming with Mathematica
function with the given point across each pair. If all such areas are greater than or equal to
zero, then a value of True is returned.
In[37]:= pointInPolygonQ poly_, p_ : Module area ,
area v1_, v2_, v3_ :
1
2
Det v1, v2, v3 . x_, y_ x, y, 1 0;
Apply And, Map area Join p , # &,
Partition poly . a_, b__ a, b, a , 2, 1
Here are the coordinates for a quadrilateral and two distinct points.
In[38]:= quad 1, 0 , 0, 1 , 1, 0 , 0, 1 ;
In[39]:= p1 0, 0 ;
p2 1, 1 ;
In[41]:= Show Graphics Line quad . a_, b__ a, b, a ,
PointSize .025 , Point p1 , Point p2 , AspectRatio Automatic ;
Finally, here are the computations for these points and polygon.
In[42]:= pointInPolygonQ quad, p1
Out[42]= True
In[43]:= pointInPolygonQ quad, p2
Out[43]= False
12. RT (for ReingoldTilford), replaces the placeTree function. In placeTree, the result was a
separation tree plus two numbers: width of the left side of the tree and width of the right side
of the tree. In RT, the result is instead a separation tree plus two lists, the first giving the width
of the left side of each level of the tree, the second giving the corresponding widths on the
right side. sep is calculated by adding the right widths of the left subtree to the left widths of
the right subtree at each level, and taking the maximum separation. drawSepTree is
unchanged.
In[44]:= IPM3`Trees`
Solutions to exercises 515
In[45]:= RT _ : , ,
In[46]:= RT _, lc_, rc_ :
With left RT lc , right RT rc , minsep 2.0 , With sep
1
2
Max 0, Max Plus #1 & zip left 3 , right 2 minsep ,
With newtree sep, left 1 , right 1 ,
leftedge Join sep , extend left 2 , right 2 , sep ,
rightedge Join sep , extend right 3 , left 3 , sep ,
newtree, leftedge, rightedge
In[47]:= placeTree _ : , 0, 0
placeTree _, lc_, rc_ :
Module left placeTree lc , right placeTree rc , minsep 1.0, sep ,
sep left 3 right 2 minsep;
sep, left 1 , right 1 , left 2
sep
2
, right 3
sep
2
In[49]:= drawSepTree , lev_, xaxis_ : Disk xaxis, lev , 0.1
drawSepTree sep_, lc_, rc_ , lev_, xaxis_ :
Join Disk xaxis, lev , 0.1 , Line xaxis, lev , xaxis sep, lev 1 ,
Line xaxis, lev , xaxis sep, lev 1 ,
drawSepTree lc, lev 1, xaxis sep , drawSepTree rc, lev 1, xaxis sep
In[51]:= drawTree t_ : drawSepTree RT t 1 , 0, 0
The auxiliary functions are zip and extend. Given the left widths of each level of the right
subtree, and the right widths of each level of the left subtree, the separation of the two subtrees
is determined by adding those numbers at each level and taking the maximum. zip is used to
join those two lists into a list of pairs; it facilitates this process.
In[52]:= zip , _ :
zip _, :
zip x1_, y1___ , x2_, y2___ : Join x1, x2 , zip y1 , y2
When the separation of a trees subtrees is determined, the lists of left and right widths of the
combined tree are computed from the corresponding lists for the subtrees. This is simple
enough for the most part: the left widths of the tree are obtained mainly by taking the left
widths of the left subtree and shifting them left; and similarly for the right widths. There is an
exception, though: if the right subtree is taller than the left subtree, the left widths of the
bottom part of the tree are obtained from the left widths of the bottom part of the right
subtree. Combining the left widths of the two subtrees to create the list of left widths of the
combined tree is done by extend.
In[55]:= extend edges1_, edges2_, sep_ : Join edges1 sep,
Take edges2, Min 0, Length edges1 Length edges2 sep
The same reasoning applies to computing the right widths, and extend is also used for that.
516 An Introduction to Programming with Mathematica
Here is the tricky tree drawn in Figure 9.5.
In[56]:= Clear a, b, c, d, e, f, g
In[57]:= t1 a, b , a, c, e, g , f , d , b ;
In[58]:= Show Graphics drawTree t1 , AspectRatio Automatic ;
13. RT is modified so that the left widths and right widths of each row take into account the width
of the labels.
In[59]:= RT x_ : ,
width x
2
,
width x
2
RT x_, lc_, rc_ :
With left RT lc , right RT rc , minsep 0.5 , With sep
1
2
Max 0, Max Plus #1 & zip left 3 , right 2 minsep ,
With newtree sep, left 1 , right 1 ,
leftedge Join
width x
2
, extend left 2 , right 2 , sep ,
rightedge Join
width x
2
, extend right 3 , left 3 , sep ,
newtree, leftedge, rightedge
width t_ : StringLength t
Drawing the following tree using the new RT and the old drawSepTree will show the
difference in the layout of the trees. However, since drawSepTree above only prints disks at
each node, a new version of it is required.
In[62]:= t1 "a", "abcdef" , "", "abcdefghij" , "abc" ;
The new version of drawSepTree draws the labels at each node instead of a disk. A complicat-
ing factor is that we can no longer just draw the lines from the center of the disk, since this
would collide with the text. So the lines are now drawn in such a way as to leave a gap between
the text and the line.
In[63]:= settext lab_ : FontForm lab, "Helvetica", 9
Solutions to exercises 517
In[64]:= drawSepTree lab_ , , lev_, xaxis_ : Text settext lab , xaxis, lev
In[65]:= drawSepTree lab_, lc_, rc_ , sep_, ls_, rs_ , lev_, xaxis_ :
With h1 If lab "", 0, .3 , h2 If lc 1 "", 0, .3 ,
h3 If rc 1 "", 0, .3 , Join Text settext lab , xaxis, lev ,
Line xaxis
sep h1
2
, lev
h1
2
, xaxis sep
sep h2
2
, lev 1
h2
2
,
Line xaxis
sep h1
2
, lev
h1
2
,
xaxis sep
sep h3
2
, lev 1
h3
2
, drawSepTree lc, ls,
lev 1, xaxis sep , drawSepTree rc, rs, lev 1, xaxis sep
In[66]:= drawTree t_ : drawSepTree t, RT t 1 , 0, 0
In[67]:= Show Graphics drawTree t1 , PlotRange All ;
a
abcdef
abcdefghij abc
9.3 Sound
1. When x is close to 2, the frequency is quite low. As x increases, the fraction 1000 x gets
larger, making the frequency of the sine function bigger. This in turn makes the tone much
higher in pitch. As x approaches 0, the function is oscillating more and more, and at 0, the
function can be thought of as oscillating infinitely often. In fact, it is oscillating so much that
the sampling routine is not able to effectively compute amplitudes and, hence, we hear noise in
this region.
In[1]:= Play Sin
1000
x
, x, 2, 2
Power::infy : Infinite expression
1
0.
encountered. More
Out[1]= Sound
518 An Introduction to Programming with Mathematica
3. To generate a tone whose rate increases one octave per second, you need the sine of a function
whose derivative doubles each second (frequency is a rate). That function is 2
t
, so here is the
command to produce the tone. You need to carefully choose a range for t that generates tones
in a reasonable range.
In[2]:= Play Sin 2
t
, t, 10, 14
Out[2]= Sound
5. Here is a function that creates a square wave with decreasing amplitudes for higher overtones.
In[3]:= SquareWave freq_, n_ : Sum
Sin freq i 2 t
i
, i, 1, n, 2
In[4]:= Plot SquareWave 440, 17 , t, 0, .01 ;
0.002 0.004 0.006 0.008 0.01
0.75
0.5
0.25
0.25
0.5
0.75
Here then, is an example of playing a square wave.
In[6]:= Play SquareWave 440, 17 , t, 0, .5
Out[6]= Sound
7. This function creates a saw-tooth wave. The user specifies the fundamental frequency and the
number of terms in the approximation.
In[7]:= SawtoothWave freq_, n_ : Sum
Sin freq i 2 t
i
, i, 1, n
In[8]:= Plot SawtoothWave 440, 17 , t, 0, .01 ;
0.002 0.004 0.006 0.008 0.01
1.5
1
0.5
0.5
1
1.5
Solutions to exercises 519
This plays the wave for a half-second duration.
In[9]:= Play SawtoothWave 440, 17 , t, 0, .5
Out[9]= Sound
Here are definitions for true sawtooth and square waves.
In[10]:= Fractional x_ : x Floor x
In[11]:= SawtoothWave[x_]:= Fractional[-x]
In[12]:= SquareWave x_ :
1
2
Sign SawtoothWave x
1
2
1
Here are plots at the fundamental frequency of 440.
In[13]:= Plot SawtoothWave 440 t , t, 0,
3
440
;
0.001 0.002 0.003 0.004 0.005 0.006
0.2
0.4
0.6
0.8
1
In[14]:= Plot SquareWave 440 t , t, 0,
3
440
;
0.001 0.002 0.003 0.004 0.005 0.006
0.6
0.8
1.2
1.4
9. Here is a function that picks out frequencies from the pentatonic scale, using essentially
brownian motion 1 f
2
to select notes.
In[15]:= pentatonic n_Integer, r_ : 2 : Module pscale, steps ,
pscale 277.183, 311.13, 369.99, 415.30, 466.16, 554.37 ;
steps Table Random Integer, r, r , n ;
pscale Mod FoldList Plus, 3, steps , 4 1
You could play a pentatonic melody as follows:
In[16]:= SetAttributes PlayTones, Listable
520 An Introduction to Programming with Mathematica
In[17]:= PlayTones freq_, time_: 0.5 : Play Sin 2 t freq , t, 0, time
In[18]:= PlayTones pentatonic 24
Out[18]= Sound , Sound , Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound , Sound ,
Sound , Sound , Sound , Sound , Sound , Sound
11. In this function, the notes are randomly chosen from the C major scale 1 f
0
and the durations
are randomly chosen from the list that represents eighth notes, quarter notes, half notes, and
whole notes (also 1 f
0
). PlayTones accepts two arguments, so MapThread threads corre-
sponding notes and durations through PlayTones.
In[19]:= tonesAndTimes n_ : Module cmajor, notes, durs ,
cmajor Table N 261.62558 2
j 12
, j, 0, 11 ;
notes : Table cmajor Random Integer, 1, 12 , n ;
durs : Table
1
2
Random Integer, 0,3
, n ;
MapThread PlayTones, notes, durs
13. Following the implementation in the text, we first create ten steps between 2 and 2 (you can
alter the range of step movements). These steps will determine how to move up or down the
list of tone durations (1/8, 1/4, 1/2, 1).
In[20]:= d10 Table Random Integer, 2, 2 , 10
Out[20]= 0, 0, 2, 1, 2, 1, 2, 2, 2, 2
In[21]:= Mod FoldList Plus, 0, d10 , 4 1
Out[21]= 1, 1, 1, 3, 2, 4, 1, 3, 1, 3, 1
In[22]:= durs
1
8
,
1
4
,
1
2
, 1 %
Out[22]=
1
8
,
1
8
,
1
8
,
1
2
,
1
4
, 1,
1
8
,
1
2
,
1
8
,
1
2
,
1
8
Here are some 1 f
2
tones.
In[23]:= s10 Table Random Integer, 2, 2 , 10
Out[23]= 2, 2, 1, 2, 1, 1, 2, 2, 1, 1
In[24]:= pos Mod FoldList Plus, 0, s10 , 13 1
Out[24]= 1, 12, 10, 9, 7, 6, 7, 5, 7, 6, 7
Solutions to exercises 521
In[25]:= Cmajor Table N 261.62558 2
j 12
, j, 0, 12
Out[25]= 261.626, 277.183, 293.665, 311.127, 329.628, 349.228,
369.994, 391.995, 415.305, 440., 466.164, 493.883, 523.251
In[26]:= Length Cmajor
Out[26]= 13
In[27]:= tones Cmajor pos
Out[27]= 261.626, 493.883, 440., 415.305, 369.994,
349.228, 369.994, 329.628, 369.994, 349.228, 369.994
In[28]:= MapThread PlayTones, tones, durs ;
And finally, here is one function that puts this all together.
In[29]:= tonesAndTimes2 n_ : Module cmajor, tones, durs, d, t ,
cmajor Table N 261.62558 2
j 12
, j, 0, 12 ;
d Table Random Integer, 2, 2 , n ;
durs
1
8
,
1
4
,
1
2
, 1 Mod FoldList Plus, 0, d , 4 1 ;
t Table Random Integer, 2, 2 , n ;
tones cmajor Mod FoldList Plus, 0, t , 13 1 ;
MapThread PlayTones, tones, durs
In[30]:= tonesAndTimes2 12 ;
10 Front end programming
10.2 The structure of cells and notebooks
1. Here is the expression to create the notebook.
In[1]:= nb NotebookPut
Notebook
Cell "Demo Notebook", "Title" ,
Cell "Section 1: Sample Cells", "Section" ,
Cell "This is a text cell", "Text" ,
Cell "2 3 5 ", "Input" ,
Cell "1 2 3", "Input"
Out[1]= NotebookObject Untitled 3
522 An Introduction to Programming with Mathematica
2. First we read the notebook from Exercise 1 into the kernel with NotebookGet.
In[2]:= nbkernel NotebookGet nb
Out[2]= Notebook
Cell CellGroupData Cell Demo Notebook, Title , Cell CellGroupData
Cell Section 1: Sample Cells, Section , Cell This is a text cell,
Text , Cell 2 3 5 , Input , Cell 1 2 3, Input , Open ,
Open , FrontEndVersion 5.1 for Microsoft Windows,
ScreenRectangle 0., 1024. , 0., 681.
Then we do a substitution on cells that contain "Section" as their second argument (their
style) and finally use NotebookPut to display the resulting notebook in the front end.
In[3]:= NotebookPut nbkernel . Cell str_, "Section" Cell str, "Subsection"
Out[3]= NotebookObject Untitled 4
10.3 Cell data types
1. Here is the notebook object with three ValueBoxes.
In[1]:= nb NotebookPut
Notebook
Cell TextData
"The current version is ", ValueBox "$Version" , "Text" ,
Cell TextData "The operating system is ",
ValueBox "$OperatingSystem" , "Text" ,
Cell TextData "Current user is ", ValueBox "$UserName" , "Text"
Out[1]= NotebookObject Untitled 5
10.4 GridBoxes
1. There are several ways of approaching this problem. One way is to create a function that
contains the formatting rules for the heading.
In[1]:= headstyle str_ :
StyleBox MakeBoxes #, StandardForm & str , FontFamily "Helvetica",
FontWeight "Bold", FontColor RGBColor 0, 0, 1 , FontSize 10 ;
Here are some sample strings for the heading.
In[2]:= headings "first", "second", "third" ;
In[3]:= data " ", " ", " " ,
1.234, 2.3451, 3.4567801 , SqrtBox " " , "
x
y
", " n " ;
Solutions to exercises 523
We now need to create a list of the headings together with their styles and prepend it to the
original data. This way the headings will be the first row of the new data set.
In[4]:= Prepend data, Map headstyle, headings
Out[4]= StyleBox "first", FontFamily Helvetica,
FontWeight Bold, FontColor RGBColor 0, 0, 1 , FontSize 10 ,
StyleBox "second", FontFamily Helvetica, FontWeight Bold,
FontColor RGBColor 0, 0, 1 , FontSize 10 ,
StyleBox "third", FontFamily Helvetica, FontWeight Bold,
FontColor RGBColor 0, 0, 1 , FontSize 10 ,
, , , 1.234, 2.3451, 3.45678 , SqrtBox ,
x
y
, n
In[5]:= ShowTable data_, headings_List : DisplayForm StyleBox
GridBox Prepend data, Map headstyle, headings ,
GridFrame 2, GridFrameMargins 1, 1 , 1, 1 ,
RowLines 1, ColumnLines 1 ,
FontFamily "Times",
Background GrayLevel .8 , SingleLetterItalics True
In[6]:= ShowTable data, "first", "second", "third"
Out[6]//DisplayForm=
first second third
1.234 2.3451 3.4567801
x
y
n
A cleaner approach would be to set up the headings as an option to ShowTable. In addition,
the header formatting should be incorporated into ShowTable. Here is one approach.
In[7]:= Options ShowTable Headings ;
In[8]:= ShowTable data_, opts___?OptionQ : Module headstyle, headings ,
headstyle str_ :
StyleBox MakeBoxes #, StandardForm & str, FontFamily "Helvetica",
FontWeight "Bold", FontColor RGBColor 0, 0, 1 , FontSize 10 ;
headings Headings . Flatten opts . Options ShowTable ;
DisplayForm StyleBox
GridBox Prepend data, Map headstyle, headings ,
GridFrame 2, GridFrameMargins 1, 1 , 1, 1 ,
RowLines 1, ColumnLines 1 ,
FontFamily "Times",
Background GrayLevel .8 , SingleLetterItalics True
524 An Introduction to Programming with Mathematica
In[9]:= ShowTable data, Headings "premier", "deuxieme", "troisieme"
Out[9]//DisplayForm=
premier deuxieme troisieme
1.234 2.3451 3.4567801
x
y
n
3. First, here is the table of all possible truth values for three variables. We will generalize this
below.
In[10]:= ins Distribute Table True, False , 3 , List, List, List
Out[10]= True, True, True , True, True, False ,
True, False, True , True, False, False , False, True, True ,
False, True, False , False, False, True , False, False, False
Here is the logical expression.
In[11]:= expr Implies Or A, B , C
Out[11]= Implies A B, C
This creates a set of rules for all possible truth value combinations.
In[12]:= vars A, B, C ;
Map Thread vars # &, ins
Out[13]= A True, B True, C True , A True, B True, C False ,
A True, B False, C True , A True, B False, C False ,
A False, B True, C True , A False, B True, C False ,
A False, B False, C True , A False, B False, C False
And here we substitute these rules into the logical expression we are working with.
In[14]:= expr . %
Out[14]= True, False, True, False, True, False, True, True
Here then is the TruthTable function.
In[15]:= TruthTable expr_, vars_ : Module len Length vars , n ,
ins Distribute Table True, False , len , List, List, List ;
res expr . Thread vars #1 & ins;
DisplayForm GridBox Prepend Transpose Append Transpose ins ,
If MemberQ True, False , #1 , " ", #1 & res .
True "T", False "F" , Append vars,
TraditionalForm expr , GridFrame True,
RowLines Prepend Table 0, Length res 1 , 2 ,
ColumnLines Append Table 0, Length vars 1 , 2
Solutions to exercises 525
In[16]:= TruthTable Implies A B, C , A, B, C
Out[16]//DisplayForm=
A B C A B C
T T T T
T T F F
T F T T
T F F F
F T T T
F T F F
F F T T
F F F T
10.5 Buttons
1. Here is the code for the Plot3D template button.
In[1]:= ButtonBox "Plot3D fun, x,xmin,xmax , y,ymin,ymax ", Active True
DisplayForm
Out[1]//DisplayForm=
Plot3D fun, x, xmin, xmax , y, ymin, ymax
Alternately, you can use placeholders.
In[2]:= ButtonBox "Plot3D , , , , , , ", Active True DisplayForm
Out[2]//DisplayForm=
Plot3D , , , , , ,
2. Here is the code to create the Expand button.
In[3]:= ButtonBox "Expand ", Active True,
ButtonStyle "CopyEvaluateCell" DisplayForm
Out[3]//DisplayForm=
Expand
Selecting the expression below and then clicking the Expand button will cause a new input
cell to be created with Expand wrapped around the selected expression; then that cell will be
evaluated to produce the expanded polynomial below.
In[4]:=
5
In[5]:= Expand
5
Out[5]=
5
5
4
10
3 2
10
2 3
5
4 5
5
4
20
3
30
2 2
20
3
5
4
10
3 2
30
2 2
30
2 2
10
3 2
10
2 3
20
3
10
2 3
5
4
5
4 5
526 An Introduction to Programming with Mathematica
3. Here is the code for the palette.
Cell[BoxData[GridBox[{
{
ButtonBox[
RowBox[{"Expand", "[", " ", "]"}],
ButtonStyle->"CopyEvaluateCell",
Active->True, ButtonEvaluator->Automatic]},
{
ButtonBox[
RowBox[{"Factor", "[", " ", "]"}],
ButtonStyle->"CopyEvaluateCell",
Active->True, ButtonEvaluator->Automatic]},
{
ButtonBox[
RowBox[{"Apart", "[", " ", "]"}],
ButtonStyle->"CopyEvaluateCell",
Active->True, ButtonEvaluator->Automatic]},
{
ButtonBox[
RowBox[{"Together", "[", " ", "]"}],
ButtonStyle->"CopyEvaluateCell",
Active->True, ButtonEvaluator->Automatic]}
},
RowSpacings->0,
ColumnSpacings->0]], "Input"]
Here is how the palette looks when the above expression is formatted.
In[6]:=
Expand
Factor
Apart
Together
If you wanted to turn this into a free-standing palette, select the above cell and choose Gener-
ate Palette from Selection from the File menu.
Notice that in the code for the palette, each of the buttons used the same three options,
ButtonStyle, Active, and ButtonEvaluator, with identical values. Using ButtonBox
Options, we can set each value once at the GridBox level and each of the buttons will
inherit the option, This cleans up the code considerably.
Solutions to exercises 527
Cell[BoxData[GridBox[{
{
ButtonBox[
RowBox[{"Expand", "[", " ", "]"}]]},
{
ButtonBox[
RowBox[{"Factor", "[", " ", "]"}]]},
{
ButtonBox[
RowBox[{"Apart", "[", " ", "]"}]]},
{
ButtonBox[
RowBox[{"Together", "[", " ", "]"}]]}
},
RowSpacings->0,
ColumnSpacings->0]], "Input",
ButtonBoxOptions->{ButtonStyle->"CopyEvaluateCell",
Active->True, ButtonEvaluator->Automatic}
]
Finally, here is the formatted palette with an input cell and the result of selecting that input
cell and clicking the Together[ ] button.
In[7]:=
Expand
Factor
Apart
Together
In[8]:=
1
x
1
y
In[9]:= Together
1
x
1
y
Out[9]=
x y
x y
528 An Introduction to Programming with Mathematica
11 Examples and applications
11.1 Manipulating data files
1. We first borrow the options fromReadList that we wish to pass into ReadSolarData.
In[1]:= Options ReadSolarData
WordSeparators ",",
RecordLists True,
RecordSeparators "\r\n", "\n", "\r"
;
In[2]:= ReadSolarData file_, opts___?OptionQ : Module ws, rl, rs, raw, data ,
ws, rl, rs WordSeparators, RecordLists, RecordSeparators .
Flatten opts . Options ReadSolarData ;
raw ReadList file, Word, RecordLists rl,
RecordSeparators rs, WordSeparators ws ;
data Select raw, StringTake # 1 , 1 "\"" & ;
Map ToExpression, data, 2
In[3]:= datafile ToFileName "IPM3", "DataFiles" , "23232.txt"
Out[3]= IPM3\DataFiles\23232.txt
In[4]:= data ReadSolarData datafile ;
Now we can use the GetData function developed in Section 11.1 to select those records that
fall between certain dates.
In[5]:= GetData dat_, m1_, y1_ , m2_, y2_ :
Select dat, # 1 y1 && # 2 m1 # 1 y2 && # 2 m2 &
For example, here are the records between November 1976 and March 1977.
In[6]:= GetData data, 11, 76 , 3, 77
Out[6]= 76, 11, 2.5, 3.5, 3.9, 4.1, 3.6, 3.5, 4.3,
4.6, 4.8, 4.8, 2.7, 2.2, 3.1, 3.2 , 76, 12, 2.3, 3.7, 4.4,
4.7, 4.4, 3.5, 4.6, 5.1, 5.4, 5.5, 3.6, 2.5, 3.8, 4.1 ,
77, 1, 1.9, 2.5, 2.7, 2.8, 2.4, 2.3, 2.8, 3., 3.1, 3.1, 1.5, 1.1, 1.6, 1.7 ,
77, 2, 3.3, 4.5, 4.9, 5.1, 4.3, 4.7,
5.6, 5.9, 6.1, 6.1, 3.4, 3.1, 4.1, 4.2 ,
77, 3, 4.8, 5.8, 6.1, 6., 4.4, 6.8, 7.6, 7.8, 7.7, 7.8, 4.2, 4.8, 5.6, 5.6
2. First we need to read the data using the function created in the previous exercise.
In[7]:= datafile ToFileName "IPM3", "DataFiles" , "23232.txt"
Out[7]= IPM3\DataFiles\23232.txt
Solutions to exercises 529
In[8]:= data ReadSolarData datafile ;
Here is the first row of the extracted data.
In[9]:= data 1
Out[9]= 61, 1, 1.6, 1.9, 2., 2., 1.6, 1.7, 2., 2., 2.1, 2.1, 0.7, 0.5, 0.8, 0.8
For 1961, we want to extract the elements in the sixth column and add them; then repeat for
each of the successive years. For example, this function selects all those rows that start with 61,
then pulls off all sixth column elements.
In[10]:= Part Select data, # 1 61 & , All, 6
Out[10]= 2., 4.4, 5.1, 6.4, 5.7, 6., 6.3, 6.2, 6.8, 5.9, 4.2, 2.3
In[11]:= Length %
Out[11]= 12
The total solar radiation for 1961 is the sum of these values.
In[12]:= Apply Plus, Part Select data, # 1 61 & , All, 6
Out[12]= 61.3
Here are the minimum and maximum years (from the first column).
In[13]:= ymin, ymax Map Min # , Max # &, Part data, All, 1 , 0
Out[13]= 61, 90
In[14]:= yearlydata Table
Apply Plus, Part Select data, # 1 y & , All, 6 , y, ymin, ymax
Out[14]= 61.3, 63.1, 56.9, 63.9, 61.2, 63.6, 61.1, 60.7, 63.2,
62.4, 63.5, 61., 60.7, 63.4, 63.6, 66.6, 62.9, 62.2, 63.4,
63.1, 61.6, 58.8, 55.7, 63.1, 62., 63., 64.5, 65.3, 64.5, 67.4
In[15]:= Graphics`MultipleListPlot`
In[16]:= MultipleListPlot yearlydata, SymbolShape Stem ;
5 10 15 20 25 30
10
20
30
40
50
60
530 An Introduction to Programming with Mathematica
3. We first load the necessary packages.
In[17]:= Graphics`MultipleListPlot`
In[18]:= << Utilities`FilterOptions`
In[19]:= PlotSolarData dat1_, dat2_, opts___?OptionQ : Module months ,
months 1, "Jan" , 2, "Feb" , 3, "Mar" ,
4, "Apr" , 5, "May" , 6, "Jun" , 7, "Jul" , 8, "Aug" ,
9, "Sep" , 10, "Oct" , 11, "Nov" , 12, "Dec" ;
MultipleListPlot dat1, dat2, FilterOptions MultipleListPlot, opts ,
PlotJoined True, AspectRatio Automatic,
Ticks months, Automatic , AxesLabel None, "kWh m
2
day" ;
Read in the file.
In[20]:= datafile ToFileName "IPM3", "DataFiles" , "23232.txt"
Out[20]= IPM3\DataFiles\23232.txt
Use ReadSolarData from the previous exercise to strip out all lines that start with a quote
character and insure each element is a number.
In[21]:= data ReadSolarData datafile ;
Using GetData developed in Section 11.1, we extract the sixth column for all dates between
January 1980 and December 1980.
In[22]:= d1 Part GetData data, 1, 80 , 12, 80 , All, 6
Out[22]= 2.7, 3.6, 5.9, 5.7, 5.9, 5.9, 6.1, 6.7, 6.8, 6., 4.7, 3.1
Similarly for data collected in 1981.
In[23]:= d2 Part GetData data, 1, 81 , 12, 81 , All, 6
Out[23]= 2.4, 4.3, 4.7, 6.2, 6.2, 6.2, 6.3, 6.8, 6.8, 5.8, 3.7, 2.2
Finally, here is the plot.
In[24]:= PlotSolarData d1, d2
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
4
5
6
kWh m
2
day
Out[24]= Graphics
Solutions to exercises 531
11.2 Random walks
1. First defining walk1DOffLattice and then inserting an If statement immediately follow-
ing dim==1 will do the trick.
In[1]:= walk1DOffLattice n_ :
FoldList Plus, 0, Table Random Real, 1, 1 , n
In[2]:= walk1D n_ : NestList # 1
Random Integer
&, 0, n
In[3]:= walk2D n_ :
Module NSEW 0, 1 , 1, 0 , 0, 1 , 1, 0 ,
FoldList Plus, 0, 0 ,
NSEW Table Random Integer, 1, 4 , n
In[4]:= walk2DOffLattice n_ :
FoldList Plus, 0, 0 ,
Map Cos # , Sin # &, Table Random Real, 0, 2 , n
In[5]:= walk3D n_ : Module NSEW3 2 Vertices Cube ,
FoldList Plus, 0, 0, 0 , NSEW3 Table Random Integer, 1, 8 , n
In[6]:= walk3DOffLattice n_ : FoldList Plus, 0, 0, 0 ,
Map Cos # , Sin # ,
#
2
&, Table Random Real, 2 , 2 , n
In[7]:= Options RandomWalk LatticeWalk True, Dimension 2
Out[7]= LatticeWalk True, Dimension 2
In[8]:= RandomWalk n_, opts___?OptionQ : Module dim, latticeQ ,
If Not IntegerQ n && n 0 ,
Message RandomWalk::rwn, n , latticeQ, dim
LatticeWalk, Dimension . Flatten opts, Options RandomWalk ;
Which
dim 1, If latticeQ, walk1D n , walk1DOffLattice n ,
dim 2, If latticeQ, walk2D n , walk2DOffLattice n ,
dim 3, If latticeQ, walk3D n , walk3DOffLattice n
2. The output fromRandomWalk with the option Dimension set to 1 is a one-dimensional list
of integers.
In[9]:= RandomWalk 10, Dimension 1
Out[9]= 0, 1, 2, 1, 2, 3, 2, 1, 0, 1, 0
532 An Introduction to Programming with Mathematica
This list can be passed directly to ListPlot.
In[10]:= ShowWalk coords_, opts___ : Which
Length Dimensions coords 1,
ListPlot coords, opts, PlotJoined True ,
Dimensions coords 2 2,
Show Graphics Line coords , opts, AspectRatio Automatic ,
Dimensions coords 2 3,
Show Graphics3D Line coords , opts, AspectRatio Automatic
In[11]:= ShowWalk RandomWalk 1000, Dimension 1 ;
200 400 600 800 1000
30
25
20
15
10
5
3. First we write down the formulas for representing a point on the unit sphere.
In[12]:= x _, _ : 1 Cos
2
Cos ;
y _, _ : 1 Cos
2
Sin ;
z _ : Cos ;
This checks that the formulas are correct:
In[15]:= Simplify x ,
2
y ,
2
z
2
Out[15]= 1
The next step is to create a pair of angles between 0 and 2 .
In[16]:= ran Table Random Real, 0, 2 , 2
Out[16]= 3.65812, 1.45235
This applies the functions x, y, and z to this pair of angles.
In[17]:= Apply x #1, #2 , y #1, #2 , z #1 &, ran
Out[17]= 0.0583615, 0.490403, 0.869539
Solutions to exercises 533
Starting at the origin and folding Plus across this function gives the following.
In[18]:= n 3;
FoldList Plus, 0, 0, 0 , Apply x #1, #2 , y #1, #2 , z #1 &,
Table Random Real, 0, 2 , n , 2 , 1
Out[19]= 0, 0, 0 , 0.0583035, 0.166546, 0.984308 ,
0.86393, 0.535037, 1.10136 , 0.64296, 0.540758, 0.126093
Here then is the rewritten walk3DOffLattice with this code inserted.
In[20]:= <<Graphics`Polyhedra`
In[21]:= walk3DOffLattice n_ : Module x, y, z ,
x _, _ : 1 Cos
2
Cos ;
y _, _ : 1 Cos
2
Sin ;
z _ : Cos ;
FoldList Plus, 0, 0, 0 , Apply x #1, #2 , y #1, #2 , z #1 &,
Table Random Real, 0, 2 , n , 2 , 1
In[22]:= ShowWalk
RandomWalk 2500, Dimension 3, LatticeWalk False ;
5.
In[23]:= AnimateWalk coords_, opts___ : Scan
Show Graphics RGBColor 1, 0, 0 , PointSize .02 , Point coords #1 ,
Line Take coords, #1 , opts, AspectRatio Automatic,
PlotRange Map Min #1 .2, Max #1 .2 &, Transpose coords &,
Range 2, Length coords
In[24]:= AnimateWalk RandomWalk 50, LatticeWalk False ;
534 An Introduction to Programming with Mathematica
11.3 The Game of Life
1.
In[1]:= Options LifeGraphics
Colors 1 RGBColor 1, 0, 0 , 0 RGBColor 0, 0, 0 ;
In[2]:= LifeGraphics lis_, opts___?OptionQ : Module colors ,
colors Colors . Flatten opts, Options LifeGraphics ;
Map
Graphics RasterArray
Reverse # . colors ,
AspectRatio Automatic &, lis
In[3]:= Options LifeGraphics
Out[3]= Colors 1 RGBColor 1, 0, 0 , 0 RGBColor 0, 0, 0
In[4]:= LifeGame n_Integer?Positive, steps_ :
Module gameboard, liveNeighbors, update ,
gameboard Table Random Integer , n , n ;
liveNeighbors mat_ : Apply Plus, Map RotateRight mat, # &,
1, 1 , 1, 0 , 1, 1 ,
0, 1 , 0, 1 , 1, 1 , 1, 0 , 1, 1 ;
update 1, 2 : 1;
update _, 3 : 1;
update _, _ : 0;
SetAttributes update, Listable ;
FixedPointList update #, liveNeighbors # &, gameboard, steps
In[5]:= Show Last LifeGraphics LifeGame 10, 5 ;
In[6]:= Graphics`Colors`
Solutions to exercises 535
In[7]:= Show Last LifeGraphics LifeGame 10, 5 ,
Colors 0 Blue, 1 Green
Out[7]= Graphics
In[8]:= Utilities`FilterOptions`
In[9]:= Options LifeGraphics
Colors 1 RGBColor 1, 0, 0 , 0 RGBColor 0, 0, 0 ;
In[10]:= AnimateLife lis_, opts___?OptionQ :
Scan Show, LifeGraphics lis, FilterOptions LifeGraphics, opts
2. We will use essentially the same function as before, but we will overload the function by
providing a definition for the case when a third argument is provided.
In[11]:= LifeGame n_, steps_, lifeform_List :
Module init Table 0, n , n , gameboard, liveNeighbors, update ,
gameboard ReplacePart init, 1, lifeform ;
liveNeighbors mat_ : Apply Plus, Map RotateRight mat, # &, 1, 1 ,
1, 0 , 1, 1 , 0, 1 , 0, 1 , 1, 1 , 1, 0 , 1, 1 ;
update 1, 2 : 1;
update _, 3 : 1;
update _, _ : 0;
Attributes update Listable;
FixedPointList update #, liveNeighbors # &, gameboard, steps
If LifeGame is called with two arguments, then the definition given earlier will be applied
(random initial game board). If LifeGame is called with three arguments, then this definition
above will be matched.
Here is a game played on a 50 50 board, starting with a glider object initially at lattice site
(20, 20), and played for ten generations.
In[12]:= glider x_, y_ : x, y , x 1, y , x 2, y , x 2, y 1 , x 1, y 2
In[13]:= lg50 LifeGame 50, 10, glider 20, 20 ;
This game could then be animated by evaluating AnimateLife[lg50].
536 An Introduction to Programming with Mathematica
12 Writing packages
12.5 Writing your own packages
1. Here are the definitions for the auxiliary collatz function.
In[1]:= collatz n_?EvenQ : n 2
In[2]:= collatz n_?OddQ : 3 n 1
2. This is essentially the definition given in the solution to Exercise 5 from Section 5.3.
In[3]:= CollatzSequence n_ : NestWhileList collatz, n, # 1 &
In[4]:= CollatzSequence 7
Out[4]= 7, 22, 11, 34, 17, 52, 26, 13, 40, 20, 10, 5, 16, 8, 4, 2, 1
3. First we write the usage message for CollatzSequence, our public function. Notice that we
write no usage message for the private collatz function.
In[5]:= CollatzSequence::usage
"CollatzSequence n computes the sequence of Collatz iterates
starting with initial value n. The sequence terminates
as soon as it reaches the value 1.";
Here is the warning message that will be issued whenever CollatzSequence is passed an
argument that is not a positive integer.
In[6]:= CollatzSequence::notint
"First argument, `1`, to CollatzSequence must be a positive integer.";
4. Here is the modified definition which now issues the warning message created in Exercise 3
whenever the argument n is not a positive integer.
In[7]:= CollatzSequence n_ :
If IntegerQ n && n 0,
NestWhileList collatz, n, # 1 & ,
Message CollatzSequence::notint, n
The following case covers the situation when CollatzSequence is passed two or more argu-
ments. Note that it uses the built-in argx message, which is issued whenever built-in func-
tions are passed the wrong number of arguments.
In[8]:= CollatzSequence _, args__ ; Message
CollatzSequence::argx, CollatzSequence, Length args 1 : Null
Solutions to exercises 537
5. The package begins by giving usage messages for every exported function. The functions to be
exported are mentioned here before the subcontext Private` is entered so that name
CollatzSequence has context Collatz`. Notice that collatz is not mentioned here and
hence will not be accessible to the user of this package.
In[9]:= Quit
In[1]:= BeginPackage "IPM3`Collatz`" ;
In[2]:= CollatzSequence::usage
"CollatzSequence n computes the sequence of Collatz iterates
starting with initial value n. The sequence terminates
as soon as it reaches the value 1.";
In[3]:= CollatzSequence::notint
"First argument, `1`, to CollatzSequence must be a positive integer.";
A new context IPM3`Collatz`Private` is then begun within IPM3`Collatz. All of the
definitions of this package are given within this new context. The context IPM3`Collatz`
CollatzSequence is defined within the System` context. The context of collatz, on
the other hand, is IPM3`Collatz`Private`.
In[4]:= Begin "`Private`" ;
In[5]:= collatz n_?EvenQ : n 2
In[6]:= collatz n_?OddQ : 3 n 1
In[7]:= CollatzSequence n_ :
If IntegerQ n && n 0,
NestWhileList collatz, n, # 1 & ,
Message CollatzSequence::notint, n
In[8]:= CollatzSequence _, args__ ; Message
CollatzSequence::argx, CollatzSequence, Length args 1 : Null
In[9]:= End ;
In[10]:= EndPackage
After the End[] and EndPackage[] functions are evaluated, $Context and $Context
Path revert to whatever they were before, except that IPM3`Collatz` is added to $Con
textPath. Users can refer to CollatzSequence using its short name, but they can only
refer to the auxiliary function collatz by its full name. The intent is to discourage clients
from using collatz at all, and doing so should definitely be avoided, since the author of the
package may change or remove auxiliary definitions at a later time.
538 An Introduction to Programming with Mathematica
Index
|, Alternatives, 159
___, BlankNullSequence, 153
__, BlankSequence, 153
;, CompoundExpression, 22
/;, Condition, 133
/;, ReplaceAll, 164
//., ReplaceRepeated, 166
:>, RuleDelayed, 165
#, Slot (argument to pure function), 102
? (information escape), 26
^^ (number base), 227
` (number mark), 237
->, Rule, 165
Abbott, Paul, 279
Aborting computations, 19
Abs, 225
AbsoluteDashing, 273
AbsoluteThickness, 273
Accuracy, 235
Affine transformations, 175
Alternatives, and patterns, 156
Alternatives (|), 159
Animations, 354
Anonymous functions, 102
Append, 65
Apply, 82
level specification, 83
Approximate numbers, 225
Argument checking, 139
Argument in complex number, Arg, 225
ArrayDepth, 59
ArrayPlot, 175, 249, 476
ArrayRules, 247
Arrays
packed, 250
sparse, 247
Artificial life (a-life), 366
ASCII codes, 72, 204
AspectRatio, 275
Attributes
setting, (SetAttributes), 81
threading across expressions (Listable), 81
AuthorTools, 365
Auto-correlation, 302
Auxiliary functions, 96
Bach, J.S., 305
BaseForm, 227
Begin, 404
BeginPackage, 407
Binary shifts, 232
Binary trees, 202
Bisection method for root finding, 129
Bitwise logical operators, BitXor, 108
BlankNullSequence (___), 153
Blanks, 151
BlankSequence (__), 153
Block, 99
Bowling program, 10
BoxData, 321
Boxes
FractionBox, 322
SqrtBox, 322
SubsuperscriptBox, 322
superscripts, 321
Brownian motion, 303
Browser categories, 364
ButtonBox, 332
ButtonCell, 338
ButtonData, 334
ButtonEvaluator, 335
ButtonFunction, 335
ButtonNotebook, 338
Buttons
actions, 334
activating, 333
as templates, 333
creating from menus, 332
embedding code, 335
evaluation options, 335
front end parsing, 336
hyperlinks, 334
placeholders, 333
structure, 332
using front end commands, 340
ButtonStyle, 334
Caesar cipher, 171
Calkins, Harry, xx
Cartesian products, 169
Cascading Ifs, (Which), 136
Cases, 152, 279
Cell, 314
Cell brackets, 17
Cell expressions, 314
BoxData, 321
embedding evaluations, 320
GraphicsData, 322
options, 315
TextData, 320
Cellular automata
evolution of, 169
visualizing, CAGraphics, 169
Character (ASCII) codes, 72
Characters, 71
Chi-square test, 234
Ciphers, 170
Circle, 271
Clearing values, (Clear), 81
Closed paths, 287
CMYKColor, 273
Codes
fixed-length, 205
variable-length, 205
Collatz numbers, 163, 219
Collatz sequences, 142
Color wheel, 278
Combining plots, 8
Complement, 69
Complex numbers, 225
conjugate, 225
internal representation, 226
magnitude, 225
phase angle, 225
plotting, 295
real and imaginary parts, 225
Compound functions, 96
CompoundExpression (;), 22
Compressing lists, 131
Computations
aborting, 19
interrupting, 19
numeric, 1
symbolic, 2
Condition (/;), 133
Condition number of a matrix, 266
Conditional definitions, 133
Conditional functions, (If), 131
Conditional pattern matching, 156
Conditions, in patterns, 158
Conjugate, 225
Context, current, 403
Context, 403
Context-free grammars, 376
Contexts, 401
540 An Introduction to Programming with Mathematica
exiting subcontexts, 402
Global`, 401
of a symbol, 403
search path, 401
starting new, 404
Converting date formats, 101
Convex hull, 296
Convex polygons, 296
Counting change, 111, 167
DampingFactor, 117
Dashing, 273
Data
converting to different formats, 347
determining structure, 345
extracting parts, 347
fitting to a model, 7
importing, 6, 342
plotting, 6, 282
plotting log-log, 7
removing outliers, 103
selecting based on criteria, 348
solar radiation, 341
visualizing, 348
Date, 320
Dealing cards, 94
Debugging, 420
DeclarePackage, 400
Decoding, run-length (runDecode), 192
Default values, 357
Definitions, multiple associated with a symbol, 134
Delete, 63
Derivatives, programming symbolic, 193
Diameter of point set, 105
DigitCharacter, 162
Digits of numbers, 227
Dimensions, 58
Directives, 272
Disk, 271
Display channels, 323
DisplayForm, 325
Distance function, 95, 105
Distribute, 105
Divergence (div), 85
Do loops, (Do), 117
return values, 120
Documentation
creating for applications, 363
directory structure, 363
Dot product, (Dot), 84
Drop, 63
Dynamic programming, 215
DAndria, Lou, xx
Efficiency issues
programs, 125
recursion, 188
Encoding
characters, 207
run-length, 186
run-length (user-defined split), 190
run-length (with Split), 187
strings, 207
Encoding text, 170
Encryption schemes, 170
End, 402
EndPackage, 407
Entering input, 18
fractions, 19
superscripts, 20
Epilog, 283
Eratosthenes, sieving, 142
Error checking, using Message, 295
Error-trapping, 361
Errors, 423
arguments to functions, 26
mispelling, 25
syntax, 26
Euclidean algorithm, 129
Evaluating input, 16
Evaluation, order of, 417
EvaluationMonitor, 128, 256
EvaluationNotebook, 318
Exact numbers, 238
Index 541
Expressions
displaying structure, (TreeForm), 59
evaluation of, 417
Factor, 2
Fibonacci, Leonardo, 177
Fibonacci numbers, 128, 177
FilterOptions, 282, 510
FindFit, 7
FindRoot, 116
First, 63
Fitting data to a model, 7
Fixed point iteration, (FixedPoint), 86
Fixed precision numbers, 240
Flatten, 65
Floating point numbers, 225
Fold, 87
FoldList, 87
FontFamily, 276
FontSize, 276
For, 143
FractionBox, 322
Frequency modulation (FM) synthesis, 306
FromCharacterCode, 72
Front end, 24
Front end tokens, 340
FrontEndExecute, 340
Function, 102
Functional programming, features, 77
Functions
adding options, 357
anonymous, 102
arguments, 88
assignments, 89
auxiliary, 96
checking arguments, 139
compound, 96
iteration, 86
listening to, 299
localizing constants, (With), 99
localizing names, (Module), 98
localizing values, (Block), 99
nested calls, 89
overloading, 174, 502
piecewise, 138
pure, 102
syntax, 88
user-defined, 88
Game of Life, 366
animating, 373
gliders, 374
visualizing, 372
Gardner, Martin, 302
Gaussian elimination, 200, 260
Global context, 401
Golden ratio, 275
Graphics
directives, 272
displaying to new window, 323
options, 273
primitives, 270
programming, 269
structure of, 269
styles in text, 276
text in, 276
Graphics, 272
GraphicsData, 322
GrayLevel, 273
Greatest common divisor, Euclidean algorithm, 129
GridBoxes
displaying, 325
formatting, 325
options, 325
structure, 324
Hamming distance, 107
Handles, 312
Hayes, Allan, 232
Help Browser, 27
browser categories, 364
documentation, 363
Hilbert matrices, 200, 263
Horners method, 106
542 An Introduction to Programming with Mathematica
Hue, 273
Huffman encoding, 204
Hunt, Andy, xx
Hyperlinks, creating, 334
If, 131
IgnoreCase, 454
Ill-conditioned matrices, 260, 263
Im, 225
Import, 6, 342
In prompt, 18
Infix operator, 23
Inner products, generalized, Inner, 84
Input
entering, 18
evaluating, 18
infix operator, 23
postfix operator, 23
prefix operator, 23
syntax, 23
traditional representations, 24
Inputs, syntax, 19
Insert, 65
IntegerDigits, 227
Integers
extended precision, 239
machine, 239
word size, 239
Integrate, 3
Interleaving lists, 183
Intermediate Value Theorem, 129
Interrupting computations, 19
Intersection, 24, 69
IPM3 packages
how to install, xviii
how to load, xix
where to find, xviii
Irrational numbers, listening to, 301
Iteration
fixed point, (FixedPoint), 86
functions, 86
functions with two arguments, (Fold), 87
Join, 69
Josephus problem, 109, 233
Kernel, 24
killing, 19
Knapp, Rob, xx
Last, 63
Length, 58
LetterCharacter, 162
Level specifications, Map and Apply, 83
Lexical analysis, 378
Lichtblau, Dan, xx
Line, 271
Linear congruential method, 233
Linear systems, solving by Gaussian elimination, 200
LinearSolve, 200
List, 54
Listable attribute, 81
ListPlot, 6
Lists
combining, (Union), 69
component assignment, 66
concatenating, (Join), 69
creating, 55
discarding elements, 63
displaying, 57
elements, 55
extracting elements, 61
flattening, (Flatten), 65
interleaving, 183
internal form, 54
intersection, (Intersection), 69
locating elements, 60
measuring, 58
notation, 21
partitioning, (Partition), 64
replacing elements, (ReplacePart), 65
reversing order, (Reverse), 64
rotating, 64
sorting, 63, 172
Index 543
transposing, (Transpose), 65
Localizing constants, (With), 99
Localizing names, (Module), 98
Localizing values, (Block), 99
Log-log plots, 7
Loops
Do, 117
While, 123
Lower triangular matrices, 211
Machine numbers, 235
MachinePrecision, 235
Maeder, Roman, xv, xx
Mandelbrot, Benit, 302
Map, 78
level specification, 83
MapThread, 79
MatchQ, 151
Mathematica
evaluating input, 16
features, xi
front end, 24
getting help, 26
Help Browser, 27
kernel, 24
notebooks, 15
quitting session, 18
starting up, 15
Mathematica Information Center, xix
Mathematica newsgroup, xix
Mathematical expressions, traditional representations,
24
MathLink, 25
Matrices
condition number, 266
Hilbert, 200, 263
ill-conditioned, 260, 263
lower triangular, 211
multiplication, (Dot), 84
nonsingular, 262
norms, 266
upper triangular, 211
visualizing, (matrixPlot), 175
MatrixForm, 57
Max norm, 266
MaxIterations, 255
Merge sort, 198
Merging lists, 198
Message, 140
Messaging, 361
Module, 98
Monitoring evaluations, 256
Morse code, 205
Multiclause definitions, 134
Multiple precision numbers, 241
MultipleListPlot, 349
Musical scales
equal tempered C major, 302
pentatonic, 307
N, 234
Name collisions, 398
Named patterns, 151
Names, 398
Nest, 86
NestList, 86
NestWhile, 127
NestWhileList, 127
Newtons method, 9, 116
accelerating, 265
controlling precision and accuracy, 257
Noise, 301
white, 302
Nondeterministic algorithms, 287
Nonsingular matrices, 262
Norms
definition, 266
l , 266
matrix, 266
max, 266
vector, 266
Notebook, 12, 310
Notebook expressions
as objects, 312
creating, 311
544 An Introduction to Programming with Mathematica
evaluating selections, 318
listing open, 313
manipulating, 313
moving around within, 316
options, 313
reading into kernel, 312
structure, 310
NotebookCreate, 313
NotebookGet, 312
NotebookPut, 12, 311
Notebooks, 15
Notebooks, 313
NotebookWrite, 314
Novak, John, xx
complete problems, 287
Number mark (`), 237
NumberQ, 227
Numbers
approximate, 225, 234
arrays of, 247
attributes, 226
bases of, 227
complex, 225
digits of, 227
exact vs. approximate, 238
Fibonacci, 177
fixed precision, 240
machine, 235
multiple precision, 241
random, 229
real, 225
representation of approximate, 236
roundoff error, 242
setting precision, 242
size limits on machine, 240
types, 224
variable precision, 241
Numerical computations, 1
NumericQ, 194, 226
Off, 25
On, 26
Options
adding to functions, 357
extracting values, 359
filtering, 282
graphics, 273
inheriting, 285
OrderedWordQ, 73
Orthogonal polynomials, 263
Out prompt, 18
Outer products, generalized, Outer, 84
Outliers, removing from datasets, 103
OutputForm, 152
Overloading function definitions, 502
Overloading functions, 174
, 287
Packages, 395
automatic loading, 400
BaseConvert package, 414
contexts, 401
determining contents, 397
displaying names, 398
exporting functions for public use, 409
importing other packages, 408
loading, 396
localizing names, 395
manipulating contexts, 406
name collisions, 398
notation, 396
removing names, 399
shadowing errors, 398
usage statements, 409
Packed arrays, 250
identifying, 251
memory savings, 251
speed improvements, 251
working with built-in functions, 252
PackedArrayQ, 251
Palindromes, 73
Parametric functions, plotting, 5
ParametricPlot3D, 5
Parse trees, 378, 386
Index 545
Part, 61
Partial pivoting, 262
Partition, 64
Pascals triangle, displaying traditionally, 327
Patterns
alternatives, 156
attaching a condition, 158
defining, 151
matching, (Cases), 152
matching, (MatchQ), 151
matching sequences, 153
named, 151
string, 161
Perfect numbers, searching for, 101
Perfect shuffle, 93
PerfectQ, 8
PerfectSearch, 8
Permutations, random, (randomPermutation), 172
Pick, 454
Picture-Description Language (PDL), 374
Piecewise, 138
Pivoting, 262
in solving linear systems, 210
scaled, 267
Play, 299
Plot, 4
Plots, combining, 8
Plotting
complex roots, 295
data, 282
functions of one variable, 4
functions represented parametrically, 5
Point, 271
Points, classifying in plane, 145
PointSize, 273
Polygon, 271
Polygons
convex, 296
regular, 9
Polyhedra, 278
Polynomials
multiplication using Horners method, 106
orthogonal, 263
Position, 60, 155
Postfix input operator, 23
Precedence, arithmetic operators, 20
Precision, 235
PrecisionGoal, 254
Predicates, used in pattern matching, 156
Prefix input operator, 23
Prepend, 65
Prime numbers, computing with Sieve of
Eratosthenes, 142
PrimePi, 144
Printing values, (Print), 120
Printing variables, 422
Procedures, 115
Programming
buttons, 10
efficiency issues, 125
symbolic documents, 10
Programs
abs, 131
addPairs, 181
addTriples, 182
applyChar, 133
areEltsEven, 104
balanced, 213
bisect, 129
CAGraphics, 169
cardDeck, 92, 182
cartesianProduct, 169
chooseWithoutReplacement, 106
coins, 167
Collatz, 163
ComplexListPlot, 295
ComplexRootPlot, 295
complexToPolar, 231
compress, 131
conditionNumber, 266
convertToDate, 101, 114
CountChange, 167
DataPlot, 285
deal, 94, 97, 182
decode, 171
546 An Introduction to Programming with Mathematica
diameter, 105
distance, 95, 105
div, 85
drawSepTree, 294
encode, 171
encodeChar, 207
encodeString, 207
findRoot, 124, 127
findRootList, 127
FindSubsequence, 155
fold (defined using recursion), 220
gcd, 129, 148
HammingDistance, 108
HilbertMatrix, 200, 263
incrementNumbers, 132
interleave, 183
interleave (recursive definition), 183
LifeGame, 372
listsort, 172
LUdecomp1, 212
LUdecomp2, 212
map (defined using recursion), 219
matrixPlot, 175
maxima, 91, 167
maxima (recursive definition), 183
maxPairs, 183
merge, 198
MergeSort, 199
multAllPairs, 183
multPairwise, 181
nest (defined using recursion), 219
newton, 258
numbertree, 213
PalindromeQ, 73
PascalTable, 327
PerfectSearch, 101
PlayTones, 303
PlotSolarData, 350
pocketChange, 112
pointInPolygonQ, 296
pointLoc, 145
PointPlot, 289
prefixMatch, 191
randomPermutation, 121, 172
RandomSparseArray, 253
RandomWalk, 358
ReadSolarData, 350
removeRepetitions, 132
RepUnit, 105, 326
reverse, 130
RootPlot, 281
rotatePlot, 176
rotatePlot3D, 176
rotateRight, 130
rotateRows, 131
runDecode, 192
runEncode, 186
ShowPoints, 283
showTree, 294
ShowWalk, 360
shuffle, 93
Sieve, 144
signum, 141
simpleClosedPath, 291
solvePP, 262
split, 190
stringMemberQ, 207
subsets, 184, 192
sumElements, 181
sumEveryOtherElement, 182
sumOddElements, 182
sumsOfCubes, 232
survivor (to Josephus problem), 110, 233
transpose, 130
TriangleForm, 329
tridiagonalMatrix, 253
TruthTable, 330
Pure functions, 102
Quadrants, 145
Quadratic congruential method, 233
Quantile functions, 230
Quitting Mathematica session, 18
Index 547
Random, 229
Random number generators
linear congruential, 233
middle-square, 234
quadratic congruential, 233
testing, 234
Random numbers, 229
alternate distributions, 230
Random permutations, 121, 172
Random sampling, 220
Random walks
animation, 354
off-lattice, 353
one-dimensional, 351
three-dimensional, 356
two-dimensional, 352
visualizing, 353, 360
Range, 55
Raster, 271
Rational numbers
internal representation, 224
representation, 239
sound of, 301
Re, 225
ReadList, 343
RealDigits, 227
Reap, 422
RecordLists, 344
RecordSeparators, 344
Rectangle, 271
Recursion
base cases, 179
caching values, 215
counting operations, 216
defining functions, 177
dynamic programming, 215
efficiency issues, 188
list functions, 180
remembering values, 215
symbolic computations, 192
tail, (using Rest), 180
Reduce, 3
Regular polygons, 9
RegularExpression, 71, 162
Remove, 399
Removing symbols, (Remove), 81
Rep units, 326
Repeating units, (RepUnit), 105
ReplaceAll (/;), 150, 164
Replacement rules, 164
ReplacePart, 65
ReplaceRepeated (//.), 166
Rest, 63
Reverse, 64
RGBColor, 273
Root finding
FindRoot, 116
Newtons method, 9
visualizing, 279
RotateLeft, 64
RotateRight, 64
Roundoff errors, 242
Rule (->), 165
RuleDelayed (:>), 165, 166
Rules
as options to functions, 149
as output to built-in functions, 149
delayed, 166
immediate, 166
repeated application, 166
transformation, 164
Run-length encoding, 186
SampleRate, 300
Sawtooth waves, 306
Scaled pivoting, 267
Scaling noise, 302
Scott, Dana, xx
Select, 60, 103
SelectionEvaluate, 14, 318
SelectionMove, 13, 316
Separation tree, 293
SetAttributes, 81
548 An Introduction to Programming with Mathematica
SetPrecision, 242
Shadowing errors, 398
Sieve of Eratosthenes, 142
Simple closed paths, 287
Simplify, 3
Simplifying algebraic expressions, 3
Solve, 3
Solving equations
and symbolic derivatives, 259
Newtons method, 257
secant method, 259
Solving linear systems
Gaussian elimination, 200, 260
lower triangular (solveLower), 211
LU-decomposition, 212
pivoting, 210, 262
upper triangular (solveUpper), 211
Sort, 63
Sorting
comparing schemes, 173
lists, 63
listsort, 172
merge sort, 198
points in plane by polar angle, 290
strings, 174
Sound
1 f , 305
auto-correlation, 302
Brownian music, 303
periodic functions, 301
physics of, 298
sampling rates, 300
sawtooth wave, 306
square wave, 306
Sow, 422
Sparse arrays, 247
memory savings, 249
representation, 247
rules, 247
speed improvements, 250
visualizing, 249
SparseArray, 247
Split, 187, 280
SqrtBox, 322
Square waves, 306
StringCases, 71
StringDrop, 70
StringExpression, 161
StringInsert, 71
StringJoin, 71
StringLength, 70
StringMatchQ, 161
StringPosition, 71
StringReplace, 71
StringReverse, 70
Strings
concatenating, 71
converting from ASCII codes, 72
converting to ASCII code, 72
converting to characters, 71
data type, 70
extracting characters, 70
ignoring case of, 454
InputForm of, 70
inserting characters, 71
length, 70
locating characters, 71
regular expressions, 71, 162
replacing characters, 71
reversing, 70
sorting, 174
StringTake, 70
StyleForm, 276
Subsequences, finding in a sequence, 154
Subsets, 184, 192
Subsets, 484
Substitution, ReplaceAll, 150
SubsuperscriptBox, 322
SuperscriptBox, 321
Sutner, Klaus, xx
Switch, 136
Symbolic computations, 2
Symbolic documents, programming, 10
Symbols
Index 549
clearing values, (Clear), 81
removing, (Remove), 81
System parameters, setting, 252
SystemOptions, 252
Table, 56
TableForm, 57
Tail recursion, 180
Take, 62
Text, 271, 276
TextData, 320
The Mathematica Journal, xix
Thickness, 273
Thread, 79, 171
Timing, 109
ToCharacterCode, 72
ToExpression, 347
ToFileName, 6
Tokens, 381
Trace, 80
TracePrint, 178, 421
Tracing evaluation, 420
Tracing evaluations, (Trace), 80
Transformation rules, 164
Transformations, affine, 175
Transpose, 65
Traveling salesman problems, 287
TreeForm, 59
Trees
balanced, 213
binary, 202
depth-first ordering, 213
drawing, 292
finding width, 292
height of, 203
Huffman encoding, 204
labels, 202
nodes, 202
printing, 203
separation, 293
visualizing, 202
weight of node, 209
TrigReduce, 2
Truth tables, constructing, 330
Typeset expressions, entering from keyboard, 20
Union, 69
Upper triangular matrices, 211
Usage messages, 409
User-defined functions, 88
ValueBox, 320
van der Pol equations, 222
Variable precision numbers, 241
Villegas, Robby, xx
von Neumann, John, 234
Which, 136, 359
While loops, (While), 123
White noise, 302
With, 99
Withoff, Dave, xx
WordSeparators, 344
WorkingPrecision, 223, 254
Zizza, Frank, 187
$BaseDirectory, 342, 363
$Context, 403
$ContextPath, 401
$MachinePrecision, 235
$MaxMachineNumber, 240
$MinMachineNumber, 240
$Path, 342
$RecursionLimit, 188, 257
$UserBaseDirectory, 343, 363
550 An Introduction to Programming with Mathematica