0% found this document useful (0 votes)

198 views

Mathematics and Algorithms

Mathematics And Algorithms

Uploaded by

Arengga Pratama

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

198 views

Mathematics and Algorithms

Mathematics And Algorithms

Uploaded by

Arengga Pratama

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 468

Wolfram Mathematica Tutorial Collection

MATHEMATICS AND ALGORITHMS

For use with Wolfram Mathematica 7.0 and later.

For the latest updates and corrections to this manual:

visit reference.wolfram.com
For information on additional copies of this documentation:
visit the Customer Service website at www.wolfram.com/services/customerservice
or email Customer Service at [email protected]
Comments on this manual are welcomed at:
[email protected]

Printed in the United States of America.

15 14 13 12 11 10 9 8 7 6 5 4 3 2

2008 Wolfram Research, Inc.

All rights reserved. No part of this document may be reproduced or transmitted, in any form or by any means,
electronic, mechanical, photocopying, recording or otherwise, without the prior written permission of the copyright
holder.
Wolfram Research is the holder of the copyright to the Wolfram Mathematica software system ("Software") 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 Software unless
pursuant to the terms of a license granted by Wolfram Research or as otherwise authorized by law is an infringement
of the copyright.
Wolfram Research, Inc. and Wolfram Media, Inc. ("Wolfram") make no representations, express,
statutory, or implied, with respect to the Software (or any aspect thereof), including, without limitation,
any implied warranties of merchantability, interoperability, or fitness for a particular purpose, all of
which are expressly disclaimed. Wolfram does not warrant that the functions of the Software will meet
your requirements or that the operation of the Software will be uninterrupted or error free. As such,
Wolfram does not recommend the use of the software described in this document for applications in
which errors or omissions could threaten life, injury or significant loss.
Mathematica, MathLink, and MathSource are registered trademarks of Wolfram Research, Inc. J/Link, MathLM,
.NET/Link, and webMathematica are trademarks of Wolfram Research, Inc. Windows is a registered trademark of
Microsoft Corporation in the United States and other countries. Macintosh is a registered trademark of Apple
Computer, Inc. All other trademarks used herein are the property of their respective owners. Mathematica is not
associated with Mathematica Policy Research, Inc.
Contents

Numbers
Types of Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Numeric Quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Digits in Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Exact and Approximate Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Numerical Precision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Arbitrary-Precision Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Arbitrary-Precision Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Machine-Precision Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Interval Arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Indeterminate and Infinite Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Controlling Numerical Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

Algebraic Calculations
Symbolic Computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Values for Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Transforming Algebraic Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Simplifying Algebraic Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Putting Expressions into Different Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Simplifying with Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Picking Out Pieces of Algebraic Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Controlling the Display of Large Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Using Symbols to Tag Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

Algebraic Manipulation
Structural Operations on Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Finding the Structure of a Polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Structural Operations on Rational Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Algebraic Operations on Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Polynomials Modulo Primes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Symmetric Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Polynomials over Algebraic Number Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Trigonometric Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Expressions Involving Complex Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Logical and Piecewise Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Simplification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Using Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

Manipulating Equations and Inequalities

Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Solving Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
The Representation of Equations and Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Equations in One Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Counting and Isolating Polynomial Roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Algebraic Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Simultaneous Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Generic and Non-Generic Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Eliminating Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Relational and Logical Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Solving Logical Combinations of Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Equations and Inequalities over Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
The Representation of Solution Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Quantifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Minimization and Maximization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

Linear Algebra
Constructing Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Getting and Setting Pieces of Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Scalars, Vectors and Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
Operations on Scalars, Vectors and Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
Multiplying Vectors and Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Vector Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Matrix Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Basic Matrix Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
Solving Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Eigenvalues and Eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Advanced Matrix Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Sparse Arrays: Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

Series, Limits and Residues

Sums and Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Making Power Series Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
The Representation of Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Operations on Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Operations on Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Composition and Inversion of Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Converting Power Series to Normal Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Solving Equations Involving Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Summation of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Solving Recurrence Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Finding Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
Residues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Pad Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

Calculus
Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Total Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Derivatives of Unknown Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
The Representation of Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Defining Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Indefinite Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Integrals That Can and Cannot Be Done . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Definite Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
Integrals over Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
Manipulating Integrals in Symbolic Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
Integral Transforms and Related Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
Generalized Functions and Related Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

Numerical Operations on Functions

Arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Numerical Mathematics in Mathematica . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
The Uncertainties of Numerical Mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
Introduction to Numerical Sums, Products, and Integrals . . . . . . . . . . . . . . . . . . . . . . . 264
Numerical Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Numerical Evaluation of Sums and Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Numerical Equation Solving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
Numerical Solution of Polynomial Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
Numerical Root Finding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Introduction to Numerical Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Numerical Solution of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Numerical Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Controlling the Precision of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
Monitoring and Selecting Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Functions with Sensitive Dependence on Their Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295

Numerical Operations on Data

Numerical Operations on Data
Basic Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
Discrete Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
Continuous Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Partitioning Data into Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
Using Nearest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
Manipulating Numerical Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
Curve Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
Statistical Model Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Approximate Functions and Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Discrete Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Convolutions and Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Cellular Automata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374

Mathematical Functions
Naming Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
Generic and Nongeneric Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
Numerical Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
Piecewise Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
Pseudorandom Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
Integer and Number Theoretic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
Combinatorial Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
Elementary Transcendental Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Functions That Do Not Have Unique Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
Mathematical Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
Orthogonal Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
Special Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
Elliptic Integrals and Elliptic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
Mathieu and Related Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
Working with Special Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458
Numbers

Types of Numbers
Four underlying types of numbers are built into Mathematica.

Integer arbitrary-length exact integer

Rational integer/integer in lowest terms

Real approximate real number, with any specified precision

Complex complex number of the form number + number I

Intrinsic types of numbers in Mathematica.

Rational numbers always consist of a ratio of two integers, reduced to lowest terms.
In[1]:= 12 344 2222
6172
Out[1]=
1111

Approximate real numbers are distinguished by the presence of an explicit decimal point.
In[2]:= 5456.
Out[2]= 5456.

An approximate real number can have any number of digits.

In[3]:= 4.54543523454543523453452345234543
Out[3]= 4.5454352345454352345345234523454

Complex numbers can have integer or rational components.

In[4]:= 4+78I
7
Out[4]= 4+
8

They can also have approximate real number components.

In[5]:= 4 + 5.6 I
Out[5]= 4 + 5.6
2 Mathematics and Algorithms

123 an exact integer

123. an approximate real number
123.0000000000000 an approximate real number with a certain precision
123.+0. I a complex number with approximate real number
components

Several versions of the number 123.

You can distinguish different types of numbers in Mathematica by looking at their heads.
(Although numbers in Mathematica have heads like other expressions, they do not have explicit
elements which you can extract.)

The object 123 is taken to be an exact integer, with head Integer .

In[6]:= Head@123D
Out[6]= Integer

The presence of an explicit decimal point makes Mathematica treat 123. as an approximate
real number, with head Real .
In[7]:= [email protected]
Out[7]= Real

NumberQ@xD test whether x is any kind of number

IntegerQ@xD test whether x is an integer
EvenQ@xD test whether x is even
OddQ@xD test whether x is odd
PrimeQ@xD test whether x is a prime integer
Head@xD===type test the type of a number

Tests for different types of numbers.

NumberQ@xD tests for any kind of number.

In[8]:= [email protected]
Out[8]= True

5. is treated as a Real , so IntegerQ gives False .

In[9]:= [email protected]
Out[9]= False

If you use complex numbers extensively, there is one subtlety you should be aware of. When
you enter a number like 123., Mathematica treats it as an approximate real number, but as-
sumes that its imaginary part is exactly zero. Sometimes you may want to enter approximate
Mathematics and Algorithms 3

When the imaginary part is the exact integer 0, Mathematica simplifies complex numbers to
real ones.
In[10]:= Head@123 + 0 ID
Out[10]= Integer

Here the imaginary part is only zero to a certain precision, so Mathematica retains the complex
number form.
In[11]:= Head@123. + 0. ID
Out[11]= Complex

The distinction between complex numbers whose imaginary parts are exactly zero, or are only
zero to a certain precision, may seem like a pedantic one. However, when we discuss, for
example, the interpretation of powers and roots of complex numbers in "Functions That Do Not
Have Unique Values", the distinction will become significant.

One way to find out the type of a number in Mathematica is just to pick out its head using
Head@exprD. For many purposes, however, it is better to use functions like IntegerQ which
explicitly test for particular types. Functions like this are set up to return True if their argument
is manifestly of the required type, and to return False otherwise. As a result, IntegerQ@xD will
give False, unless x has an explicit integer value.

Complex Numbers
You can enter complex numbers in Mathematica just by including the constant I, equal to -1 .
Make sure that you type a capital I.

If you are using notebooks, you can also enter I as by typing Esc ii Esc (see "Mathematical
Notation in Notebooks: Numerical Calculations"). The form is normally what is used in output.

Note that an ordinary i means a variable named i, not -1 .

4 Mathematics and Algorithms

This gives the imaginary number result 2 i.

In[1]:= Sqrt@- 4D
Out[1]= 2

This gives the ratio of two complex numbers.

In[2]:= H4 + 3 IL H2 - IL
Out[2]= 1+2

Here is the numerical value of a complex exponential.

In[3]:= Exp@2 + 9 ID N
Out[3]= -6.73239 + 3.04517

x+I y the complex number x + i y

Re@zD real part
Im@zD imaginary part
Conjugate@zD complex conjugate z* or z
Abs@zD absolute value z
Arg@zD the argument j in z ei j

Complex number operations.

Numeric Quantities

NumberQ@exprD test whether expr is explicitly a number

NumericQ@exprD test whether expr has a numerical value

Testing for numeric quantities.

Pi is a symbol, so Pi + 3 is not explicitly a number.

In[1]:= NumberQ@Pi + 3D
Out[1]= False

It does however have a numerical value.

In[2]:= NumericQ@Pi + 3D
Out[2]= True

This finds the explicit numerical value of Pi + 3.

Mathematics and Algorithms 5

This finds the explicit numerical value of Pi + 3.

In[3]:= N@Pi + 3D
Out[3]= 6.14159

Mathematica knows that constants such as Pi are numeric quantities. It also knows that stan-
dard mathematical functions such as Log and Sin have numerical values when their arguments
are numerical.

Log@2 + xD contains x, and is therefore not a numeric quantity.

In[4]:= 8NumericQ@Log@2DD, NumericQ@Log@2 + xDD<
Out[4]= 8True, False<

Many functions implicitly use the numerical values of numeric quantities.

In[5]:= Min@Exp@2D, Log@2D, Sqrt@2DD
Out[5]= Log@2D

In general, Mathematica assumes that any function which has the attribute NumericFunction
will yield numerical values when its arguments are numerical. All standard mathematical func-
tions in Mathematica already have this attribute. But when you define your own functions, you
can explicitly set the attribute to tell Mathematica to assume that these functions will have
numerical values when their arguments are numerical.

Digits in Numbers

IntegerDigits@nD a list of the decimal digits in the integer n

IntegerDigits@n,bD the digits of n in base b
IntegerDigits@n,b,lenD the list of digits padded on the left with zeros to give total
length len
IntegerLength@nD the number of decimal digits in n
IntegerLength@n,bD the number of base b digits in n
IntegerExponent@n,bD the number of zeros at the end of n in base b
RealDigits@xD a list of the decimal digits in the approximate real number
x, together with the number of digits to the left of the
decimal point
6 Mathematics and Algorithms

RealDigits@x,bD the digits of x in base b

RealDigits@x,b,lenD the first len digits of x in base b
RealDigits@x,b,len,nD the first len digits starting with the coefficient of bn
FromDigits@listD construct a number from its decimal digit sequence
FromDigits@list,bD construct a number from its digit sequence in base b
FromDigits@"string"D construct an integer from a string of digits
FromDigits@"string",bD construct an integer from a string of digits in base b
IntegerString@nD a string of the decimal digits in the integer n
IntegerString@n,bD a string of the digits of n in base b

Converting between numbers and lists or strings of digits.

Here is the list of base 16 digits for an integer.

In[1]:= IntegerDigits@1 234 135 634, 16D
Out[1]= 84, 9, 8, 15, 6, 10, 5, 2<

This gives a list of digits, together with the number of digits that appear to the left of the
decimal point.
In[2]:= [email protected]
Out[2]= 881, 2, 3, 4, 5, 6, 7, 8, 9, 0, 1, 2, 3, 4, 5, 6<, 3<

Here is the binary digit sequence for 56, padded with zeros so that it is of total length 8.
In[3]:= IntegerDigits@56, 2, 8D
Out[3]= 80, 0, 1, 1, 1, 0, 0, 0<

This reconstructs the original number from its binary digit sequence.
In[4]:= FromDigits@%, 2D
Out[4]= 56

Here is 56 as a binary string.

In[5]:= IntegerString@56, 2D
Out[5]= 111000

This reconstructs the original number again.

In[6]:= FromDigits@%, 2D
Out[6]= 56
Mathematics and Algorithms 7

b^^nnnn a number in base b

BaseForm@x,bD print with x in base b
IntegerString@n,bD a string representing n in base b

Numbers in other bases.

When the base is larger than 10, extra digits are represented by letters a|z.

The number 100 1012 in base 2 is 37 in base 10.

In[7]:= 2^^100101
Out[7]= 37

This prints 37 in base 2.

In[8]:= BaseForm@37, 2D
Out[8]//BaseForm= 1001012

This gives the base-2 representation as a string.

In[9]:= IntegerString@37, 2D
Out[9]= 100101

Here is a number in base 16.

In[10]:= 16^^ffffaa00
Out[10]= 4 294 945 280

You can do computations with numbers in base 16. Here the result is given in base 10.
In[11]:= 16^^fffaa2 + 16^^ff - 1
Out[11]= 16 776 096

This gives the result in base 16.

In[12]:= BaseForm@%, 16D
Out[12]//BaseForm= fffba016

You can give approximate real numbers, as well as integers, in other bases.
In[13]:= 2^^101.100101
Out[13]= 5.57813

Here are the first few digits of 2 in octal.

8 Mathematics and Algorithms

Here are the first few digits of 2 in octal.

In[14]:= BaseForm@N@Sqrt@2D, 30D, 8D
Out[14]//BaseForm= 1.3240474631771674622042627661154678

This gives an explicit list of the first 15 octal digits.

In[15]:= RealDigits@Sqrt@2D, 8, 15D
Out[15]= 881, 3, 2, 4, 0, 4, 7, 4, 6, 3, 1, 7, 7, 1, 6<, 1<

This gives 15 octal digits starting with the coefficient of 8-10 .

In[16]:= RealDigits@Sqrt@2D, 8, 15, - 10D
Out[16]= 881, 7, 7, 1, 6, 7, 4, 6, 2, 2, 0, 4, 2, 6, 2<, -9<

"Output Formats for Numbers" describes how to print numbers in various formats. If you want
to create your own formats, you will often need to use MantissaExponent to separate the
pieces of real numbers.

MantissaExponent@xD give a list containing the mantissa and exponent of x

MantissaExponent@x,bD give the mantissa and exponent in base b

Separating the mantissa and exponent of numbers.

This gives a list in which the mantissa and exponent of the number are separated.
In[17]:= [email protected] 10 ^ 125D
Out[17]= 80.345, 126<

Exact and Approximate Results

A standard electronic calculator does all your calculations to a particular accuracy, say 10
decimal digits. With Mathematica, however, you can often get exact results.

Mathematica gives an exact result for 2100 , even though it has 31 decimal digits.
In[1]:= 2 ^ 100
Out[1]= 1 267 650 600 228 229 401 496 703 205 376

You can tell Mathematica to give you an approximate numerical result, just as a calculator
would, by ending your input with N. The N stands for "numerical". It must be a capital letter.
"Special Ways to Input Expressions" will explain what the means.
Mathematics and Algorithms 9

This gives an approximate numerical result.

In[2]:= 2 ^ 100 N

Out[2]= 1.26765 1030

Mathematica can give results in terms of rational numbers.

In[3]:= 13+27
13
Out[3]=
21

N always gives the approximate numerical result.

In[4]:= 1 3 + 2 7 N
Out[4]= 0.619048

exprN give an approximate numerical value for expr

Getting numerical approximations.

When you type in an integer like 7, Mathematica assumes that it is exact. If you type in a
number like 4.5, with an explicit decimal point, Mathematica assumes that it is accurate only to
a fixed number of decimal places.

This is taken to be an exact rational number, and reduced to its lowest terms.
In[5]:= 452 62
226
Out[5]=
31

Whenever you give a number with an explicit decimal point, Mathematica produces an approxi-
mate numerical result.
In[6]:= 452.3 62
Out[6]= 7.29516
10 Mathematics and Algorithms

Here again, the presence of the decimal point makes Mathematica give you an approximate
numerical result.
In[7]:= 452. 62
Out[7]= 7.29032

When any number in an arithmetic expression is given with an explicit decimal point, you get an
approximate numerical result for the whole expression.
In[8]:= 1. + 452 62
Out[8]= 8.29032

Numerical Precision
As discussed in "Exact and Approximate Results", Mathematica can handle approximate real
numbers with any number of digits. In general, the precision of an approximate real number is
the effective number of decimal digits in it that are treated as significant for computations. The
accuracy is the effective number of these digits that appear to the right of the decimal point.
Note that to achieve full consistency in the treatment of numbers, precision and accuracy often
have values that do not correspond to integer numbers of digits.

Precision@xD the total number of significant decimal digits in x

Accuracy@xD the number of significant decimal digits to the right of the
decimal point in x

Precision and accuracy of real numbers.

This generates a number with 30-digit precision.

In[1]:= x = N@Pi ^ 10, 30D
Out[1]= 93 648.0474760830209737166901849

This gives the precision of the number.

In[2]:= Precision@xD
Out[2]= 30.

The accuracy is lower since only some of the digits are to the right of the decimal point.
In[3]:= Accuracy@xD
Out[3]= 25.0285

This number has all its digits to the right of the decimal point.
Mathematics and Algorithms 11

This number has all its digits to the right of the decimal point.
In[4]:= x 10 ^ 6
Out[4]= 0.0936480474760830209737166901849

Now the accuracy is larger than the precision.

In[5]:= 8Precision@%D, Accuracy@%D<
Out[5]= 830., 31.0285<

An approximate real number always has some uncertainty in its value, associated with digits
beyond those known. One can think of precision as providing a measure of the relative size of
this uncertainty. Accuracy gives a measure of the absolute size of the uncertainty.

Mathematica is set up so that if a number x has uncertainty d, then its true value can lie any-
where in an interval of size d from x - d 2 to x + d 2. An approximate number with accuracy a is
defined to have uncertainty 10-a , while a nonzero approximate number with precision p is
defined to have uncertainty x 10-p .

Precision@xD -log10 Hd x L

Accuracy@xD -log10 HdL

Definitions of precision and accuracy in terms of uncertainty.

Adding or subtracting a quantity smaller than the uncertainty has no visible effect.
In[6]:= 8x - 10 ^ - 26, x, x + 10 ^ - 26<
Out[6]= 893 648.0474760830209737166901849,
93 648.0474760830209737166901849, 93 648.0474760830209737166901849<

N@expr,nD evaluate expr to n-digit precision using arbitrary-precision

numbers
N@exprD evaluate expr numerically using machine-precision numbers

Numerical evaluation with arbitrary-precision and machine-precision numbers.

Mathematica distinguishes two kinds of approximate real numbers: arbitrary-precision num-

bers, and machine-precision numbers or machine numbers. Arbitrary-precision numbers can
contain any number of digits, and maintain information on their precision. Machine numbers, on
the other hand, always contain the same number of digits, and maintain no information on their
precision.

Here is a machine-number approximation to p.

12 Mathematics and Algorithms

Here is a machine-number approximation to p.

In[7]:= N@PiD
Out[7]= 3.14159

These are both arbitrary-precision numbers.

In[8]:= 8N@Pi, 4D, N@Pi, 20D<
Out[8]= 83.142, 3.1415926535897932385<

As discussed in more detail below, machine numbers work by making direct use of the numeri-
cal capabilities of your underlying computer system. As a result, computations with them can
often be done more quickly. They are however much less flexible than arbitrary-precision num-
bers, and difficult numerical analysis can be needed to determine whether results obtained with
them are correct.

MachinePrecision the precision specification used to indicate machine

numbers
$MachinePrecision the effective precision for machine numbers on your
computer system
MachineNumberQ@xD test whether x is a machine number

Machine numbers.

This returns the symbol MachinePrecision to indicate a machine number.

In[9]:= Precision@N@PiDD
Out[9]= MachinePrecision

On this computer, machine numbers have slightly less than 16 decimal digits.
In[10]:= $MachinePrecision
Out[10]= 15.9546

When you enter an approximate real number, Mathematica has to decide whether to treat it as
a machine number or an arbitrary-precision number. Unless you specify otherwise, if you give
less than $MachinePrecision digits, Mathematica will treat the number as machine precision,
and if you give more digits, it will treat the number as arbitrary precision.
Mathematics and Algorithms 13

123.4 a machine-precision number

123.45678901234567890 an arbitrary-precision number on some computer systems
123.45678901234567890` a machine-precision number on all computer systems
123.456`200 an arbitrary-precision number with 200 digits of precision
123.456``200 an arbitrary-precision number with 200 digits of accuracy
1.234*^6 a machine-precision number in scientific notation
(1.234106 )
1.234`200*^6 a number in scientific notation with 200 digits of precision
2^^101.111`200 a number in base 2 with 200 binary digits of precision

2^^101.111`200*^6 a number in base-2 scientific notation (101.1112 26 )

Input forms for numbers.

When Mathematica prints out numbers, it usually tries to give them in a form that will be as
easy as possible to read. But if you want to take numbers that are printed out by Mathematica,
and then later use them as input to Mathematica, you need to make sure that no information
gets lost.

In standard output form, Mathematica prints a number like this to six digits.
In[11]:= N@PiD
Out[11]= 3.14159

In input form, Mathematica prints all the digits it knows.

In[12]:= InputForm@%D
Out[12]//InputForm= 3.141592653589793

Here is an arbitrary-precision number in standard output form.

In[13]:= N@Pi, 20D
Out[13]= 3.1415926535897932385

In input form, Mathematica explicitly indicates the precision of the number, and gives extra
digits to make sure the number can be reconstructed correctly.
In[14]:= InputForm@%D
Out[14]//InputForm= 3.1415926535897932384626433832795028842`20.
14 Mathematics and Algorithms

This makes Mathematica not explicitly indicate precision.

In[15]:= InputForm@%, NumberMarks -> FalseD
Out[15]//InputForm= 3.14159265358979323846

InputFormAexpr,NumberMarks ->TrueE
use ` marks in all approximate numbers
InputFormAexpr,NumberMarks ->AutomaticE
use ` only in arbitrary-precision numbers
InputFormAexpr,NumberMarks ->FalseE
never use ` marks

Controlling printing of numbers.

The default setting for the NumberMarks option, both in InputForm and in functions such as
ToString and OpenWrite is given by the value of $NumberMarks. By resetting $NumberMarks,
therefore, you can globally change the way that numbers are printed in InputForm.

This makes Mathematica by default always include number marks in input form.
In[16]:= $NumberMarks = True
Out[16]= True

Even a machine-precision number is now printed with an explicit number mark.

In[17]:= InputForm@N@PiDD
Out[17]//InputForm= 3.141592653589793`

Even with no number marks, InputForm still uses * ^ for scientific notation.
In[18]:= InputForm@N@Exp@600D, 20D, NumberMarks -> FalseD
Out[18]//InputForm= 3.7730203009299398234*^260

In doing numerical computations, it is inevitable that you will sometimes end up with results
that are less precise than you want. Particularly when you get numerical results that are very
close to zero, you may well want to assume that the results should be exactly zero. The func-
tion Chop allows you to replace approximate real numbers that are close to zero by the exact
integer 0.
Mathematics and Algorithms 15

Chop@exprD replace all approximate real numbers in expr with magni-

tude less than 10-10 by 0
Chop@expr,dxD replace numbers with magnitude less than dx by 0

Removing numbers close to zero.

This computation gives a small imaginary part.

In[19]:= Exp@N@2 Pi IDD

Out[19]= 1. - 2.44921 10-16

You can get rid of the imaginary part using Chop .

In[20]:= Chop@%D
Out[20]= 1.

Arbitrary-Precision Calculations
When you use N to get a numerical result, Mathematica does what a standard calculator
would do: it gives you a result to a fixed number of significant figures. You can also tell Mathe-
matica exactly how many significant figures to keep in a particular calculation. This allows you
to get numerical results in Mathematica to any degree of precision.

exprN or N@exprD approximate numerical value of expr

N@expr,nD numerical value of expr calculated with n-digit precision

Numerical evaluation functions.

This gives the numerical value of p to a fixed number of significant digits. Typing N@PiD is
exactly equivalent to Pi N .
In[1]:= N@PiD
Out[1]= 3.14159

This gives p to 40 digits.

In[2]:= N@Pi, 40D
Out[2]= 3.141592653589793238462643383279502884197
16 Mathematics and Algorithms

Here is 7 to 30 digits.
In[3]:= N@Sqrt@7D, 30D
Out[3]= 2.64575131106459059050161575364

Doing any kind of numerical calculation can introduce small roundoff errors into your results.
When you increase the numerical precision, these errors typically become correspondingly
smaller. Making sure that you get the same answer when you increase numerical precision is
often a good way to check your results.

The quantity ep 163 turns out to be very close to an integer. To check that the result is not, in
fact, an integer, you have to use sufficient numerical precision.
In[4]:= N@Exp@Pi Sqrt@163DD, 40D
17
Out[4]= 2.625374126407687439999999999992500725972 10

Arbitrary-Precision Numbers
When you do calculations with arbitrary-precision numbers, Mathematica keeps track of preci-
sion at all points. In general, Mathematica tries to give you results which have the highest
possible precision, given the precision of the input you provided.

Mathematica treats arbitrary-precision numbers as representing the values of quantities where

a certain number of digits are known, and the rest are unknown. In general, an arbitrary-
precision number x is taken to have Precision@xD digits which are known exactly, followed by
an infinite number of digits which are completely unknown.

This computes p to 10-digit precision.

In[1]:= N@Pi, 10D
Out[1]= 3.141592654

After a certain point, all digits are indeterminate.

In[2]:= RealDigits@%, 10, 13D
Out[2]= 883, 1, 4, 1, 5, 9, 2, 6, 5, 3, Indeterminate, Indeterminate, Indeterminate<, 1<

When you do a computation, Mathematica keeps track of which digits in your result could be
affected by unknown digits in your input. It sets the precision of your result so that no affected

digits are ever included. This procedure ensures that all digits returned by Mathematica are
correct, whatever the values of the unknown digits may be.
Mathematics and Algorithms 17

digits are ever included. This procedure ensures that all digits returned by Mathematica are
correct, whatever the values of the unknown digits may be.

This evaluates GH1 7L to 30-digit precision.

In[3]:= N@Gamma@1 7D, 30D
Out[3]= 6.54806294024782443771409334943

The result has a precision of exactly 30 digits.

In[4]:= Precision@%D
Out[4]= 30.

If you give input only to a few digits of precision, Mathematica cannot give you such high-
precision output.
In[5]:= N@[email protected], 30D
Out[5]= 6.58965

If you want Mathematica to assume that the argument is exactly 142 1000, then you have to
say so explicitly.
In[6]:= N@Gamma@142 1000D, 30D
Out[6]= 6.58964729492039788328481917496

In many computations, the precision of the results you get progressively degrades as a result of
"roundoff error". A typical case of this occurs if you subtract two numbers that are close
together. The result you get depends on high-order digits in each number, and typically has far
fewer digits of precision than either of the original numbers.

Both input numbers have a precision of around 20 digits, but the result has much lower preci-
sion.
In[7]:= 1.11111111111111111111 - 1.11111111111111111000
-18
Out[7]= 1.1 10

Adding extra digits in one number but not the other is not sufficient to allow extra digits to be
found in the result.
In[8]:= 1.11111111111111111111345 - 1.11111111111111111000
-18
Out[8]= 1.1 10

The precision of the output from a function can depend in a complicated way on the precision of
the input. Functions that vary rapidly typically give less precise output, since the variation of

the output associated with uncertainties in the input is larger. Functions that are close to con-
stants can actually give output that is more precise than their input.
18 Mathematics and Algorithms

the output associated with uncertainties in the input is larger. Functions that are close to con-
stants can actually give output that is more precise than their input.

Here is a case where the output is less precise than the input.
In[9]:= Sin@111 111 111.0000000000000000D
Out[9]= -0.2975351033349432

Here is e-40 evaluated to 20-digit precision.

In[10]:= N@Exp@- 40D, 20D
-18
Out[10]= 4.2483542552915889953 10

The result you get by adding the exact integer 1 has a higher precision.
In[11]:= 1+%
Out[11]= 1.0000000000000000042483542552915889953

It is worth realizing that different ways of doing the same calculation can end up giving you
results with very different precisions. Typically, if you once lose precision in a calculation, it is
essentially impossible to regain it; in losing precision, you are effectively losing information
about your result.

Here is a 40-digit number that is close to 1.

In[12]:= x = N@1 - 10 ^ - 30, 40D
Out[12]= 0.9999999999999999999999999999990000000000

Adding 1 to it gives another 40-digit number.

In[13]:= 1+x
Out[13]= 1.999999999999999999999999999999000000000

The original precision has been maintained.

In[14]:= Precision@%D
Out[14]= 40.301

This way of computing 1 + x loses precision.

In[15]:= Hx ^ 2 - 1L Hx - 1L
Out[15]= 2.000000000

The result obtained in this way has quite low precision.

Mathematics and Algorithms 19

The result obtained in this way has quite low precision.

In[16]:= Precision@%D
Out[16]= 9.69897

The fact that different ways of doing the same calculation can give you different numerical
answers means, among other things, that comparisons between approximate real numbers
must be treated with care. In testing whether two real numbers are "equal", Mathematica
effectively finds their difference, and tests whether the result is "consistent with zero" to the
precision given.

These numbers are equal to the precision given.

In[17]:= 3 == 3.000000000000000000
Out[17]= True

The internal algorithms that Mathematica uses to evaluate mathematical functions are set up to
maintain as much precision as possible. In most cases, built-in Mathematica functions will give
you results that have as much precision as can be justified on the basis of your input. In some
cases, however, it is simply impractical to do this, and Mathematica will give you results that
have lower precision. If you give higher-precision input, Mathematica will use higher precision
in its internal calculations, and you will usually be able to get a higher-precision result.

N@exprD evaluate expr numerically to machine precision

N@expr,nD evaluate expr numerically trying to get a result with n digits
of precision

Numerical evaluation.

If you start with an expression that contains only integers and other exact numeric quantities,
then N@expr, nD will in almost all cases succeed in giving you a result to n digits of precision. You
should realize, however, that to do this Mathematica sometimes has to perform internal interme-
diate calculations to much higher precision.

The global variable $MaxExtraPrecision specifies how many additional digits should be allowed
in such intermediate calculations.

variable default value

$MaxExtraPrecision 50 maximum additional precision to use

Controlling precision in intermediate calculations.

Mathematica automatically increases the precision that it uses internally in order to get the
correct answer here.
20 Mathematics and Algorithms

Mathematica automatically increases the precision that it uses internally in order to get the
correct answer here.
In[18]:= N@Sin@10 ^ 40D, 30D
Out[18]= -0.569633400953636327308034181574

Using the default setting $MaxExtraPrecision = 50, Mathematica cannot get the correct
answer here.
In[19]:= N@Sin@10 ^ 100D, 30D

N::meprec : Internal precision limit $MaxExtraPrecision = 50.` reached while evaluating

Sin@100000000000000000000000000000000000000000000000000000000000000000000000
00000000000000000000000000000D.
Out[19]= 0.

This tells Mathematica that it can use more digits in its internal calculations.
In[20]:= $MaxExtraPrecision = 200
Out[20]= 200

Now it gets the correct answer.

In[21]:= N@Sin@10 ^ 100D, 30D
Out[21]= -0.372376123661276688262086695553

This resets $MaxExtraPrecision to its default value.

In[22]:= $MaxExtraPrecision = 50
Out[22]= 50

Even when you are doing computations that give exact results, Mathematica still occasionally
uses approximate numbers for some of its internal calculations, so that the value of
$MaxExtraPrecision can thus have an effect.

Mathematica works this out using bounds from approximate numbers.

In[23]:= Sin@Exp@100DD > 0
Out[23]= True

With the default value of $MaxExtraPrecision , Mathematica cannot work this out.
In[24]:= Sin@Exp@200DD > 0

N::meprec : Internal precision limit $MaxExtraPrecision = 50.` reached while evaluating -SinA200 E.
200
Out[24]= SinA E > 0

Temporarily resetting $MaxExtraPrecision allows Mathematica to get the result.

Mathematics and Algorithms 21

Temporarily resetting $MaxExtraPrecision allows Mathematica to get the result.

In[25]:= Block@8$MaxExtraPrecision = 100<, Sin@Exp@200DD > 0D
Out[25]= False

In doing calculations that degrade precision, it is possible to end up with numbers that have no
significant digits at all. But even in such cases, Mathematica still maintains information on the
accuracy of the numbers. Given a number with no significant digits, but accuracy a, Mathemat-
ica can then still tell that the actual value of the number must be in the range
8- 10-a , + 10-a < 2. Mathematica by default prints such numbers in the form 0. 10e .

Here is a number with 20-digit precision.

In[26]:= x = N@Exp@50D, 20D
21
Out[26]= 5.1847055285870724641 10

Here there are no significant digits left.

In[27]:= Sin@xD x
-22
Out[27]= 0. 10

But Mathematica still keeps track of the accuracy of the result.

In[28]:= Accuracy@%D
Out[28]= 21.7147

Adding the result to an exact 1 gives a number with quite high precision.
In[29]:= 1 + %%
Out[29]= 1.000000000000000000000

One subtlety in characterizing numbers by their precision is that any number that is consistent
with zero must be treated as having zero precision. The reason for this is that such a number
has no digits that can be recognized as significant, since all its known digits are just zero.

This gives a number whose value is consistent with zero.

In[30]:= d = N@Pi, 20D - Pi
-20
Out[30]= 0. 10
22 Mathematics and Algorithms

The number has no recognizable significant digits of precision.

In[31]:= Precision@dD
Out[31]= 0.

But it still has a definite accuracy, that characterizes the uncertainty in it.
In[32]:= Accuracy@dD
Out[32]= 19.5029

If you do computations whose results are likely to be near zero, it can be convenient to specify
the accuracy, rather than the precision, that you want to get.

N@expr,pD evaluate expr to precision p

N@expr,8p,a<D evaluate expr to at most precision p and accuracy a
NAexpr,9Infinity,a=E evaluate expr to any precision but to accuracy a

Specifying accuracy as well as precision.

Here is a symbolic expression.

In[33]:= u = ArcTan@1 3D - ArcCot@3D
1
Out[33]= -ArcCot@3D + ArcTanB F
3

This shows that the expression is equivalent to zero.

In[34]:= FullSimplify@uD
Out[34]= 0

N cannot guarantee to get a result to precision 20.

In[35]:= N@u, 20D

N::meprec :
1
Internal precision limit $MaxExtraPrecision = 50.` reached while evaluating -ArcCot@3D + ArcTanB F.
3
-71
Out[35]= 0. 10

But it can get a result to accuracy 20.

In[36]:= N@u, 8Infinity, 20<D
-20
Out[36]= 0. 10

When Mathematica works out the potential effect of unknown digits in arbitrary-precision num-
bers, it assumes by default that these digits are completely independent in different numbers.
Mathematics and Algorithms 23

In particular, if two numbers are generated in the same way in a computation, some of their
unknown digits may be equal. Then, when these numbers are, for example, subtracted, the
unknown digits may cancel. By assuming that the unknown digits are always independent,
however, Mathematica will miss such cancellations.

Here is a number computed to 20-digit precision.

In[37]:= d = N@3 ^ - 30, 20D
-15
Out[37]= 4.8569357496188611379 10

The quantity 1 + d has about 34-digit precision.

In[38]:= Precision@1 + dD
Out[38]= 34.3136

This quantity has lower precision, since Mathematica assumes that the unknown digits in each
number d are independent.
In[39]:= Precision@H1 + dL - dD
Out[39]= 34.0126

Numerical algorithms sometimes rely on cancellations between unknown digits in different

numbers yielding results of higher precision. If you can be sure that certain unknown digits will
eventually cancel, then you can explicitly introduce fixed digits in place of the unknown ones.
You can carry these fixed digits through your computation, then let them cancel, and get a
result of higher precision.

SetPrecision@x,nD create a number with n decimal digits of precision, padding

with base-2 zeros if necessary
SetAccuracy @x,nD create a number with n decimal digits of accuracy

Functions for modifying precision and accuracy.

This introduces 10 more digits in d.

In[40]:= d = SetPrecision@d, 30D
-15
Out[40]= 4.85693574961886113790624266497 10

The digits that were added cancel out here.

24 Mathematics and Algorithms

The digits that were added cancel out here.

In[41]:= H1 + dL - d
Out[41]= 1.00000000000000000000000000000000000000000000

The precision of the result is now about 44 digits, rather than 34.
In[42]:= Precision@%D
Out[42]= 44.0126

SetPrecision works by adding digits which are zero in base 2. Sometimes, Mathematica stores
slightly more digits in an arbitrary-precision number than it displays, and in such cases,
SetPrecision will use these extra digits before introducing zeros.

This creates a number with a precision of 40 decimal digits. The extra digits come from conver-
sion to base 10.
In[43]:= [email protected], 40D
Out[43]= 0.4000000000000000222044604925031308084726

variable default value

$MaxPrecision Infinity maximum total precision to be used
$MinPrecision 0 minimum precision to be used

Global precision-control parameters.

By making the global assignment $MinPrecision = n, you can effectively apply

SetPrecision@expr, nD at every step in a computation. This means that even when the number
of correct digits in an arbitrary-precision number drops below n, the number will always be
padded to have n digits.

If you set $MaxPrecision = n as well as $MinPrecision = n, then you can force all arbitrary-
precision numbers to have a fixed precision of n digits. In effect, what this does is to make
Mathematica treat arbitrary-precision numbers in much the same way as it treats machine
numbers~but with more digits of precision.

Fixed-precision computation can make some calculations more efficient, but without careful
analysis you can never be sure how many digits are correct in the results you get.
Mathematics and Algorithms 25

Here is a small number with 20-digit precision.

In[44]:= k = N@Exp@- 60D, 20D
-27
Out[44]= 8.7565107626965203385 10

With Mathematica's usual arithmetic, this works fine.

In[45]:= Evaluate@1 + kD - 1
-27
Out[45]= 8.7565107626965203385 10

This tells Mathematica to use fixed-precision arithmetic.

In[46]:= $MinPrecision = $MaxPrecision = 20
Out[46]= 20

The first few digits are correct, but the rest are wrong.
In[47]:= Evaluate@1 + kD - 1
-27
Out[47]= 8.7565107626963908935 10

Machine-Precision Numbers
Whenever machine-precision numbers appear in a calculation, the whole calculation is typically
done in machine precision. Mathematica will then give machine-precision numbers as the result.

Whenever the input contains any machine-precision numbers, Mathematica does the computa-
tion to machine precision.
In[1]:= 1.4444444444444444444 ^ 5.7
Out[1]= 8.13382

[email protected] yields a machine-precision result, so the N is irrelevant.

In[2]:= N@[email protected], 30D
Out[2]= 1.02338

This gives a higher-precision result.

In[3]:= N@Zeta@56 10D, 30D
Out[3]= 1.02337547922702991086041788103

When you do calculations with arbitrary-precision numbers, as discussed in "Arbitrary-Precision

Numbers", Mathematica always keeps track of the precision of your results, and gives only
those digits which are known to be correct, given the precision of your input. When you do
calculations with machine-precision numbers, however, Mathematica always gives you a
machine-precision result, whether or not all the digits in the result can, in fact, be determined
to be correct on the basis of your input.

This subtracts two machine-precision numbers.

In[4]:= diff = 1.11111111 - 1.11111000

Out[4]= 1.11 10-6

The result is taken to have machine precision.

In[5]:= Precision@diffD
Out[5]= MachinePrecision

Here are all the digits in the result.

In[6]:= InputForm@diffD
Out[6]//InputForm= 1.1099999999153454`*^-6

The fact that you can get spurious digits in machine-precision numerical calculations with Mathe-
matica is in many respects quite unsatisfactory. The ultimate reason, however, that Mathemat-
ica uses fixed precision for these calculations is a matter of computational efficiency.

Mathematica is usually set up to insulate you as much as possible from the details of the com-
puter system you are using. In dealing with machine-precision numbers, you would lose too
much, however, if Mathematica did not make use of some specific features of your computer.

The important point is that almost all computers have special hardware or microcode for doing
floating-point calculations to a particular fixed precision. Mathematica makes use of these
features when doing machine-precision numerical calculations.

The typical arrangement is that all machine-precision numbers in Mathematica are represented
as "double-precision floating-point numbers" in the underlying computer system. On most
current computers, such numbers contain a total of 64 binary bits, typically yielding 16 decimal
digits of mantissa.

The main advantage of using the built-in floating-point capabilities of your computer is speed.
Arbitrary-precision numerical calculations, which do not make such direct use of these capabili-
ties, are usually many times slower than machine-precision calculations.
Mathematics and Algorithms 27

There are several disadvantages of using built-in floating-point capabilities. One already
mentioned is that it forces all numbers to have a fixed precision, independent of what precision
can be justified for them.

A second disadvantage is that the treatment of machine-precision numbers can vary slightly
from one computer system to another. In working with machine-precision numbers, Mathemat-
ica is at the mercy of the floating-point arithmetic system of each particular computer. If float-
ing-point arithmetic is done differently on two computers, you may get slightly different results
for machine-precision Mathematica calculations on those computers.

$MachinePrecision the number of decimal digits of precision

$MachineEpsilon the minimum positive machine-precision number which can

be added to 1.0 to give a result distinguishable from 1.0
$MaxMachineNumber the maximum machine-precision number

$MinMachineNumber the minimum positive machine-precision number

$MaxNumber the maximum magnitude of an arbitrary-precision number

$MinNumber the minimum magnitude of a positive arbitrary-precision

number

Properties of numbers on a particular computer system.

Since machine-precision numbers on any particular computer system are represented by a

definite number of binary bits, numbers which are too close together will have the same bit
pattern, and so cannot be distinguished. The parameter $MachineEpsilon gives the distance
between 1.0 and the closest number which has a distinct binary representation.

This gives the value of $MachineEpsilon for the computer system on which these examples
are run.
In[7]:= $MachineEpsilon

Out[7]= 2.22045 10-16

Although this prints as 1., Mathematica knows that the result is larger than 1.
In[8]:= 1. + $MachineEpsilon
Out[8]= 1.

InputForm reveals that the result is not exactly 1.

28 Mathematics and Algorithms

InputForm reveals that the result is not exactly 1.

In[9]:= % InputForm
Out[9]//InputForm= 1.0000000000000002

Subtracting 1 gives $MachineEpsilon .

In[10]:= % - 1.

Out[10]= 2.22045 10-16

This prints as 1. also.

In[11]:= 1. + $MachineEpsilon 2
Out[11]= 1.

In this case, however, the result is not distinguished from 1. to machine precision.
In[12]:= % InputForm
Out[12]//InputForm= 1.

Subtracting 1 from the result yields 0.

In[13]:= % - 1.
Out[13]= 0.

Machine numbers have not only limited precision, but also limited magnitude. If you generate a
number which lies outside the range specified by $MinMachineNumber and $MaxMachineNumber,
Mathematica will automatically convert the number to arbitrary-precision form.

This is the maximum machine-precision number which can be handled on the computer system
used for this example.
In[14]:= $MaxMachineNumber

Out[14]= 1.79769 10308

Mathematica automatically converts any result larger than $MaxMachineNumber to arbitrary

precision.
In[15]:= 2 $MaxMachineNumber
308
Out[15]= 3.595386269724631 10
Mathematics and Algorithms 29

Here is another computation whose result is outside the range of machine-precision numbers.
In[16]:= [email protected]

Out[16]= 1.970071114017 10434

Interval Arithmetic

Interval@8min,max<D the interval from min to max

Interval@8min1 ,max1 <,8min2 ,max2 <,D
the union of intervals from min1 to max1 , min2 to max2 ,

Representations of real intervals.

This represents all numbers between -2 and +5.

In[1]:= Interval@8- 2, 5<D
Out[1]= Interval@8-2, 5<D

The square of any number between -2 and +5 is always between 0 and 25.
In[2]:= Interval@8- 2, 5<D ^ 2
Out[2]= Interval@80, 25<D

Taking the reciprocal gives two distinct intervals.

In[3]:= 1 Interval@8- 2, 5<D
1 1
Out[3]= IntervalB:-, - >, : , >F
2 5

Abs folds the intervals back together again.

In[4]:= Abs@%D
1
Out[4]= IntervalB: , >F
5

You can use intervals in many kinds of functions.

In[5]:= Solve@3 x + 2 == Interval@8- 2, 5<D, xD
4
Out[5]= ::x IntervalB:- , 1>F>>
3
30 Mathematics and Algorithms

Some functions automatically generate intervals.

In[6]:= Limit@Sin@1 xD, x -> 0D
Out[6]= Interval@8-1, 1<D

IntervalUnion@interval1 ,interval2 ,D
find the union of several intervals
IntervalIntersection@interval1 ,interval2 ,D
find the intersection of several intervals
IntervalMemberQ@interval,xD test whether the point x lies within an interval
IntervalMemberQ@interval1 ,interval2 D
test whether interval2 lies completely within interval1

Operations on intervals.

This finds the overlap of the two intervals.

In[7]:= IntervalIntersection@Interval@83, 7<D, Interval@8- 2, 5<DD
Out[7]= Interval@83, 5<D

You can use Max and Min to find the end points of intervals.
In[8]:= Max@%D
Out[8]= 5

This finds out which of a list of intervals contains the point 7.

In[9]:= IntervalMemberQ@Table@Interval@8i, i + 1<D, 8i, 1, 20, 3<D, 7D
Out[9]= 8False, False, True, False, False, False, False<

You can use intervals not only with exact quantities but also with approximate numbers. Even
with machine-precision numbers, Mathematica always tries to do rounding in such a way as to
preserve the validity of results.

This shows explicitly the interval treated by Mathematica as the machine-precision number 0.
In[10]:= [email protected]
-308
Out[10]= IntervalA9-2.22507 10 , 2.22507 10-308 =E
Mathematics and Algorithms 31

This shows the corresponding interval around 100., shifted back to zero.
In[11]:= [email protected] - 100
-14 -14
Out[11]= IntervalA9-1.42109 10 , 1.42109 10 =E

The same kind of thing works with numbers of any precision.

In[12]:= Interval@N@Pi, 50DD - Pi
-50 -50
Out[12]= IntervalA9-0. 10 , 0. 10 =E

With ordinary machine-precision arithmetic, this computation gives an incorrect result.

In[13]:= Sin@N@PiDD
-16
Out[13]= 1.22461 10

The interval generated here, however, includes the correct value of 0.

In[14]:= Sin@Interval@N@PiDDD
-16 -16
Out[14]= IntervalA9-3.21629 10 , 5.6655 10 =E

Indeterminate and Infinite Results

If you type in an expression like 0 0, Mathematica prints a message, and returns the result
Indeterminate .
In[1]:= 00

1
Power::infy : Infinite expression encountered.
0

::indet : Indeterminate expression 0 ComplexInfinity encountered.

Out[1]= Indeterminate

An expression like 0 0 is an example of an indeterminate numerical result. If you type in 0 0,

there is no way for Mathematica to know what answer you want. If you got 0 0 by taking the
limit of x x as x 0, then you might want the answer 1. On the other hand, if you got 0 0
instead as the limit of 2 x x, then you probably want the answer 2. The expression 0 0 on its
own does not contain enough information to choose between these and other cases. As a result,
its value must be considered indeterminate.

Whenever an indeterminate result is produced in an arithmetic computation, Mathematica

prints a warning message, and then returns Indeterminate as the result of the computation. If

you ever try to use Indeterminate in an arithmetic computation, you always get the result
Indeterminate. A single indeterminate expression effectively "poisons" any arithmetic computa -
tion. (The symbol Indeterminate plays a role in Mathematica similar to the "not a number"
object in the IEEE Floating Point Standard.)

The usual laws of arithmetic simplification are suspended in the case of Indeterminate .
In[2]:= Indeterminate - Indeterminate
Out[2]= Indeterminate

Indeterminate "poisons" any arithmetic computation, and leads to an indeterminate result.

In[3]:= 2 Indeterminate - 7
Out[3]= Indeterminate

When you do arithmetic computations inside Mathematica programs, it is often important to be

able to tell whether indeterminate results were generated in the computations. You can do this
by using the function Check discussed in "Messages" to test whether any warning messages
associated with indeterminate results were produced.

You can use Check inside a program to test whether warning messages are generated in a
computation.
In[4]:= Check@H7 - 7L H8 - 8L, meaninglessD

1
Power::infy : Infinite expression encountered.
0

::indet : Indeterminate expression 0 ComplexInfinity encountered.

Out[4]= meaningless

Indeterminate an indeterminate numerical result

Infinity a positive infinite quantity

-Infinity a negative infinite quantity (DirectedInfinity@- 1D)
DirectedInfinity@rD an infinite quantity with complex direction r
ComplexInfinity an infinite quantity with an undetermined direction
DirectedInfinity@D equivalent to ComplexInfinity

Indeterminate and infinite quantities.

There are many situations where it is convenient to be able to do calculations with infinite
quantities. The symbol Infinity in Mathematica represents a positive infinite quantity. You can

use it to specify such things as limits of sums and integrals. You can also do some arithmetic
calculations with it.
Mathematics and Algorithms 33

use it to specify such things as limits of sums and integrals. You can also do some arithmetic
calculations with it.

Here is an integral with an infinite limit.

In[5]:= Integrate@1 x ^ 3, 8x, 1, Infinity<D
1
Out[5]=
2

Mathematica knows that 1 = 0.

In[6]:= 1 Infinity
Out[6]= 0

If you try to find the difference between two infinite quantities, you get an indeterminate result.
In[7]:= Infinity - Infinity

::indet : Indeterminate expression - + encountered.

Out[7]= Indeterminate

There are a number of subtle points that arise in handling infinite quantities. One of them
concerns the "direction" of an infinite quantity. When you do an infinite integral, you typically
think of performing the integration along a path in the complex plane that goes to infinity in
some direction. In this case, it is important to distinguish different versions of infinity that
correspond to different directions in the complex plane. + and - are two examples, but for
some purposes one also needs i and so on.

In Mathematica, infinite quantities can have a "direction", specified by a complex number.

When you type in the symbol Infinity, representing a positive infinite quantity, this is con-
verted internally to the form DirectedInfinity@1D, which represents an infinite quantity in the
+1 direction. Similarly, - Infinity becomes DirectedInfinity@- 1D, and I Infinity becomes
DirectedInfinity@ID. Although the DirectedInfinity form is always used internally, the
standard output format for DirectedInfinity@rD is r Infinity.

Infinity is converted internally to DirectedInfinity@1D.

In[8]:= Infinity FullForm
Out[8]//FullForm= DirectedInfinity@1D

Although the notion of a "directed infinity" is often useful, it is not always available. If you type
in 1 0, you get an infinite result, but there is no way to determine the "direction" of the infinity

ity. Mathematica represents the result of 1 0 as DirectedInfinity@D. In standard output

form, this undirected infinity is printed out as ComplexInfinity.
34 Mathematics and Algorithms

Mathematica represents the result of 1 0 as DirectedInfinity @D. In standard output

form, this undirected infinity is printed out as ComplexInfinity.

1 0 gives an undirected form of infinity.

In[9]:= 10

1
Power::infy : Infinite expression encountered.
0
Out[9]= ComplexInfinity

Controlling Numerical Evaluation

NHoldAll prevent any arguments of a function from being affected

by N
NHoldFirst prevent the first argument from being affected

NHoldRest prevent all but the first argument from being affected

Attributes for controlling numerical evaluation.

Usually N goes inside functions and gets applied to each of their arguments.
In[1]:= N@f@2 3, PiDD
Out[1]= [email protected], 3.14159D

This tells Mathematica not to apply N to the first argument of f.

In[2]:= SetAttributes@f, NHoldFirstD

Now the first argument of f is left in its exact form.

In[3]:= N@f@2 3, PiDD
2
Out[3]= fB , 3.14159F
3
Mathematics and Algorithms 35

Algebraic Calculations

Symbolic Computation
One of the important features of Mathematica is that it can do symbolic, as well as numerical
calculations. This means that it can handle algebraic formulas as well as numbers.

Here is a typical numerical computation.

In[1]:= 3 + 62 - 1
Out[1]= 64

This is a symbolic computation.

In[2]:= 3x-x+2
Out[2]= 2+2x

Numerical computation 62 + 3 - 1 64
Symbolic computation 3 x + 2 x - x + 22

Numerical and symbolic computations.

You can type any algebraic expression into Mathematica.

In[3]:= -1 + 2 x + x^3

Out[3]= -1 + 2 x + x3

Mathematica automatically carries out basic algebraic simplifications. Here it combines x2 and
-4 x2 to get -3 x2 .
In[4]:= x^2 + x - 4 x^2

Out[4]= x - 3 x2

You can type in any algebraic expression, using the operators listed in "Arithmetic". You can
use spaces to denote multiplication. Be careful not to forget the space in x y. If you type in xy
with no space, Mathematica will interpret this as a single symbol, with the name xy, not as a
product of the two symbols x and y.
36 Mathematics and Algorithms

Mathematica rearranges and combines terms using the standard rules of algebra.
In[5]:= x y + 2 x^2 y + y^2 x^2 - 2 y x

Out[5]= -x y + 2 x2 y + x2 y2

Here is another algebraic expression.

In[6]:= Hx + 2 y + 1L Hx - 2L ^ 2

Out[6]= H-2 + xL2 H1 + x + 2 yL

The function Expand multiplies out products and powers.

In[7]:= Expand@%D

Out[7]= 4 - 3 x2 + x3 + 8 y - 8 x y + 2 x2 y

Factor does essentially the inverse of Expand.

In[8]:= Factor@%D

Out[8]= H-2 + xL2 H1 + x + 2 yL

When you type in more complicated expressions, it is important that you put parentheses in the
right places. Thus, for example, you have to give the expression x4 y in the form x ^ H4 yL. If you
leave out the parentheses, you get x4 y instead. It never hurts to put in too many parentheses,
but to find out exactly when you need to use parentheses, look at "Operator Input Forms".

Here is a more complicated formula, requiring several parentheses.

In[9]:= Sqrt@2D 9801 H4 nL ! H1103 + 26 390 nL Hn ! ^ 4 396 ^ H4 nLL
1
-8 n
22 99-2-4 n H1103 + 26 390 nL H4 nL !
Out[9]=
Hn !L4

When you type in an expression, Mathematica automatically applies its large repertoire of rules
for transforming expressions. These rules include the standard rules of algebra, such as x - x = 0,
together with much more sophisticated rules involving higher mathematical functions.

4
Mathematica uses standard rules of algebra to replace J 1 + x N by H1 + xL2 .
In[10]:= Sqrt@1 + xD ^ 4

Out[10]= H1 + xL2
Mathematics and Algorithms 37

Mathematica knows no rules for this expression, so it leaves the expression in the original form
you gave.
In[11]:= Log@1 + Cos@xDD
Out[11]= Log@1 + Cos@xDD

The notion of transformation rules is a very general one. In fact, you can think of the whole of
Mathematica as simply a system for applying a collection of transformation rules to many
different kinds of expressions.

The general principle that Mathematica follows is simple to state. It takes any expression you
input, and gets results by applying a succession of transformation rules, stopping when it
knows no more transformation rules that can be applied.

Take any expression, and apply transformation rules until the result no longer changes.

The fundamental principle of Mathematica.

Values for Symbols

When Mathematica transforms an expression such as x + x into 2 x, it is treating the variable x
in a purely symbolic or formal fashion. In such cases, x is a symbol which can stand for any
expression.

Often, however, you need to replace a symbol like x with a definite "value". Sometimes this
value will be a number; often it will be another expression.

To take an expression such as 1 + 2 x and replace the symbol x that appears in it with a definite
value, you can create a Mathematica transformation rule, and then apply this rule to the expres-
sion. To replace x with the value 3, you would create the transformation rule x -> 3. You must
type -> as a pair of characters, with no space in between. You can think of x -> 3 as being a
rule in which "x goes to 3".

To apply a transformation rule to a particular Mathematica expression, you type expr . rule. The
"replacement operator" . is typed as a pair of characters, with no space in between.

This uses the transformation rule x -> 3 in the expression 1 + 2 x.

In[1]:= 1 + 2 x . x -> 3
Out[1]= 7

You can replace x with any expression. Here every occurrence of x is replaced by 2 - y.
38 Mathematics and Algorithms

You can replace x with any expression. Here every occurrence of x is replaced by 2 - y.
In[2]:= 1 + x + x ^ 2 . x -> 2 - y

Out[2]= 3 + H2 - yL2 - y

Here is a transformation rule. Mathematica treats it like any other symbolic expression.
In[3]:= x -> 3 + y
Out[3]= x3+y

This applies the transformation rule on the previous line to the expression x ^ 2 - 9.
In[4]:= x ^ 2 - 9 . %

Out[4]= -9 + H3 + yL2

expr.x->value replace x by value in the expression expr

expr.8x->xval,y->yval< perform several replacements

Replacing symbols by values in expressions.

You can apply rules together by putting the rules in a list.

In[5]:= Hx + yL Hx - yL ^ 2 . 8x -> 3, y -> 1 - a<

Out[5]= H4 - aL H2 + aL2

The replacement operator . allows you to apply transformation rules to a particular expres-
sion. Sometimes, however, you will want to define transformation rules that should always be
applied. For example, you might want to replace x with 3 whenever x occurs.

As discussed in "Defining Variables", you can do this by assigning the value 3 to x using x = 3.
Once you have made the assignment x = 3, x will always be replaced by 3, whenever it
appears.

This assigns the value 3 to x.

In[6]:= x=3
Out[6]= 3

Now x will automatically be replaced by 3 wherever it appears.

In[7]:= x^2 - 1
Out[7]= 8

This assigns the expression 1 + a to be the value of x.

Mathematics and Algorithms 39

This assigns the expression 1 + a to be the value of x.

In[8]:= x = 1+a
Out[8]= 1+a

Now x is replaced by 1 + a.
In[9]:= x^2 - 1

Out[9]= -1 + H1 + aL2

You can define the value of a symbol to be any expression, not just a number. You should
realize that once you have given such a definition, the definition will continue to be used when-
ever the symbol appears, until you explicitly change or remove the definition. For most people,
forgetting to remove values you have assigned to symbols is the single most common source of
mistakes in using Mathematica.

x=value define a value for x which will always be used

x=. remove any value defined for x

Assigning values to symbols.

The symbol x still has the value you assigned to it.

In[10]:= x+5-2x
Out[10]= 6 + a - 2 H1 + aL

This removes the value you assigned to x.

In[11]:= x =.

Now x has no value defined, so it can be used as a purely symbolic variable.

In[12]:= x+5-2x
Out[12]= 5 - x

A symbol such as x can serve many different purposes in Mathematica, and in fact, much of the
flexibility of Mathematica comes from being able to mix these purposes at will. However, you
need to keep some of the different uses of x straight in order to avoid making mistakes. The
most important distinction is between the use of x as a name for another expression, and as a
symbolic variable that stands only for itself.

Traditional programming languages that do not support symbolic computation allow variables to
be used only as names for objects, typically numbers, that have been assigned as values for
them. In Mathematica, however, x can also be treated as a purely formal variable, to which
various transformation rules can be applied. Of course, if you explicitly give a definition, such as
40 Mathematics and Algorithms

You should understand that explicit definitions such as x = 3 have a global effect. On the other
hand, a replacement such as expr . x -> 3 affects only the specific expression expr. It is usually
much easier to keep things straight if you avoid using explicit definitions except when abso-
lutely necessary.

You can always mix replacements with assignments. With assignments, you can give names to
expressions in which you want to do replacements, or to rules that you want to use to do the
replacements.

This assigns a value to the symbol t.

In[13]:= t = 1 + x^2

Out[13]= 1 + x2

This finds the value of t, and then replaces x by 2 in it.

In[14]:= t . x -> 2
Out[14]= 5

This finds the value of t for a different value of x.

In[15]:= t . x -> 5 a

Out[15]= 1 + 25 a2

This finds the value of t when x is replaced by Pi , and then evaluates the result numerically.
In[16]:= t . x -> Pi N
Out[16]= 10.8696

Transforming Algebraic Expressions

There are often many different ways to write the same algebraic expression. As one example,
the expression H1 + xL2 can be written as 1 + 2 x + x2 . Mathematica provides a large collection of
functions for converting between different forms of algebraic expressions.
Mathematics and Algorithms 41

Expand@exprD multiply out products and powers, writing the result as a

sum of terms
Factor@exprD write expr as a product of minimal factors

Two common functions for transforming algebraic expressions.

Expand gives the "expanded form", with products and powers multiplied out.
In[1]:= Expand@H1 + xL ^ 2D

Out[1]= 1 + 2 x + x2

Factor recovers the original form.

In[2]:= Factor@%D

Out[2]= H1 + xL2

It is easy to generate complicated expressions with Expand.

In[3]:= Expand@H1 + x + 3 yL ^ 4D

Out[3]= 1 + 4 x + 6 x2 + 4 x3 + x4 + 12 y + 36 x y + 36 x2 y + 12 x3 y + 54 y2 + 108 x y2 + 54 x2 y2 + 108 y3 + 108 x y3 + 81 y4

Factor often gives you simpler expressions.

In[4]:= Factor@%D

Out[4]= H1 + x + 3 yL4

There are some cases, though, where Factor can give you more complicated expressions.
In[5]:= Factor@x ^ 10 - 1D

Out[5]= H-1 + xL H1 + xL I1 - x + x2 - x3 + x4 M I1 + x + x2 + x3 + x4 M

In this case, Expand gives the "simpler" form.

In[6]:= Expand@%D

Out[6]= -1 + x10

Simplifying Algebraic Expressions

There are many situations where you want to write a particular algebraic expression in the
simplest possible form. Although it is difficult to know exactly what one means in all cases by

the "simplest form", a worthwhile practical procedure is to look at many different forms of an
expression, and pick out the one that involves the smallest number of parts.
42 Mathematics and Algorithms

the "simplest form", a worthwhile practical procedure is to look at many different forms of an
expression, and pick out the one that involves the smallest number of parts.

Simplify@exprD try to find the simplest form of expr by applying various

standard algebraic transformations
FullSimplify@exprD try to find the simplest form by applying a wide range of
transformations

Simplifying algebraic expressions.

Simplify writes x2 + 2 x + 1 in factored form.

In[1]:= Simplify@x ^ 2 + 2 x + 1D

Out[1]= H1 + xL2

Simplify leaves x10 - 1 in expanded form, since for this expression, the factored form is
larger.
In[2]:= Simplify@x ^ 10 - 1D

Out[2]= -1 + x10

You can often use Simplify to "clean up" complicated expressions that you get as the results
of computations.

Here is the integral of 1 Ix4 - 1M. Integrals are discussed in more detail in "Integration".
In[3]:= Integrate@1 Hx ^ 4 - 1L, xD
ArcTan@xD 1 1
Out[3]= - + Log@-1 + xD - Log@1 + xD
2 4 4

Differentiating the result from Integrate should give back your original expression. In this
case, as is common, you get a more complicated version of the expression.
In[4]:= D@%, xD
1 1 1
Out[4]= - -
4 H-1 + xL 4 H1 + xL 2 I1 + x2 M

Simplify succeeds in getting back the original, simpler, form of the expression.
In[5]:= Simplify@%D
1
Out[5]=
-1 + x4

Simplify is set up to try various standard algebraic transformations on the expressions you
give. Sometimes, however, it can take more sophisticated transformations to make progress in
finding the simplest form of an expression.
Mathematics and Algorithms 43

FullSimplify tries a much wider range of transformations, involving not only algebraic func-
tions, but also many other kinds of functions.

Simplify does nothing to this expression.

In[6]:= Simplify@Gamma@xD Gamma@1 - xDD
Out[6]= Gamma@1 - xD Gamma@xD

FullSimplify , however, transforms it to a simpler form.

In[7]:= FullSimplify@Gamma@xD Gamma@1 - xDD
Out[7]= p Csc@p xD

For fairly small expressions, FullSimplify will often succeed in making some remarkable
simplifications. But for larger expressions, it can become unmanageably slow.

The reason for this is that to do its job, FullSimplify effectively has to try combining every
part of an expression with every other, and for large expressions the number of cases that it
has to consider can be astronomically large.

Simplify also has a difficult task to do, but it is set up to avoid some of the most time-consum-
ing transformations that are tried by FullSimplify. For simple algebraic calculations, there-
fore, you may often find it convenient to apply Simplify quite routinely to your results.

In more complicated calculations, however, even Simplify, let alone FullSimplify, may end
up needing to try a very large number of different forms, and therefore taking a long time. In
such cases, you typically need to do more controlled simplification, and use your knowledge of
the form you want to get to guide the process.

Putting Expressions into Different Forms

Complicated algebraic expressions can usually be written in many different ways. Mathematica
provides a variety of functions for converting expressions from one form to another.

In many applications, the most common of these functions are Expand, Factor and Simplify.
44 Mathematics and Algorithms

However, particularly when you have rational expressions that contain quotients, you may need
to use other functions.

Expand@exprD multiply out products and powers

ExpandAll@exprD apply Expand everywhere
Factor@exprD reduce to a product of factors
Together@exprD put all terms over a common denominator
Apart@exprD separate into terms with simple denominators
Cancel@exprD cancel common factors between numerators and
denominators
Simplify@exprD try a sequence of algebraic transformations and give the
smallest form of expr found

Functions for transforming algebraic expressions.

Here is a rational expression that can be written in many different forms.

In[1]:= e = Hx - 1L ^ 2 H2 + xL HH1 + xL Hx - 3L ^ 2L
H-1 + xL2 H2 + xL
Out[1]=
H-3 + xL2 H1 + xL

Expand expands out the numerator, but leaves the denominator in factored form.
In[2]:= Expand@eD
2 3x x3
Out[2]= - +
H-3 + xL2 H1 + xL H-3 + xL2 H1 + xL H-3 + xL2 H1 + xL

ExpandAll expands out everything, including the denominator.

In[3]:= ExpandAll@eD
2 3x x3
Out[3]= - +
9 + 3 x - 5 x2 + x3 9 + 3 x - 5 x2 + x3 9 + 3 x - 5 x2 + x3

Together collects all the terms together over a common denominator.

In[4]:= Together@%D
2 - 3 x + x3
Out[4]=
H-3 + xL2 H1 + xL
Mathematics and Algorithms 45

Apart breaks the expression apart into terms with simple denominators.
In[5]:= Apart@%D
5 19 1
Out[5]= 1 + + +
H-3 + xL2 4 H-3 + xL 4 H1 + xL

Factor factors everything, in this case reproducing the original form.

In[6]:= Factor@%D
H-1 + xL2 H2 + xL
Out[6]=
H-3 + xL2 H1 + xL

According to Simplify , this is the simplest way to write the original expression.
In[7]:= Simplify@eD
H-1 + xL2 H2 + xL
Out[7]=
H-3 + xL2 H1 + xL

Getting expressions into the form you want is something of an art. In most cases, it is best
simply to experiment, trying different transformations until you get what you want. Often you
will be able to use palettes in the front end to do this.

When you have an expression with a single variable, you can choose to write it as a sum of
terms, a product, and so on. If you have an expression with several variables, there is an even
wider selection of possible forms. You can, for example, choose to group terms in the expres-
sion so that one or another of the variables is "dominant".

Collect@expr,xD group together powers of x

FactorTerms @expr,xD pull out factors that do not depend on x

Rearranging expressions in several variables.

Here is an algebraic expression in two variables.

In[8]:= v = Expand@H3 + 2 xL ^ 2 Hx + 2 yL ^ 2D
2 3 4 2 3 2 2 2 2
Out[8]= 9 x + 12 x + 4 x + 36 x y + 48 x y + 16 x y + 36 y + 48 x y + 16 x y

This groups together terms in v that involve the same power of x.

In[9]:= Collect@v, xD
4 2 3 2 2 2
Out[9]= 4 x + 36 y + x H12 + 16 yL + x I9 + 48 y + 16 y M + x I36 y + 48 y M

This groups together powers of y.

46 Mathematics and Algorithms

This groups together powers of y.

In[10]:= Collect@v, yD
2 3 4 2 3 2 2
Out[10]= 9 x + 12 x + 4 x + I36 x + 48 x + 16 x M y + I36 + 48 x + 16 x M y

This factors out the piece that does not depend on y.

In[11]:= FactorTerms@v, yD
2 2 2
Out[11]= I9 + 12 x + 4 x M Ix + 4 x y + 4 y M

As we have seen, even when you restrict yourself to polynomials and rational expressions,
there are many different ways to write any particular expression. If you consider more compli-
cated expressions, involving, for example, higher mathematical functions, the variety of possi-
ble forms becomes still greater. As a result, it is totally infeasible to have a specific function
built into Mathematica to produce each possible form. Rather, Mathematica allows you to con-
struct arbitrary sets of transformation rules for converting between different forms. Many Mathe-
matica packages include such rules; the details of how to construct them for yourself are given
in "Transformation Rules and Definitions".

There are nevertheless a few additional built-in Mathematica functions for transforming
expressions.

TrigExpand@exprD expand out trigonometric expressions into a sum of terms

TrigFactor@exprD factor trigonometric expressions into products of terms
TrigReduce@exprD reduce trigonometric expressions using multiple angles
TrigToExp@exprD convert trigonometric functions to exponentials
ExpToTrig@exprD convert exponentials to trigonometric functions
FunctionExpand@exprD expand out special and other functions
ComplexExpand@exprD perform expansions assuming that all variables are real
PowerExpand @exprD transform Hx yL p into x p y p , etc.

Some other functions for transforming expressions.

This expands out the trigonometric expression, writing it so that all functions have argument x.
In[12]:= TrigExpand@Tan@xD Cos@2 xDD
3 Tan@xD 1
Out[12]= Cos@xD Sin@xD - - Sin@xD2 Tan@xD
2 2 2
Mathematics and Algorithms 47

This uses trigonometric identities to generate a factored form of the expression.

In[13]:= TrigFactor@%D
p p
Out[13]= 2 SinB - xF SinB + xF Tan@xD
4 4

This reduces the expression by using multiple angles.

In[14]:= TrigReduce@%D
1
Out[14]= - Sec@xD HSin@xD - Sin@3 xDL
2

This expands the sine assuming that x and y are both real.
In[15]:= ComplexExpand@Sin@x + I yDD
Out[15]= Cosh@yD Sin@xD + Cos@xD Sinh@yD

This does the expansion allowing x and y to be complex.

In[16]:= ComplexExpand@Sin@x + I yD, 8x, y<D
Out[16]= -Cosh@Im@xD + Re@yDD Sin@Im@yD - Re@xDD + Cos@Im@yD - Re@xDD Sinh@Im@xD + Re@yDD

The transformations on expressions done by functions like Expand and Factor are always
correct, whatever values the symbolic variables in the expressions may have. Sometimes,
however, it is useful to perform transformations that are only correct for some possible values
of symbolic variables. One such transformation is performed by PowerExpand .

Mathematica does not automatically expand out non-integer powers of products.

In[17]:= Sqrt@x yD

Out[17]= xy

PowerExpand does the expansion.

In[18]:= PowerExpand@%D

Out[18]= x y
48 Mathematics and Algorithms

Simplifying with Assumptions

Simplify@expr,assumD simplify expr with assumptions

Simplifying with assumptions.

Mathematica does not automatically simplify this, since it is only true for some values of x.
In[1]:= Simplify@Sqrt@x ^ 2DD

Out[1]= x2

x2 is equal to x for x 0, but not otherwise.

In[2]:= 8Sqrt@4 ^ 2D, Sqrt@H- 4L ^ 2D<
Out[2]= 84, 4<

This tells Simplify to make the assumption x > 0, so that simplification can proceed.
In[3]:= Simplify@Sqrt@x ^ 2D, x > 0D
Out[3]= x

No automatic simplification can be done on this expression.

In[4]:= 2 a + 2 Sqrt@a - Sqrt@- bDD Sqrt@a + Sqrt@- bDD

Out[4]= 2a+2 a- -b a+ -b

If a and b are assumed to be positive, the expression can however be simplified.

In[5]:= Simplify@%, a > 0 && b > 0D

Out[5]= 2 a+ a2 + b

Here is a simple example involving trigonometric functions.

In[6]:= Simplify@ArcSin@Sin@xDD, - Pi 2 < x < Pi 2D
Out[6]= x
Mathematics and Algorithms 49

Element@x,domD state that x is an element of the domain dom

Element@8x1 ,x2 ,<,domD state that all the xi are elements of the domain dom

Reals real numbers

Integers integers

Primes prime numbers

Some domains used in assumptions.

This simplifies x2 assuming that x is a real number.

In[7]:= Simplify@Sqrt@x ^ 2D, Element@x, RealsDD
Out[7]= Abs@xD

This simplifies the sine assuming that n is an integer.

In[8]:= Simplify@Sin@x + 2 n PiD, Element@n, IntegersDD
Out[8]= Sin@xD

With the assumptions given, Fermats little theorem can be used.

In[9]:= Simplify@Mod@a ^ p, pD, Element@a, IntegersD && Element@p, PrimesDD
Out[9]= Mod@a, pD

This uses the fact that sin HxL, but not arcsin HxL, is real when x is real.
In[10]:= Simplify@Re@8Sin@xD, ArcSin@xD<D, Element@x, RealsDD
Out[10]= 8Sin@xD, Re@ArcSin@xDD<

Picking Out Pieces of Algebraic Expressions

Coefficient @expr, formD coefficient of form in expr

Exponent@expr, formD maximum power of form in expr

Part@expr,nD or expr@@nDD nth term of expr

Functions to pick out pieces of polynomials.

50 Mathematics and Algorithms

Here is an algebraic expression.

In[1]:= e = Expand@H1 + 3 x + 4 y ^ 2L ^ 2D

Out[1]= 1 + 6 x + 9 x2 + 8 y2 + 24 x y2 + 16 y4

This gives the coefficient of x in e.

In[2]:= Coefficient@e, xD

Out[2]= 6 + 24 y2

Exponent@expr, yD gives the highest power of y that appears in expr.

In[3]:= Exponent@e, yD
Out[3]= 4

This gives the fourth term in e.

In[4]:= Part@e, 4D

Out[4]= 8 y2

You may notice that the function Part@expr, nD used to pick out the nth term in a sum is the
same as the function described in "Manipulating Elements of Lists" for picking out elements in
lists. This is no coincidence. In fact, as discussed in "Manipulating Expressions like Lists," every
Mathematica expression can be manipulated structurally much like a list. However, as discussed
in "Manipulating Expressions like Lists," you must be careful, because Mathematica often shows
algebraic expressions in a form that is different from the way it treats them internally.

Coefficient works even with polynomials that are not explicitly expanded out.
In[5]:= Coefficient@H1 + 3 x + 4 y ^ 2L ^ 2, xD

Out[5]= 6 + 24 y2

Numerator@exprD numerator of expr

Denominator @exprD denominator of expr

Functions to pick out pieces of rational expressions.

Here is a rational expression.

In[6]:= r = H1 + xL H2 H2 - yLL
1+x
Out[6]=
2 H2 - yL

Denominator picks out the denominator.

Mathematics and Algorithms 51

Denominator picks out the denominator.

In[7]:= Denominator@%D
Out[7]= 2 H2 - yL

Denominator gives 1 for expressions that are not explicit quotients.

In[8]:= Denominator@1 x + 2 yD
Out[8]= 1

Controlling the Display of Large Expressions

When you do symbolic calculations, it is quite easy to end up with extremely complicated expres -
sions. Often, you will not even want to see the complete result of a computation.

If you end your input with a semicolon, Mathematica will do the computation you asked for, but
will not display the result. You can nevertheless use % or Out@nD to refer to the result.

By default, the Mathematica front end will display any outputs which are excessively large in a
shortened form inside an interface which allows you to refine the display of the output.

Mathematica shows this output with 5138 of the terms omitted.

In[1]:= Expand@Hx + 2 y + 1L ^ 100D

A very large output was generated. Here is a sample of it:

1 + 100 x + 4950 x2 + 161 700 x3 + 3 921 225 x4 + 75 287 520 x5 + 1 192 052 400 x6 + 5138 +
1 568 717 617 782 433 884 352 170 216 652 800 y98 + 3 137 435 235 564 867 768 704 340 433 305 600 x y98 +
Out[1]= 1 568 717 617 782 433 884 352 170 216 652 800 x2 y98 + 63 382 530 011 411 470 074 835 160 268 800 y99 +
63 382 530 011 411 470 074 835 160 268 800 x y99 + 1 267 650 600 228 229 401 496 703 205 376 y100

Show Less Show More Show Full Output Set Size Limit...

The Show Less and Show More buttons allow you to decrease or increase the level of detail to which
Mathematica shows the expression. The Show Full Output button removes the interface entirely and
displays the full result, but the result may take considerable time to display. The default threshold size at
which this feature starts working may be set using the Set Size Limit option, which opens the Prefer-
ences dialog to the panel with the appropriate setting.

The large output suppression feature is implemented using the Mathematica function Short.
You can use Short directly for finer control over the display of expressions. You can also use it
for outputs which are not large enough to be suppressed by the default suppression scheme.

Ending your input with ; stops Mathematica from displaying the complicated result of the
computation.
52 Mathematics and Algorithms

Ending your input with ; stops Mathematica from displaying the complicated result of the
computation.
In[2]:= Expand@Hx + 5 y + 10L ^ 8D;

You can still refer to the result as %. Short displays a one-line outline of the result. The
<< n >> stands for n terms that have been left out.
In[3]:= % Short
2 7 7 8
Out[3]//Short= 100 000 000 + 80 000 000 x + 28 000 000 x + 39 + 6 250 000 y + 625 000 x y + 390 625 y

This shows a three-line version of the expression. More parts are now visible.
In[4]:= Short@%, 3D
2 3 4
Out[4]//Short= 100 000 000 + 80 000 000 x + 28 000 000 x + 5 600 000 x + 700 000 x +
56 000 x5 + 2800 x6 + 80 x7 + x8 + 28 + 5 250 000 x2 y5 + 175 000 x3 y5 +
43 750 000 y6 + 8 750 000 x y6 + 437 500 x2 y6 + 6 250 000 y7 + 625 000 x y7 + 390 625 y8

This gives the total number of terms in the sum.

In[5]:= Length@%D
Out[5]= 45

command; execute command , but do not print the result

exprShort show a one-line outline form of expr
Short@expr,nD show an n-line outline of expr

Some ways to shorten your output.

Using Symbols to Tag Objects

There are many ways to use symbols in Mathematica. Here we use symbols as "tags" for differ-
ent types of objects.

Working with physical units gives one simple example. When you specify the length of an
object, you want to give not only a number, but also the units in which the length is measured.
In standard notation, you might write a length as 12 meters.

You can imitate this notation almost directly in Mathematica. You can for example simply use a
symbol meters to indicate the units of your measurement.
Mathematics and Algorithms 53

The symbol meters here acts as a tag, which indicates the units used.
In[1]:= 12 meters
Out[1]= 12 meters

You can add lengths like this.

In[2]:= % + 5.3 meters
Out[2]= 17.3 meters

This gives a speed.

In[3]:= % H25 secondsL
0.692 meters
Out[3]=
seconds

This converts to a speed in feet per second.

In[4]:= % . meters -> 3.28084 feet
2.27034 feet
Out[4]=
seconds

There is in fact a Mathematica package that allows you to work with units. The package defines
many symbols that represent standard types of units.

Load the Mathematica package for handling units.

In[5]:= << Units`

The package uses standardized names for units.

In[6]:= 12 Meter Second
12 Meter
Out[6]=
Second
54 Mathematics and Algorithms

The function Convert@expr, unitsD converts to the specified units.

In[7]:= Convert@%, Mile HourD
37 500 Mile
Out[7]=
1397 Hour

Usually you have to give prefixes for units as separate words.

In[8]:= Convert@3 Kilo Meter Hour, Inch MinuteD
250 000 Inch
Out[8]=
127 Minute
Mathematics and Algorithms 55

Algebraic Manipulation

Structural Operations on Polynomials

Expand@polyD expand out products and powers

Factor@polyD factor completely
FactorTerms @polyD pull out any overall numerical factor
FactorTerms @poly,8x,y,<D pull out any overall factor that does not depend on x, y,
Collect@poly,xD arrange a polynomial as a sum of powers of x
Collect@poly,8x,y,<D arrange a polynomial as a sum of powers of x, y,

Structural operations on polynomials.

Here is a polynomial in one variable.

In[1]:= H2 + 4 x ^ 2L ^ 2 Hx - 1L ^ 3
2
Out[1]= H-1 + xL3 I2 + 4 x2 M

Expand expands out products and powers, writing the polynomial as a simple sum of terms.
In[2]:= t = Expand@%D

Out[2]= -4 + 12 x - 28 x2 + 52 x3 - 64 x4 + 64 x5 - 48 x6 + 16 x7

Factor performs complete factoring of the polynomial.

In[3]:= Factor@tD
2
Out[3]= 4 H-1 + xL3 I1 + 2 x2 M

FactorTerms pulls out the overall numerical factor from t.

In[4]:= FactorTerms@tD

Out[4]= 4 I-1 + 3 x - 7 x2 + 13 x3 - 16 x4 + 16 x5 - 12 x6 + 4 x7 M

There are several ways to write any polynomial. The functions Expand, FactorTerms and
Factor give three common ways. Expand writes a polynomial as a simple sum of terms, with all
products expanded out. FactorTerms pulls out common factors from each term. Factor does
complete factoring, writing the polynomial as a product of terms, each of as low degree as
possible.

When you have a polynomial in more than one variable, you can put the polynomial in different
forms by essentially choosing different variables to be "dominant". Collect@poly, xD takes a
polynomial in several variables and rewrites it as a sum of terms containing different powers of
the "dominant variable" x.

Here is a polynomial in two variables.

In[5]:= Expand@H1 + 2 x + yL ^ 3D

Out[5]= 1 + 6 x + 12 x2 + 8 x3 + 3 y + 12 x y + 12 x2 y + 3 y2 + 6 x y2 + y3

Collect reorganizes the polynomial so that x is the "dominant variable".

In[6]:= Collect@%, xD

Out[6]= 1 + 8 x3 + 3 y + 3 y2 + y3 + x2 H12 + 12 yL + x I6 + 12 y + 6 y2 M

If you specify a list of variables, Collect will effectively write the expression as a polynomial
in these variables.
In[7]:= Collect@Expand@H1 + x + 2 y + 3 zL ^ 3D, 8x, y<D

Out[7]= 1 + x3 + 8 y3 + 9 z + 27 z2 + 27 z3 + x2 H3 + 6 y + 9 zL +
y2 H12 + 36 zL + y I6 + 36 z + 54 z2 M + x I3 + 12 y2 + 18 z + 27 z2 + y H12 + 36 zLM

Expand@poly,pattD expand out poly avoiding those parts which do not contain
terms matching patt

Controlling polynomial expansion.

This avoids expanding parts which do not contain x.

In[8]:= Expand@Hx + 1L ^ 2 Hy + 1L ^ 2, xD

Out[8]= H1 + yL2 + 2 x H1 + yL2 + x2 H1 + yL2

This avoids expanding parts which do not contain objects matching b@_D.
In[9]:= Expand@Ha@1D + a@2D + 1L ^ 2 H1 + b@1DL ^ 2, b@_DD

Out[9]= H1 + a@1D + a@2DL2 + 2 H1 + a@1D + a@2DL2 b@1D + H1 + a@1D + a@2DL2 b@1D2

c
PowerExpand @exprD expand out Ha bLc and Iab M in expr

PowerExpand Aexpr,Assumptions ->assumE

expand out expr assuming assum

Expanding powers and logarithms.

Mathematica does not automatically expand out expressions of the form Ha bL ^ c except when c is
an integer. In general it is only correct to do this expansion if a and b are positive reals. Never-
Mathematics and Algorithms 57

Mathematica does not automatically expand out this expression.

In[10]:= Hx yL ^ n
n
Out[10]= Hx yL

PowerExpand does the expansion, effectively assuming that x and y are positive reals.
In[11]:= PowerExpand@%D
n n
Out[11]= x y

Log is not automatically expanded out.

In[12]:= Log@%D
n n
Out[12]= LogAx y E

PowerExpand does the expansion.

In[13]:= PowerExpand@%D
Out[13]= n Log@xD + n Log@yD

PowerExpand returns a result correct for the given assumptions.

In[14]:= PowerExpand@%%, Assumptions 8x > 0, n < 0<D
1 Im@n Log@xDD Im@n Log@yDD
Out[14]= 2 p FloorB - - F + n Log@xD + n Log@yD
2 2p 2p

Collect@poly,pattD collect separately terms involving each object that matches

patt
Collect@poly,patt,hD apply h to each final coefficient obtained

Ways of collecting terms.

Here is an expression involving various functions f.

In[15]:= t = 3 + x f@1D + x ^ 2 f@1D + y f@2D ^ 2 + z f@2D ^ 2

Out[15]= 3 + x f@1D + x2 f@1D + y f@2D2 + z f@2D2

58 Mathematics and Algorithms

This collects terms that match f@_D.

In[16]:= Collect@t, f@_DD

Out[16]= 3 + Ix + x2 M f@1D + Hy + zL f@2D2

This applies Factor to each coefficient obtained.

In[17]:= Collect@t, f@_D, FactorD

Out[17]= 3 + x H1 + xL f@1D + Hy + zL f@2D2

HornerForm@expr,xD puts expr into Horner form with respect to x

Horner form.

Horner form is a way of arranging a polynomial that allows numerical values to be computed
more efficiently by minimizing the number of multiplications.

This gives the Horner form of a polynomial.

In[18]:= HornerForm@1 + 2 x + 3 x ^ 2 + 4 x ^ 3, xD
Out[18]= 1 + x H2 + x H3 + 4 xLL

Finding the Structure of a Polynomial

PolynomialQ @expr,xD test whether expr is a polynomial in x

PolynomialQ @expr,8x1 ,x2 ,<D test whether expr is a polynomial in the xi

Variables@polyD a list of the variables in poly

Exponent@poly,xD the maximum exponent with which x appears in poly
Coefficient @poly,exprD the coefficient of expr in poly
Coefficient @poly,expr,nD the coefficient of exprn in poly
Coefficient @poly,expr,0D the term in poly independent of expr
CoefficientList@poly,8x1 ,x2 ,<D generate an array of the coefficients of the xi in poly

CoefficientRules@poly,8x1 ,x2 ,<D get exponent vectors and coefficients of monomials

Finding the structure of polynomials written in expanded form.

Here is a polynomial in two variables.

In[1]:= t = H1 + xL ^ 3 H1 - y - xL ^ 2
3 2
Out[1]= H1 + xL H1 - x - yL

This is the polynomial in expanded form.

Mathematics and Algorithms 59

This is the polynomial in expanded form.

In[2]:= Expand@tD

Out[2]= 1 + x - 2 x2 - 2 x3 + x4 + x5 - 2 y - 4 x y + 4 x3 y + 2 x4 y + y2 + 3 x y2 + 3 x2 y2 + x3 y2

PolynomialQ reports that t is a polynomial in x.

In[3]:= PolynomialQ@t, xD
Out[3]= True

This expression, however, is not a polynomial in x.

In[4]:= PolynomialQ@x + Sin@xD, xD
Out[4]= False

Variables gives a list of the variables in the polynomial t.

In[5]:= Variables@tD
Out[5]= 8x, y<

This gives the maximum exponent with which x appears in the polynomial t. For a polynomial
in one variable, Exponent gives the degree of the polynomial.
In[6]:= Exponent@t, xD
Out[6]= 5

Coefficient @poly, exprD gives the total coefficient with which expr appears in poly. In this
case, the result is a sum of two terms.
In[7]:= Coefficient@t, x ^ 2D

Out[7]= -2 + 3 y2

This is equivalent to Coefficient @t, x ^ 2D.

In[8]:= Coefficient@t, x, 2D

Out[8]= -2 + 3 y2

This picks out the coefficient of x0 in t.

In[9]:= Coefficient@t, x, 0D

Out[9]= 1 - 2 y + y2
60 Mathematics and Algorithms

CoefficientList gives a list of the coefficients of each power of x, starting with x0 .

In[10]:= CoefficientList@1 + 3 x ^ 2 + 4 x ^ 4, xD
Out[10]= 81, 0, 3, 0, 4<

For multivariate polynomials, CoefficientList gives an array of the coefficients for each
power of each variable.
In[11]:= CoefficientList@t, 8x, y<D
Out[11]= 881, -2, 1<, 81, -4, 3<, 8-2, 0, 3<, 8-2, 4, 1<, 81, 2, 0<, 81, 0, 0<<

CoefficientRules includes only those monomials that have nonzero coefficients.

In[12]:= CoefficientRules@t, 8x, y<D
Out[12]= 885, 0< 1, 84, 1< 2, 84, 0< 1, 83, 2< 1, 83, 1< 4, 83, 0< -2, 82, 2< 3,
82, 0< -2, 81, 2< 3, 81, 1< -4, 81, 0< 1, 80, 2< 1, 80, 1< -2, 80, 0< 1<

It is important to notice that the functions in this tutorial will often work even on polynomials
that are not explicitly given in expanded form.

Many of the functions also work on expressions that are not strictly polynomials.

Without giving specific integer values to a, b and c, this expression cannot strictly be consid-
ered a polynomial.
In[13]:= x^a + x^b + y^c

Out[13]= xa + xb + yc

Exponent@expr, xD still gives the maximum exponent of x in expr, but here has to write the
result in symbolic form.
In[14]:= Exponent@%, xD
Out[14]= Max@0, a, bD

Structural Operations on Rational Expressions

For ordinary polynomials, Factor and Expand give the most important forms. For rational
expressions, there are many different forms that can be useful.
Mathematics and Algorithms 61

ExpandNumerator@exprD expand numerators only

ExpandDenominator@exprD expand denominators only
Expand@exprD expand numerators, dividing the denominator into each
term
ExpandAll@exprD expand numerators and denominators completely

Different kinds of expansion for rational expressions.

Here is a rational expression.

In[1]:= t = H1 + xL ^ 2 H1 - xL + 3 x ^ 2 H1 + xL ^ 2 + H2 - xL ^ 2
3 x2 H1 + xL2
Out[1]= H2 - xL2 + +
H1 + xL2 1-x

ExpandNumerator writes the numerator of each term in expanded form.

In[2]:= ExpandNumerator@tD
3 x2 1 + 2 x + x2
Out[2]= 4 - 4 x + x2 + +
H1 + xL2 1-x

Expand expands the numerator of each term, and divides all the terms by the appropriate
denominators.
In[3]:= Expand@tD
1 2x x2 3 x2
Out[3]= 4+ -4x+ + x2 + +
1-x 1-x 1-x H1 + xL2

ExpandDenominator expands out the denominator of each term.

In[4]:= ExpandDenominator@tD
H1 + xL2 3 x2
Out[4]= H2 - xL2 + +
1-x 1 + 2 x + x2

ExpandAll does all possible expansions in the numerator and denominator of each term.
In[5]:= ExpandAll@tD
1 2x x2 3 x2
Out[5]= 4+ -4x+ + x2 + +
1-x 1-x 1-x 1 + 2 x + x2

ExpandAllAexpr,pattE , etc. avoid expanding parts which contain no terms matching patt

Controlling expansion.

This avoids expanding the term which does not contain z.

62 Mathematics and Algorithms

This avoids expanding the term which does not contain z.

In[6]:= ExpandAll@Hx + 1L ^ 2 y ^ 2 + Hz + 1L ^ 2 z ^ 2, zD
H1 + xL2 1 2
Out[6]= 1+ + +
2 2 z
y z

Together@exprD combine all terms over a common denominator

Apart@exprD write an expression as a sum of terms with simple
denominators
Cancel@exprD cancel common factors between numerators and
denominators
Factor@exprD perform a complete factoring

Structural operations on rational expressions.

Here is a rational expression.

In[7]:= u = H- 4 x + x ^ 2L H- x + x ^ 2L + H- 4 + 3 x + x ^ 2L H- 1 + x ^ 2L
-4 x + x2 -4 + 3 x + x2
Out[7]= +
-x + x2 -1 + x2

Together puts all terms over a common denominator.

In[8]:= Together@uD
2 I-4 + x2 M
Out[8]=
H-1 + xL H1 + xL

You can use Factor to factor the numerator and denominator of the resulting expression.
In[9]:= Factor@%D
2 H-2 + xL H2 + xL
Out[9]=
H-1 + xL H1 + xL

Apart writes the expression as a sum of terms, with each term having as simple a denomina-
tor as possible.
In[10]:= Apart@uD
3 3
Out[10]= 2- +
-1 + x 1+x
Mathematics and Algorithms 63

Cancel cancels any common factors between numerators and denominators.

In[11]:= Cancel@uD
-4 + x 4+x
Out[11]= +
-1 + x 1+x

Factor first puts all terms over a common denominator, then factors the result.
In[12]:= Factor@%D
2 H-2 + xL H2 + xL
Out[12]=
H-1 + xL H1 + xL

In mathematical terms, Apart decomposes a rational expression into "partial fractions".

In expressions with several variables, you can use Apart@expr, varD to do partial fraction decom-
positions with respect to different variables.

Here is a rational expression in two variables.

In[13]:= v = Hx ^ 2 + y ^ 2L Hx + x yL
x2 + y2
Out[13]=
x+xy

This gives the partial fraction decomposition with respect to x.

In[14]:= Apart@v, xD
x y2
Out[14]= +
1+y x H1 + yL

Here is the partial fraction decomposition with respect to y.

In[15]:= Apart@v, yD
1 y 1 + x2
Out[15]= - + +
x x x H1 + yL

Algebraic Operations on Polynomials

For many kinds of practical calculations, the only operations you will need to perform on polyno-
mials are essentially the structural ones already discussed.

If you do more advanced algebra with polynomials, however, you will have to use the algebraic
operations discussed in this tutorial.

You should realize that most of the operations discussed in this tutorial work only on ordinary
polynomials, with integer exponents and rational-number coefficients for each term.
64 Mathematics and Algorithms

You should realize that most of the operations discussed in this tutorial work only on ordinary
polynomials, with integer exponents and rational-number coefficients for each term.

PolynomialQuotient@poly1 ,poly2 ,xD find the result of dividing the polynomial poly1 in x by poly2 ,
dropping any remainder term
PolynomialRemainder@ find the remainder from dividing the polynomial poly1 in x
poly1 ,poly2 ,xD by poly2

PolynomialQuotientRemainder@poly1 ,poly2 ,xD

give the quotient and remainder in a list
PolynomialMod@poly,mD reduce the polynomial poly modulo m
PolynomialGCD@poly1 ,poly2 D find the greatest common divisor of two polynomials
PolynomialLCM@poly1 ,poly2 D find the least common multiple of two polynomials
PolynomialExtendedGCD@ find the extended greatest common divisor of two
poly1 ,poly2 D polynomials
Resultant@poly1 ,poly2 ,xD find the resultant of two polynomials
Subresultants@poly1 ,poly2 ,xD find the principal subresultant coefficients of two
polynomials
Discriminant@poly,xD find the discriminant of the polynomial poly
GroebnerBasis@8poly1 ,poly2 ,<,8x1 ,x2 ,<D
find the Grbner basis for the polynomials polyi

GroebnerBasis@8poly1 ,poly2 ,<,8x1 ,x2 ,<,8y1 ,y2 ,<D

find the Grbner basis eliminating the yi

PolynomialReduce@poly,8poly1 ,poly2 ,<,8x1 ,x2 ,<D

find a minimal representation of poly in terms of the polyi

Reduction of polynomials.

pHxL bHxL
Given two polynomials pHxL and qHxL, one can always uniquely write qHxL
= aHxL + qHxL
, where the

degree of bHxL is less than the degree of qHxL. PolynomialQuotient gives the quotient aHxL, and
PolynomialRemainder gives the remainder bHxL.

This gives the remainder from dividing x2 by 1 + x.

In[1]:= PolynomialRemainder@x ^ 2, x + 1, xD
Out[1]= 1
Mathematics and Algorithms 65

Here is the quotient of x2 and x + 1, with the remainder dropped.

In[2]:= PolynomialQuotient@x ^ 2, x + 1, xD
Out[2]= -1 + x

This gives back the original expression.

In[3]:= Simplify@Hx + 1L % + %%D

Out[3]= x2

Here the result depends on whether the polynomials are considered to be in x or y.

In[4]:= 8PolynomialRemainder@x + y, x - y, xD, PolynomialRemainder@x + y, x - y, yD<
Out[4]= 82 y, 2 x<

PolynomialMod is essentially the analog for polynomials of the function Mod for integers. When
the modulus m is an integer, PolynomialMod@poly, mD simply reduces each coefficient in poly
modulo the integer m. If m is a polynomial, then PolynomialMod@poly, mD effectively tries to get
a polynomial with as low a degree as possible by subtracting from poly appropriate multiples q m
of m. The multiplier q can itself be a polynomial, but its degree is always less than the degree of
poly. PolynomialMod yields a final polynomial whose degree and leading coefficient are both as
small as possible.

This reduces x2 modulo x + 1. The result is simply the remainder from dividing the polynomials.
In[5]:= PolynomialMod@x ^ 2, x + 1D
Out[5]= 1

In this case, PolynomialMod and PolynomialRemainder do not give the same result.
In[6]:= 8PolynomialMod@x ^ 2, a x + 1D, PolynomialRemainder@x ^ 2, a x + 1, xD<
1
Out[6]= :x2 , >
a2

The main difference between PolynomialMod and PolynomialRemainder is that while the
former works simply by multiplying and subtracting polynomials, the latter uses division in
getting its results. In addition, PolynomialMod allows reduction by several moduli at the same
time. A typical case is reduction modulo both a polynomial and an integer.
66 Mathematics and Algorithms

This reduces the polynomial x2 + 1 modulo both x + 1 and 2.

In[7]:= PolynomialMod@x ^ 2 + 1, 8x + 1, 2<D
Out[7]= 0

PolynomialGCD@poly1 , poly2 D finds the highest degree polynomial that divides the polyi exactly.

It gives the analog for polynomials of the integer function GCD.

PolynomialGCD gives the greatest common divisor of the two polynomials.

In[8]:= PolynomialGCD@H1 - xL ^ 2 H1 + xL H2 + xL, H1 - xL H2 + xL H3 + xLD
Out[8]= H-1 + xL H2 + xL

PolynomialExtendedGCD gives the extended greatest common divisor of the two polynomi-
als.
In[9]:= 8g, 8r, s<< = PolynomialExtendedGCD@x ^ 3 + 2 x ^ 2 - x + 1, x ^ 4 + x + 2, xD
29 26 x 23 x2 21 x3 93 19 x 21 x2
Out[9]= :1, : - - + , - - >>
215 215 215 215 215 215 215

The returned polynomials r and s can be used to represent the GCD in terms of the original
polynomials.
In[10]:= r Hx ^ 3 + 2 x ^ 2 - x + 1L + s Hx ^ 4 + x + 2L Expand
Out[10]= 1

The function Resultant@poly1 , poly2 , xD is used in a number of classical algebraic algorithms.

The resultant of two polynomials a and b, both with leading coefficient one, is given by the
product of all the differences ai - b j between the roots of the polynomials. It turns out that for

any pair of polynomials, the resultant is always a polynomial in their coefficients. By looking at
when the resultant is zero, you can tell for what values of their parameters two polynomials
have a common root. Two polynomials with leading coefficient one have k common roots if
exactly the first k elements in the list Subresultants@poly1 , poly2 , xD are zero.

Here is the resultant with respect to y of two polynomials in x and y. The original polynomials
have a common root in y only for values of x at which the resultant vanishes.
In[11]:= Resultant@Hx - yL ^ 2 - 2, y ^ 2 - 3, yD

Out[11]= 1 - 10 x2 + x4

The function Discriminant@poly, xD is the product of the squares of the differences of its roots.
It can be used to determine whether the polynomial has any repeated roots. The discriminant is
equal to the resultant of the polynomial and its derivative, up to a factor independent of the
variable.
Mathematics and Algorithms 67

This polynomial has a repeated root, so its discriminant vanishes.

In[12]:= Discriminant@Hx - 1L ^ 2, xD
Out[12]= 0

This polynomial has distinct roots, so its discriminant is nonzero.

In[13]:= Discriminant@x ^ 4 - 1, xD
Out[13]= -256

Grbner bases appear in many modern algebraic algorithms and applications. The function
GroebnerBasis@8poly1 , poly2 , <, 8x1 , x2 , <D takes a set of polynomials, and reduces this set
to a canonical form from which many properties can conveniently be deduced. An important
feature is that the set of polynomials obtained from GroebnerBasis always has exactly the
same collection of common roots as the original set.

The Hx + yL2 is effectively redundant, and so does not appear in the Grbner basis.
In[14]:= GroebnerBasis@8Hx + yL, Hx + yL ^ 2<, 8x, y<D
Out[14]= 8x + y<

The polynomial 1 has no roots, showing that the original polynomials have no common roots.
In[15]:= GroebnerBasis@8x + y, x ^ 2 - 1, y ^ 2 - 2 x<, 8x, y<D
Out[15]= 81<

The polynomials are effectively unwound here, and can now be seen to have exactly five com-
mon roots.
In[16]:= GroebnerBasis@8x y ^ 2 + 2 x y + x ^ 2 + 1, x y + y ^ 2 + 1<, 8x, y<D

Out[16]= 9-1 - y2 + y3 + y4 + y5 , x + y2 + y3 + y4 =

PolynomialReduce@poly, 8p1 , p2 , <, 8x1 , x2 , <D yields a list 88a1 , a2 , <, b< of polynomials
with the property that b is minimal and a1 p1 + a2 p2 + + b is exactly poly.
68 Mathematics and Algorithms

This writes x2 + y2 in terms of x - y and y + a, leaving a remainder that depends only on a.

In[17]:= PolynomialReduce@x ^ 2 + y ^ 2, 8x - y, y + a<, 8x, y<D

Out[17]= 98x + y, -2 a + 2 y<, 2 a2 =

Factor@polyD factor a polynomial

FactorSquareFree@polyD write a polynomial as a product of powers of square-free
factors
FactorTerms @poly,xD factor out terms that do not depend on x
FactorList@polyD , FactorSquareFreeList@polyD , FactorTermsList@polyD
give results as lists of factors

Functions for factoring polynomials.

Factor, FactorTerms and FactorSquareFree perform various degrees of factoring on polynomi-

als. Factor does full factoring over the integers. FactorTerms extracts the "content" of the
polynomial. FactorSquareFree pulls out any multiple factors that appear.

Here is a polynomial, in expanded form.

In[18]:= t = Expand@2 H1 + xL ^ 2 H2 + xL H3 + xLD

Out[18]= 12 + 34 x + 34 x2 + 14 x3 + 2 x4

FactorTerms pulls out only the factor of 2 that does not depend on x.
In[19]:= FactorTerms@t, xD

Out[19]= 2 I6 + 17 x + 17 x2 + 7 x3 + x4 M

FactorSquareFree factors out the 2 and the term H1 + xL ^ 2, but leaves the rest unfactored.
In[20]:= FactorSquareFree@tD

Out[20]= 2 H1 + xL2 I6 + 5 x + x2 M

Factor does full factoring, recovering the original form.

In[21]:= Factor@tD

Out[21]= 2 H1 + xL2 H2 + xL H3 + xL

Particularly when you write programs that work with polynomials, you will often find it conve-
nient to pick out pieces of polynomials in a standard form. The function FactorList gives a list
of all the factors of a polynomial, together with their exponents. The first element of the list is
always the overall numerical factor for the polynomial.

The form that FactorList returns is the analog for polynomials of the form produced by
FactorInteger for integers.
Mathematics and Algorithms 69

The form that FactorList returns is the analog for polynomials of the form produced by
FactorInteger for integers.

Here is a list of the factors of the polynomial in the previous set of examples. Each element of
the list gives the factor, together with its exponent.
In[22]:= FactorList@tD
Out[22]= 882, 1<, 81 + x, 2<, 82 + x, 1<, 83 + x, 1<<

FactorApoly,GaussianIntegers->TrueE
factor a polynomial, allowing coefficients that are Gaussian
integers

Factoring polynomials with complex coefficients.

Factor and related functions usually handle only polynomials with ordinary integer or rational-
number coefficients. If you set the option GaussianIntegers -> True, however, then Factor
will allow polynomials with coefficients that are complex numbers with rational real and imagi-
nary parts. This often allows more extensive factorization to be performed.

This polynomial is irreducible when only ordinary integers are allowed.

In[23]:= Factor@1 + x ^ 2D

Out[23]= 1 + x2

When Gaussian integer coefficients are allowed, the polynomial factors.

In[24]:= Factor@1 + x ^ 2, GaussianIntegers -> TrueD
Out[24]= H- + xL H + xL

IrreduciblePolynomialQ@polyD test whether poly is an irreducible polynomial over the

rationals
IrreduciblePolynomialQApoly, test whether poly is irreducible over the Gaussian rationals
GaussianIntegersTrueE
IrreduciblePolynomialQA test irreducibility over the rationals extended by the
poly,ExtensionAutomaticE algebraic number coefficients of poly

Irreducibility testing.
70 Mathematics and Algorithms

A polynomial is irreducible over a field F if it cannot be represented as a product of two noncon-

stant polynomials with coefficients in F.

This polynomial is irreducible over the rationals.

In[25]:= IrreduciblePolynomialQ@x ^ 2 + 4D
Out[25]= True

Over the Gaussian rationals, the polynomial is reducible.

In[26]:= IrreduciblePolynomialQ@x ^ 2 + 4, GaussianIntegers -> TrueD
Out[26]= False

By default, algebraic numbers are treated as independent variables.

In[27]:= IrreduciblePolynomialQ@x ^ 2 + 2 Sqrt@2D x + 2D
Out[27]= True

Over the rationals extended by Sqrt@2D, the polynomial is reducible.

In[28]:= IrreduciblePolynomialQ@x ^ 2 + 2 Sqrt@2D x + 2, Extension AutomaticD
Out[28]= False

Cyclotomic@n,xD give the cyclotomic polynomial of order n in x

Cyclotomic polynomials.

Cyclotomic polynomials arise as "elementary polynomials" in various algebraic algorithms. The

cyclotomic polynomials are defined by Cn HxL = k Ix - e2 p i kn M, where k runs over all positive inte-

gers less than n that are relatively prime to n.

This is the cyclotomic polynomial C6 HxL.

In[29]:= Cyclotomic@6, xD

Out[29]= 1 - x + x2

C6 HxL appears in the factors of x6 - 1.

In[30]:= Factor@x ^ 6 - 1D

Out[30]= H-1 + xL H1 + xL I1 - x + x2 M I1 + x + x2 M
Mathematics and Algorithms 71

Decompose@poly,xD decompose poly, if possible, into a composition of a list of

simpler polynomials

Decomposing polynomials.

Factorization is one important way of breaking down polynomials into simpler parts. Another,
quite different, way is decomposition. When you factor a polynomial PHxL, you write it as a
product p1 HxL p2 HxL ... of polynomials pi HxL. Decomposing a polynomial QHxL consists of writing it as
a composition of polynomials of the form q1 Hq2 H ... HxL ...LL.

Here is a simple example of Decompose. The original polynomial x4 + x2 + 1 can be written as

2
the polynomial x + x + 1, where x is the polynomial x2 .
In[31]:= Decompose@x ^ 4 + x ^ 2 + 1, xD

Out[31]= 91 + x + x2 , x2 =

Here are two polynomial functions.

In[32]:= Hq1@x_D = 1 - 2 x + x ^ 4; q2@x_D = 5 x + x ^ 3;L

This gives the composition of the two functions.

In[33]:= Expand@q1@q2@xDDD

Out[33]= 1 - 10 x - 2 x3 + 625 x4 + 500 x6 + 150 x8 + 20 x10 + x12

Decompose recovers the original functions.

In[34]:= Decompose@%, xD

Out[34]= 91 - 2 x + x4 , 5 x + x3 =

Decompose@poly, xD is set up to give a list of polynomials in x, which, if composed, reproduce

the original polynomial. The original polynomial can contain variables other than x, but the
sequence of polynomials that Decompose produces are all intended to be considered as func-
tions of x.

Unlike factoring, the decomposition of polynomials is not completely unique. For example, the
two sets of polynomials pi and qi , related by q1 HxL = p1 Hx - aL and q2 HxL = p2 HxL + a give the same
result on composition, so that p1 Hp2 HxLL = q1 Hq2 HxLL. Mathematica follows the convention of absorb-
ing any constant terms into the first polynomial in the list produced by Decompose.
72 Mathematics and Algorithms

InterpolatingPolynomial@8 f1 , f2 ,<,xD
give a polynomial in x which is equal to fi when x is the
integer i
InterpolatingPolynomial@88x1 , f1 <,8x2 , f2 <,<,xD
give a polynomial in x which is equal to fi when x is xi

Generating interpolating polynomials.

This yields a quadratic polynomial which goes through the specified three points.
In[35]:= InterpolatingPolynomial@88- 1, 4<, 80, 2<, 81, 6<<, xD
Out[35]= 4 + H1 + xL H-2 + 3 xL

When x is 0, the polynomial has value 2.

In[36]:= % . x -> 0
Out[36]= 2

Polynomials Modulo Primes

Mathematica can work with polynomials whose coefficients are in the finite field Z p of integers

modulo a prime p.

PolynomialMod@poly,pD reduce the coefficients in a polynomial modulo p

ExpandApoly,Modulus->pE expand poly modulo p

FactorApoly,Modulus->pE factor poly modulo p

PolynomialGCDApoly1 ,poly2 ,Modulus->pE

find the GCD of the polyi modulo p

GroebnerBasisApolys,vars,Modulus->pE
find the Grbner basis modulo p

Functions for manipulating polynomials over finite fields.

Here is an ordinary polynomial.

In[1]:= Expand@H1 + xL ^ 6D

Out[1]= 1 + 6 x + 15 x2 + 20 x3 + 15 x4 + 6 x5 + x6
Mathematics and Algorithms 73

This reduces the coefficients modulo 2.

In[2]:= PolynomialMod@%, 2D

Out[2]= 1 + x2 + x4 + x6

Here are the factors of the resulting polynomial over the integers.
In[3]:= Factor@%D

Out[3]= I1 + x2 M I1 + x4 M

If you work modulo 2, further factoring becomes possible.

In[4]:= Factor@%, Modulus -> 2D

Out[4]= H1 + xL6

Symmetric Polynomials
A symmetric polynomial in variables x1 , , xn is a polynomial that is invariant under arbitrary
permutations of x1 , , xn . Polynomials

s1 = x1 + x2 + + xn
s2 = x1 x2 + x1 x3 + + xn-1 xn

sn = x1 x2 xn

are called elementary symmetric polynomials in variables x1 , , xn .

The fundamental theorem of symmetric polynomials says that every symmetric polynomial in
x1 , , xn can be represented as a polynomial in elementary symmetric polynomials in x1 , , xn .

When the ordering of variables is fixed, an arbitrary polynomial f can be uniquely represented
as a sum of a symmetric polynomial p, called the symmetric part of f , and a remainder q that
e e
does not contain descending monomials. A monomial c x11 xnn is called descending iff
e1 en .
74 Mathematics and Algorithms

SymmetricPolynomial@ give the kth elementary symmetric polynomial in the

k,8x1 ,,xn <D variables x1 , , xn

SymmetricReduction@ f ,8x1 ,,xn <D give a pair of polynomials 8p, q< in x1 , , xn such that
f == p + q, where p is the symmetric part and q is the
remainder
SymmetricReduction@ f ,8x1 ,,xn <,8s1 ,,sn <D
give the pair 8p, q< with the elementary symmetric polynomi -
als in p replaced by s1 , , sn

Functions for symmetric polynomial computations.

Here is the elementary symmetric polynomial of degree three in four variables.

In[1]:= SymmetricPolynomial@3, 8x, y, z, t<D
Out[1]= t x y + t x z + t y z + x y z

This writes the polynomial Hx + yL2 + Hx + zL2 + Hz + yL2 in terms of elementary symmetric polynomi-
als. The input polynomial is symmetric, so the remainder is zero.
In[2]:= SymmetricReductionAHx + yL2 + Hx + zL2 + Hz + yL2 , 8x, y, z<E
2
Out[2]= 92 Hx + y + zL - 2 Hx y + x z + y zL, 0=

Here the elementary symmetric polynomials in the symmetric part are replaced with variables
s1 , s2 , s3 . The polynomial is not symmetric, so the remainder is not zero.
In[3]:= SymmetricReductionAx5 + y5 + z4 , 8x, y, z<, 8s1 , s2 , s3 <E
5 3 2 2 4 5
Out[3]= :s1 - 5 s1 s2 + 5 s1 s2 + 5 s1 s3 - 5 s2 s3 , z - z >

SymmetricReduction can be applied to polynomials with symbolic coefficients.

In[4]:= SymmetricReductionAx5 + y5 + z4 + a x4 + b y4 + c z5 , 8x, y, z<, 8s1 , s2 , s3 <E
4 5 2 3 2 2 2 4 4 5
Out[4]= :a s1 + s1 - 4 a s1 s2 - 5 s1 s2 + 2 a s2 + 5 s1 s2 + 4 a s1 s3 + 5 s1 s3 - 5 s2 s3 , H-a + bL y + H1 - aL z + H-1 + cL z >

Polynomials over Algebraic Number Fields

Functions like Factor usually assume that all coefficients in the polynomials they produce must
involve only rational numbers. But by setting the option Extension you can extend the domain
of coefficients that will be allowed.
Mathematics and Algorithms 75

FactorApoly,Extension->9a1 ,a2 ,=E

factor poly allowing coefficients that are rational combina-
tions of the ai

Factoring polynomials over algebraic number fields.

Allowing only rational number coefficients, this polynomial cannot be factored.

In[1]:= Factor@1 + x ^ 4D

Out[1]= 1 + x4

With coefficients that can involve 2 , the polynomial can now be factored.
In[2]:= Factor@1 + x ^ 4, Extension -> 8Sqrt@2D<D

Out[2]= - -1 + 2 x - x2 1+ 2 x + x2

The polynomial can also be factored if one allows coefficients involving -1 .

In[3]:= Factor@1 + x ^ 4, Extension -> 8Sqrt@- 1D<D

Out[3]= I- + x2 M I + x2 M

GaussianIntegers -> True is equivalent to Extension -> Sqrt@- 1D.

In[4]:= Factor@1 + x ^ 4, GaussianIntegers -> TrueD

Out[4]= I- + x2 M I + x2 M

If one allows coefficients that involve both 2 and -1 the polynomial can be factored com-
pletely.
In[5]:= Factor@1 + x ^ 4, Extension -> 8Sqrt@2D, Sqrt@- 1D<D
1
Out[5]= 2 - H1 + L x 2 - H1 - L x 2 + H1 - L x 2 + H1 + L x
4

Expand gives the original polynomial back again.

In[6]:= Expand@%D

Out[6]= 1 + x4
76 Mathematics and Algorithms

FactorApoly,Extension->AutomaticE
factor poly allowing algebraic numbers in poly to appear in
coefficients

Factoring polynomials with algebraic number coefficients.

Here is a polynomial with a coefficient involving 2.

In[7]:= t = Expand@HSqrt@2D + xL ^ 2D

Out[7]= 2+2 2 x + x2

By default, Factor will not factor this polynomial.

In[8]:= Factor@tD

Out[8]= 2+2 2 x + x2

But now the field of coefficients is extended by including 2 , and the polynomial is factored.
In[9]:= Factor@t, Extension -> AutomaticD
2
Out[9]= 2 +x

Other polynomial functions work much like Factor. By default, they treat algebraic number
coefficients just like independent symbolic variables. But with the option
Extension -> Automatic they perform operations on these coefficients.

By default, Cancel does not reduce these polynomials.

In[10]:= Cancel@t Hx ^ 2 - 2LD

2+2 2 x + x2
Out[10]=
-2 + x2

But now it does.

In[11]:= Cancel@t Hx ^ 2 - 2L, Extension -> AutomaticD

- 2 -x
Out[11]=
2 -x

By default, PolynomialLCM pulls out no common factors.

In[12]:= PolynomialLCM@t, x ^ 2 - 2D

Out[12]= I-2 + x2 M 2 + 2 2 x + x2

But now it does.

Mathematics and Algorithms 77

But now it does.

In[13]:= PolynomialLCM@t, x ^ 2 - 2, Extension -> AutomaticD

Out[13]= -2 2 -2x+ 2 x2 + x3

IrreduciblePolynomialQA test whether poly is an irreducible polynomial over the

poly,ExtensionAutomaticE rationals extended by the coefficients of poly

IrreduciblePolynomialQA test whether poly is irreducible over the rationals extended

poly,Extension->8a1 ,a2 ,<E by the coefficients of poly and by a1 , a2 ,

IrreduciblePolynomialQA test irreducibility over the field of all complex numbers

poly,ExtensionAllE

Irreducibility testing.

A polynomial is irreducible over a field F if it cannot be represented as a product of two noncon-

stant polynomials with coefficients in F.

By default, algebraic numbers are treated as independent variables.

In[14]:= IrreduciblePolynomialQ@x ^ 2 + 2 Sqrt@2D x + 2D
Out[14]= True

Over the rationals extended by Sqrt@2D, the polynomial is reducible.

In[15]:= IrreduciblePolynomialQ@x ^ 2 + 2 Sqrt@2D x + 2, Extension AutomaticD
Out[15]= False

This polynomial is irreducible over the rationals.

In[16]:= IrreduciblePolynomialQ@x ^ 2 - 3D
Out[16]= True

Over the rationals extended by Sqrt@3D, the polynomial is reducible.

In[17]:= IrreduciblePolynomialQ@x ^ 2 - 3, Extension 8Sqrt@3D<D
Out[17]= False

This polynomial is irreducible over the field of all complex numbers.

In[18]:= IrreduciblePolynomialQ@x ^ 3 - 5 x y + 7, Extension AllD
Out[18]= True

Trigonometric Expressions
78 Mathematics and Algorithms

Trigonometric Expressions

TrigExpand@exprD expand trigonometric expressions out into a sum of terms

TrigFactor@exprD factor trigonometric expressions into products of terms
TrigFactorList@exprD give terms and their exponents in a list
TrigReduce@exprD reduce trigonometric expressions using multiple angles

Functions for manipulating trigonometric expressions.

This expands out a trigonometric expression.

In[1]:= TrigExpand@Sin@2 xD Cos@2 yDD
2 2
Out[1]= 2 Cos@xD Cos@yD Sin@xD - 2 Cos@xD Sin@xD Sin@yD

This factors the expression.

In[2]:= TrigFactor@%D
p p
Out[2]= 4 Cos@xD Sin@xD SinB - yF SinB + yF
4 4

And this reduces the expression to a form that is linear in the trigonometric functions.
In[3]:= TrigReduce@%D
1
Out[3]= HSin@2 x - 2 yD + Sin@2 x + 2 yDL
2

TrigExpand works on hyperbolic as well as circular functions.

In[4]:= TrigExpand@Tanh@x + yDD
Cosh@yD Sinh@xD Cosh@xD Sinh@yD
Out[4]= +
Cosh@xD Cosh@yD + Sinh@xD Sinh@yD Cosh@xD Cosh@yD + Sinh@xD Sinh@yD

TrigReduce reproduces the original form again.

In[5]:= TrigReduce@%D
Out[5]= Tanh@x + yD

Mathematica automatically uses functions like Tan whenever it can.

In[6]:= Sin@xD ^ 2 Cos@xD
Out[6]= Sin@xD Tan@xD

With TrigFactorList, however, you can see the parts of functions like Tan.
Mathematics and Algorithms 79

With TrigFactorList, however, you can see the parts of functions like Tan.
In[7]:= TrigFactorList@%D
Out[7]= 881, 1<, 8Sin@xD, 2<, 8Cos@xD, -1<<

TrigToExp@exprD write trigonometric functions in terms of exponentials

ExpToTrig@exprD write exponentials in terms of trigonometric functions

Converting to and from exponentials.

TrigToExp writes trigonometric functions in terms of exponentials.

In[8]:= TrigToExp@Tan@xDD
I- x - x M
Out[8]=
- x + x

TrigToExp also works with hyperbolic functions.

In[9]:= TrigToExp@Tanh@xDD
--x + x
Out[9]=
-x + x

ExpToTrig does the reverse, getting rid of explicit complex numbers whenever possible.
In[10]:= ExpToTrig@%D
Out[10]= Tanh@xD

ExpToTrig deals with hyperbolic as well as circular functions.

In[11]:= ExpToTrig@Exp@xD - Exp@- xDD
Out[11]= 2 Sinh@xD

You can also use ExpToTrig on purely numerical expressions.

In[12]:= ExpToTrig@H- 1L ^ H1 17LD
p p
Out[12]= CosB F + SinB F
17 17
80 Mathematics and Algorithms

Expressions Involving Complex Variables

Mathematica usually pays no attention to whether variables like x stand for real or complex
numbers. Sometimes, however, you may want to make transformations which are appropriate
only if particular variables are assumed to be either real or complex.

The function ComplexExpand expands out algebraic and trigonometric expressions, making
definite assumptions about the variables that appear.

ComplexExpand@exprD expand expr assuming that all variables are real

ComplexExpand@expr,8x1 ,x2 ,<D
expand expr assuming that the xi are complex

Expanding complex expressions.

This expands the expression, assuming that x and y are both real.
In[1]:= ComplexExpand@Tan@x + I yDD
Sin@2 xD Sinh@2 yD
Out[1]= +
Cos@2 xD + Cosh@2 yD Cos@2 xD + Cosh@2 yD

In this case, a is assumed to be real, but x is assumed to be complex, and is broken into
explicit real and imaginary parts.
In[2]:= ComplexExpand@a + x ^ 2, 8x<D

Out[2]= a - Im@xD2 + 2 Im@xD Re@xD + Re@xD2

With several complex variables, you quickly get quite complicated results.
In[3]:= ComplexExpand@Sin@xD Exp@yD, 8x, y<D
Out[3]= Re@yD Cos@Im@yDD Cosh@Im@xDD Sin@Re@xDD - Re@yD Cos@Re@xDD Sin@Im@yDD Sinh@Im@xDD +
IRe@yD Cosh@Im@xDD Sin@Im@yDD Sin@Re@xDD + Re@yD Cos@Im@yDD Cos@Re@xDD Sinh@Im@xDDM

There are several ways to write a complex variable z in terms of real parameters. As above, for
example, z can be written in the "Cartesian form" Re@zD + I Im@zD. But it can equally well be
written in the "polar form" Abs@zD Exp@I Arg@zDD.

The option TargetFunctions in ComplexExpand allows you to specify how complex variables
should be written. TargetFunctions can be set to a list of functions from the set
8Re, Im, Abs, Arg, Conjugate, Sign<. ComplexExpand will try to give results in terms of
whichever of these functions you request. The default is typically to give results in terms of Re
and Im.
The option TargetFunctions in ComplexExpand allows you to specify how complex variables
Mathematics and Algorithms 81

8Re, Im, Abs, Arg, Conjugate, Sign<. ComplexExpand will try to give results in terms of
whichever of these functions you request. The default is typically to give results in terms of Re
and Im.

This gives an expansion in Cartesian form.

In[4]:= ComplexExpand@Re@z ^ 2D, 8z<D

Out[4]= -Im@zD2 + Re@zD2

Here is an expansion in polar form.

In[5]:= ComplexExpand@Re@z ^ 2D, 8z<, TargetFunctions -> 8Abs, Arg<D

Out[5]= Abs@zD2 Cos@Arg@zDD2 - Abs@zD2 Sin@Arg@zDD2

Here is another form of expansion.

In[6]:= ComplexExpand@Re@z ^ 2D, 8z<, TargetFunctions -> ConjugateD
z2 Conjugate@zD2
Out[6]= +
2 2

Logical and Piecewise Functions

Nested logical and piecewise functions can be expanded out much like nested arithmetic func-
tions. You can do this using LogicalExpand and PiecewiseExpand.

LogicalExpand@exprD expand out logical functions in expr

PiecewiseExpand@exprD expand out piecewise functions in expr
PiecewiseExpand@expr,assumD expand out with the specified assumptions

Expanding out logical and piecewise functions.

LogicalExpand puts logical expressions into a standard disjunctive normal form (DNF), consist-
ing of an OR of ANDs.

By default, Mathematica leaves this expression unchanged.

In[1]:= Ha bL && c
Out[1]= Ha bL && c
82 Mathematics and Algorithms

LogicalExpand expands this into an OR of ANDs.

In[2]:= LogicalExpand@%D
Out[2]= Ha && cL Hb && cL

LogicalExpand works on all logical functions, always converting them into a standard OR of
ANDs form. Sometimes the results are inevitably quite large.

Xor can be expressed as an OR of ANDs.

In[3]:= LogicalExpand@Xor@a, b, cDD
Out[3]= Ha && b && cL Ha && ! b && ! cL Hb && ! a && ! cL Hc && ! a && ! bL

Any collection of nested conditionals can always in effect be flattened into a piecewise normal
form consisting of a single Piecewise object. You can do this in Mathematica using
PiecewiseExpand.

By default, Mathematica leaves this expression unchanged.

In[4]:= If@x > 0, If@x < 1, a, bD, cD
Out[4]= If@x > 0, If@x < 1, a, bD, cD

PiecewiseExpand flattens it into a single Piecewise object.

In[5]:= PiecewiseExpand@%D
a 0<x<1
Out[5]= b x1
c True

Functions like Max and Abs, as well as Clip and UnitStep, implicitly involve conditionals, and
combinations of them can again be reduced to a single Piecewise object using
PiecewiseExpand.

This gives a result as a single Piecewise object.

In[6]:= PiecewiseExpand@Max@Min@a, bD, cDD
a a - c > 0 && a - b 0
Out[6]= b a - b > 0 && b - c > 0
c True

With x assumed real, this can also be written as a Piecewise object.

In[7]:= PiecewiseExpand@Abs@xD, x RealsD
-x x < 0
Out[7]=
x True

Functions like Floor, Mod and FractionalPart can also be expressed in terms of Piecewise
objects, though in principle they can involve an infinite number of cases.
Mathematics and Algorithms 83

Functions like Floor, Mod and FractionalPart can also be expressed in terms of Piecewise
objects, though in principle they can involve an infinite number of cases.

Without a bound on x, this would yield an infinite number of cases.

In[8]:= PiecewiseExpand@Floor@x ^ 2D, 0 < x < 2D

1 1x< 2
Out[8]= 2 2 x< 3
3 x 3

Mathematica by default limits the number of cases that Mathematica will explicitly generate in
the expansion of any single piecewise function such as Floor at any stage in a computation.
You can change this limit by resetting the value of $MaxPiecewiseCases.

Simplification

Simplify@exprD try various algebraic and trigonometric transformations to

simplify an expression
FullSimplify@exprD try a much wider range of transformations

Simplifying expressions.

Mathematica does not automatically simplify an algebraic expression like this.

In[1]:= H1 - xL H1 - x ^ 2L
1-x
Out[1]=
1 - x2

Simplify performs the simplification.

In[2]:= Simplify@%D
1
Out[2]=
1+x

Simplify performs standard algebraic and trigonometric simplifications.

In[3]:= Simplify@Sin@xD ^ 2 + Cos@xD ^ 2D
Out[3]= 1
84 Mathematics and Algorithms

It does not, however, do more sophisticated transformations that involve, for example, special
functions.
In[4]:= Simplify@Gamma@1 + nD nD
Gamma@1 + nD
Out[4]=
n

FullSimplify does perform such transformations.

In[5]:= FullSimplify@%D
Out[5]= Gamma@nD

FullSimplifyAexpr,ExcludedForms->patternE
try to simplify expr, without touching subexpressions that
match pattern

Controlling simplification.

Here is an expression involving trigonometric functions and square roots.

In[6]:= t = H1 - Sin@xD ^ 2L Sqrt@Expand@H1 + Sqrt@2DL ^ 20DD

Out[6]= 22 619 537 + 15 994 428 2 I1 - Sin@xD2 M

By default, FullSimplify will try to simplify everything.

In[7]:= FullSimplify@tD

Out[7]= 3363 + 2378 2 Cos@xD2

This makes FullSimplify avoid simplifying the square roots.

In[8]:= FullSimplify@t, ExcludedForms -> Sqrt@_DD

Out[8]= 22 619 537 + 15 994 428 2 Cos@xD2

Mathematics and Algorithms 85

FullSimplifyAexpr,TimeConstraint->tE
try to simplify expr, working for at most t seconds on each
transformation
FullSimplifyAexpr,TransformationFunctions->9 f1 , f2 ,=E
use only the functions fi in trying to transform parts of expr

FullSimplifyAexpr,TransformationFunctions->9Automatic, f1 , f2 ,=E
use built-in transformations as well as the fi

SimplifyAexpr,ComplexityFunction->cE
and FullSimplifyAexpr,ComplexityFunction->cE

simplify using c to determine what form is considered

simplest

Further control of simplification.

In both Simplify and FullSimplify there is always an issue of what counts as the "simplest"
form of an expression. You can use the option ComplexityFunction -> c to provide a function
to determine this. The function will be applied to each candidate form of the expression, and
the one that gives the smallest numerical value will be considered simplest.

With its default definition of simplicity, Simplify leaves this unchanged.

In[9]:= Simplify@4 Log@10DD
Out[9]= 4 Log@10D

This now tries to minimize the number of elements in the expression.

In[10]:= Simplify@4 Log@10D, ComplexityFunction -> LeafCountD
Out[10]= Log@10 000D

Using Assumptions
Mathematica normally makes as few assumptions as possible about the objects you ask it to
manipulate. This means that the results it gives are as general as possible. But sometimes
these results are considerably more complicated than they would be if more assumptions were
made.
86 Mathematics and Algorithms

Refine@expr,assumD refine expr using assumptions

Simplify@expr,assumD simplify with assumptions
FullSimplify@expr,assumD full simplify with assumptions
FunctionExpand@expr,assumD function expand with assumptions

Doing operations with assumptions.

Simplify by default does essentially nothing with this expression.

In[1]:= Simplify@1 Sqrt@xD - Sqrt@1 xDD

1 1
Out[1]= - +
x x

The reason is that its value is quite different for different choices of x.
In[2]:= % . x -> 8- 3, - 2, - 1, 1, 2, 3<
2
Out[2]= :- , - 2 , -2 , 0, 0, 0>
3

With the assumption x > 0, Simplify can immediately reduce the expression to 0.
In[3]:= Simplify@1 Sqrt@xD - Sqrt@1 xD, x > 0D
Out[3]= 0

Without making assumptions about x and y, nothing can be done.

In[4]:= FunctionExpand@Log@x yDD
Out[4]= Log@x yD

If x and y are both assumed positive, the log can be expanded.

In[5]:= FunctionExpand@Log@x yD, x > 0 && y > 0D
Out[5]= Log@xD + Log@yD

By applying Simplify and FullSimplify with appropriate assumptions to equations and

inequalities you can in effect establish a vast range of theorems.

Without making assumptions about x the truth or falsity of this equation cannot be determined.
In[6]:= Simplify@Abs@xD == xD
Out[6]= x Abs@xD

Now Simplify can prove that the equation is true.

Mathematics and Algorithms 87

Now Simplify can prove that the equation is true.

In[7]:= Simplify@Abs@xD == x, x > 0D
Out[7]= True

This establishes the standard result that the arithmetic mean is larger than the geometric one.
In[8]:= Simplify@Hx + yL 2 >= Sqrt@x yD, x >= 0 && y >= 0D
Out[8]= True

This proves that erfHxL lies in the range H0, 1L for all positive arguments.
In[9]:= FullSimplify@0 < Erf@xD < 1, x > 0D
Out[9]= True

Simplify and FullSimplify always try to find the simplest forms of expressions. Sometimes,
however, you may just want Mathematica to follow its ordinary evaluation process, but with
certain assumptions made. You can do this using Refine. The way it works is that
Refine@expr, assumD performs the same transformations as Mathematica would perform automat -
ically if the variables in expr were replaced by numerical expressions satisfying the assumptions
assum.

There is no simpler form that Simplify can find.

In[10]:= Simplify@Log@xD, x < 0D
Out[10]= Log@xD

Refine just evaluates Log@xD as it would for any explicit negative number x.
In[11]:= Refine@Log@xD, x < 0D
Out[11]= p + Log@-xD

An important class of assumptions is those which assert that some object is an element of a
particular domain. You can set up such assumptions using x dom, where the character can
be entered as el or @ElementD.

xdom or Element@x,domD assert that x is an element of the domain dom

8x1 ,x2 ,<dom assert that all the xi are elements of the domain dom

pattdom assert that any expression which matches patt is an ele -

ment of the domain dom

Asserting that objects are elements of domains.

This confirms that p is an element of the domain of real numbers.

88 Mathematics and Algorithms

This confirms that p is an element of the domain of real numbers.

In[12]:= Pi Reals
Out[12]= True

These numbers are all elements of the domain of algebraic numbers.

In[13]:= 81, Sqrt@2D, 3 + Sqrt@5D< Algebraics
Out[13]= True

Mathematica knows that p is not an algebraic number.

In[14]:= Pi Algebraics
Out[14]= False

Current mathematics has not established whether e + p is an algebraic number or not.

In[15]:= E + Pi Algebraics
Out[15]= + p Algebraics

This represents the assertion that the symbol x is an element of the domain of real numbers.
In[16]:= x Reals
Out[16]= x Reals

Complexes the domain of complex numbers

Reals the domain of real numbers

Algebraics the domain of algebraic numbers

Rationals the domain of rational numbers

Integers the domain of integers

Primes the domain of primes

Booleans the domain of Booleans (True and False )

Domains supported by Mathematica.

If n is assumed to be an integer, sinHnpL is zero.

In[17]:= Simplify@Sin@n PiD, n IntegersD
Out[17]= 0
Mathematics and Algorithms 89

This establishes the theorem coshHxL 1 if x is assumed to be a real number.

In[18]:= Simplify@Cosh@xD >= 1, x RealsD
Out[18]= True

If you say that a variable satisfies an inequality, Mathematica will automatically assume that it
is real.
In[19]:= Simplify@x Reals, x > 0D
Out[19]= True

By using Simplify, FullSimplify and FunctionExpand with assumptions you can access
many of Mathematica's vast collection of mathematical facts.

This uses the periodicity of the tangent function.

In[20]:= Simplify@Tan@x + Pi kD, k IntegersD
Out[20]= Tan@xD

The assumption k 2 Integers implies that k must be even.

In[21]:= Simplify@Tan@x + Pi k 2D, k 2 IntegersD
Out[21]= Tan@xD

Mathematica knows that logHxL < expHxL for positive x.

In[22]:= Simplify@Log@xD < Exp@xD, x > 0D
Out[22]= True

FullSimplify accesses knowledge about special functions.

In[23]:= FullSimplify@Im@BesselJ@0, xDD, x RealsD
Out[23]= 0

Mathematica knows about discrete mathematics and number theory as well as continuous
mathematics.

This uses Wilson's theorem to simplify the result.

In[24]:= FunctionExpand@Mod@Hp - 1L !, pD, p PrimesD
Out[24]= -1 + p
90 Mathematics and Algorithms

This uses the multiplicative property of the Euler phi function.

In[25]:= FunctionExpand@EulerPhi@m nD, 8m, n< Integers && GCD@m, nD == 1D
Out[25]= EulerPhi@mD EulerPhi@nD

In something like Simplify@expr, assumD or Refine@expr, assumD you explicitly give the assump-
tions you want to use. But sometimes you may want to specify one set of assumptions to use in
a whole collection of operations. You can do this by using Assuming.

Assuming@assum,exprD use assumptions assum in the evaluation of expr

$Assumptions the default assumptions to use

Specifying assumptions with larger scopes.

This tells Simplify to use the default assumption x > 0.

In[26]:= Assuming@x > 0, Simplify@Sqrt@x ^ 2DDD
Out[26]= x

This combines the two assumptions given.

In[27]:= Assuming@x > 0, Assuming@x Integers, Refine@Floor@Sqrt@x ^ 2DDDDD
Out[27]= x

Functions like Simplify and Refine take the option Assumptions , which specifies what default
assumptions they should use. By default, the setting for this option is
Assumptions :> $Assumptions. The way Assuming then works is to assign a local value to
$Assumptions, much as in Block.

In addition to Simplify and Refine, a number of other functions take Assumptions options,
and thus can have assumptions specified for them by Assuming. Examples are
FunctionExpand, Integrate, Limit, Series, LaplaceTransform.

The assumption is automatically used in Integrate.

In[28]:= Assuming@n > 0, 1 + Integrate@x ^ n, 8x, 0, 1<D ^ 2D
1
Out[28]= 1+
H1 + nL2
Mathematics and Algorithms 91

Manipulating Equations and Inequalities

Equations
"Defining Variables" discussed assignments such as x = y which set x equal to y. Here we discuss
equations, which test equality. The equation x == y tests whether x is equal to y.

This tests whether 2 + 2 and 4 are equal. The result is the symbol True .
In[1]:= 2 + 2 == 4
Out[1]= True

It is very important that you do not confuse x = y with x == y. While x = y is an imperative state-
ment that actually causes an assignment to be done, x == y merely tests whether x and y are
equal, and causes no explicit action. If you have used the C programming language, you will
recognize that the notation for assignment and testing in Mathematica is the same as in C.

x=y assigns x to have value y

x==y tests whether x and y are equal

Assignments and tests.

This assigns x to have value 4.

In[2]:= x=4
Out[2]= 4

If you ask for x, you now get 4.

In[3]:= x
Out[3]= 4

This tests whether x is equal to 4. In this case, it is.

In[4]:= x == 4
Out[4]= True
92 Mathematics and Algorithms

x is equal to 4, not 6.
In[5]:= x == 6
Out[5]= False

This removes the value assigned to x.

In[6]:= x =.

The tests we have used so far involve only numbers, and always give a definite answer, either
True or False. You can also do tests on symbolic expressions.

Mathematica cannot get a definite result for this test unless you give x a specific numerical
value.
In[7]:= x == 5
Out[7]= x5

If you replace x by the specific numerical value 4, the test gives False .
In[8]:= % . x -> 4
Out[8]= False

Even when you do tests on symbolic expressions, there are some cases where you can get
definite results. An important one is when you test the equality of two expressions that are
identical. Whatever the numerical values of the variables in these expressions may be, Mathe-
matica knows that the expressions must always be equal.

The two expressions are identical, so the result is True , whatever the value of x may be.
In[9]:= 2 x + x ^ 2 == 2 x + x ^ 2
Out[9]= True

Mathematica does not try to tell whether these expressions are equal. In this case, using
Expand would make them have the same form.
In[10]:= 2 x + x ^ 2 == x H2 + xL

Out[10]= 2 x + x2 x H2 + xL

Expressions like x == 4 represent equations in Mathematica. There are many functions in Mathe-
matica for manipulating and solving equations.
Mathematics and Algorithms 93

This is an equation in Mathematica. "Solving Equations" discusses how to solve it for x.

In[11]:= x ^ 2 + 2 x - 7 == 0

Out[11]= -7 + 2 x + x2 0

You can assign a name to the equation.

In[12]:= eqn = %

Out[12]= -7 + 2 x + x2 0

If you ask for eqn, you now get the equation.

In[13]:= eqn

Out[13]= -7 + 2 x + x2 0

Solving Equations
An expression like x ^ 2 + 2 x - 7 == 0 represents an equation in Mathematica. You will often need
to solve equations like this, to find out for what values of x they are true.

This gives the two solutions to the quadratic equation x2 + 2 x - 7 = 0. The solutions are given as
replacements for x.
In[1]:= Solve@x ^ 2 + 2 x - 7 == 0, xD

Out[1]= ::x -1 - 2 2 >, :x -1 + 2 2 >>

Here are the numerical values of the solutions.

In[2]:= N@%D
Out[2]= 88x -3.82843<, 8x 1.82843<<

You can get a list of the actual solutions for x by applying the rules generated by Solve to x
using the replacement operator.
In[3]:= x . %
Out[3]= 8-3.82843, 1.82843<

You can equally well apply the rules to any other expression involving x.
In[4]:= x ^ 2 + 3 x . %%
Out[4]= 83.17157, 8.82843<
94 Mathematics and Algorithms

Solve@lhs==rhs,xD solve an equation, giving a list of rules for x

x.solution use the list of rules to get values for x
expr.solution use the list of rules to get values for an expression

Finding and using solutions to equations.

Solve always tries to give you explicit formulas for the solutions to equations. However, it is a
basic mathematical result that, for sufficiently complicated equations, explicit algebraic formu-
las in terms of radicals cannot be given. If you have an algebraic equation in one variable, and
the highest power of the variable is at most four, then Mathematica can always give you formu-
las for the solutions. However, if the highest power is five or more, it may be mathematically
impossible to give explicit algebraic formulas for all the solutions.

Mathematica can always solve algebraic equations in one variable when the highest power is
less than five.
In[5]:= Solve@x ^ 4 - 5 x ^ 2 - 3 == 0, xD

5 37 5 37 1 1
Out[5]= ::x - + >, :x + >, :x - -5 + 37 >, :x -5 + 37 >>
2 2 2 2 2 2

It can solve some equations that involve higher powers.

In[6]:= Solve@x ^ 6 == 1, xD
13 13 23 23
Out[6]= 98x -1<, 8x 1<, 9x -H-1L =, 9x H-1L =, 9x -H-1L =, 9x H-1L ==

There are some equations, however, for which it is mathematically impossible to find explicit
formulas for the solutions. Mathematica uses Root objects to represent the solutions in this
case.
In[7]:= Solve@2 - 4 x + x ^ 5 == 0, xD
5 5
Out[7]= 99x RootA2 - 4 1 + 1 &, 1E=, 9x RootA2 - 4 1 + 1 &, 2E=,
9x RootA2 - 4 1 + 15 &, 3E=, 9x RootA2 - 4 1 + 15 &, 4E=, 9x RootA2 - 4 1 + 15 &, 5E==

Even though you cannot get explicit formulas, you can still evaluate the solutions numerically.
In[8]:= N@%D
Out[8]= 88x -1.51851<, 8x 0.508499<, 8x 1.2436<, 8x -0.116792 - 1.43845 <, 8x -0.116792 + 1.43845 <<

In addition to being able to solve purely algebraic equations, Mathematica can also solve some
equations involving other functions.
Mathematics and Algorithms 95

After printing a warning, Mathematica returns one solution to this equation.

In[9]:= Solve@Sin@xD == a, xD

Solve::ifun : Inverse functions are being used by Solve, so some

solutions may not be found; use Reduce for complete solution information.
Out[9]= 88x ArcSin@aD<<

It is important to realize that an equation such as sinHxL = a actually has an infinite number of
possible solutions, in this case differing by multiples of 2 p. However, Solve by default returns
just one solution, but prints a message telling you that other solutions may exist. You can use
Reduce to get more information.

There is no explicit "closed form" solution for a transcendental equation like this.
In[10]:= Solve@Cos@xD == x, xD

Solve::tdep :
The equations appear to involve the variables to be solved for in an essentially non-algebraic way.
Out[10]= Solve@Cos@xD x, xD

You can find an approximate numerical solution using FindRoot , and giving a starting value
for x.
In[11]:= FindRoot@Cos@xD == x, 8x, 0<D
Out[11]= 8x 0.739085<

Solve can also handle equations involving symbolic functions. In such cases, it again prints a
warning, then gives results in terms of formal inverse functions.

Mathematica returns a result in terms of the formal inverse function of f.

In[12]:= Solve@f@x ^ 2D == a, xD

InverseFunction::ifun : Inverse functions are being used. Values may be lost for multivalued inverses.

Out[12]= ::x - fH-1L @aD >, :x fH-1L @aD >>

96 Mathematics and Algorithms

Solve@8lhs1 ==rhs1 ,lhs2 ==rhs2 ,<,8x,y,<D

solve a set of simultaneous equations for x, y,

Solving sets of simultaneous equations.

You can also use Mathematica to solve sets of simultaneous equations. You simply give the list
of equations, and specify the list of variables to solve for.

Here is a list of two simultaneous equations, to be solved for the variables x and y.
In[13]:= Solve@8a x + y == 0, 2 x + H1 - aL y == 1<, 8x, y<D
1 a
Out[13]= ::x - , y- >>
2
-2 + a - a 2 - a + a2

Here are some more complicated simultaneous equations. The two solutions are given as two
lists of replacements for x and y.
In[14]:= Solve@8x ^ 2 + y ^ 2 == 1, x + 3 y == 0<, 8x, y<D
3 1 3 1
Out[14]= ::x - , y >, :x , y- >>
10 10 10 10

This uses the solutions to evaluate the expression x + y.

In[15]:= x + y . %

2 2
Out[15]= :- , >
5 5

Mathematica can solve any set of simultaneous linear or polynomial equations.

When you are working with sets of equations in several variables, it is often convenient to
reorganize the equations by eliminating some variables between them.

This eliminates y between the two equations, giving a single equation for x.
In[16]:= Eliminate@8a x + y == 0, 2 x + H1 - aL y == 1<, yD
2
Out[16]= I2 - a + a M x 1

If you have several equations, there is no guarantee that there exists any consistent solution
for a particular variable.
Mathematics and Algorithms 97

There is no consistent solution to these equations, so Mathematica returns 8<, indicating that
the set of solutions is empty.
In[17]:= Solve@8x == 1, x == 2<, xD
Out[17]= 8<

There is also no consistent solution to these equations for almost all values of a.
In[18]:= Solve@8x == 1, x == a<, xD
Out[18]= 8<

The general question of whether a set of equations has any consistent solution is quite a subtle
one. For example, for most values of a, the equations 8x == 1, x == a< are inconsistent, so
there is no possible solution for x. However, if a is equal to 1, then the equations do have a
solution. Solve is set up to give you generic solutions to equations. It discards any solutions
that exist only when special constraints between parameters are satisfied.

If you use Reduce instead of Solve, Mathematica will however keep all the possible solutions to
a set of equations, including those that require special conditions on parameters.

This shows that the equations have a solution only when a == 1. The notation a == 1 && x == 1
represents the requirement that both a == 1 and x == 1 should be True .
In[19]:= Reduce@8x == a, x == 1<, xD
Out[19]= a 1 && x 1

This gives the complete set of possible solutions to the equation. The answer is stated in terms
of a combination of simpler equations. && indicates equations that must simultaneously be true;
indicates alternatives.
In[20]:= Reduce@a x - b == 0, xD
b
Out[20]= Hb 0 && a 0L a 0 && x
a

This gives a more complicated combination of equations.

In[21]:= Reduce@a x ^ 2 - b == 0, xD

b b
Out[21]= Hb 0 && a 0L a 0 && x - x
a a

This gives a symbolic representation of all solutions.

In[22]:= Reduce@Sin@xD == a, xD
Out[22]= C@1D Integers && Hx p - ArcSin@aD + 2 p C@1D x ArcSin@aD + 2 p C@1DL
98 Mathematics and Algorithms

Solve@lhs==rhs,xD solve an equation for x

Solve@8lhs1 ==rhs1 ,lhs2 ==rhs2 ,<,8x,y,<D
solve a set of simultaneous equations for x, y,
Eliminate@8lhs1 ==rhs1 ,lhs2 ==rhs2 ,<,8x,<D
eliminate x, in a set of simultaneous equations
Reduce@8lhs1 ==rhs1 ,lhs2 ==rhs2 ,<,8x,y,<D
give a set of simplified equations, including all possible
solutions

Functions for solving and manipulating equations.

Reduce also has powerful capabilities for handling equations specifically over real numbers or
integers. "Equations and Inequalities over Domains" discusses this in more detail.

This reduces the equation assuming x and y are complex.

In[23]:= Reduce@x ^ 2 + y ^ 2 == 1, yD

Out[23]= y - 1 - x2 y 1 - x2

This includes the conditions for x and y to be real.

In[24]:= Reduce@x ^ 2 + y ^ 2 == 1, y, RealsD

Out[24]= -1 x 1 && y - 1 - x2 y 1 - x2

This gives only the integer solutions.

In[25]:= Reduce@x ^ 2 + y ^ 2 == 1, y, IntegersD
Out[25]= Hx -1 && y 0L Hx 0 && y -1L Hx 0 && y 1L Hx 1 && y 0L

The Representation of Equations and Solutions

Mathematica treats equations as logical statements. If you type in an equation like
x ^ 2 + 3 x == 2, Mathematica interprets this as a logical statement which asserts that x ^ 2 + 3 x is
equal to 2. If you have assigned an explicit value to x, say x = 4, then Mathematica can explic-
itly determine that the logical statement x ^ 2 + 3 x == 2 is False.

If you have not assigned any explicit value to x, however, Mathematica cannot work out
whether x ^ 2 + 3 x == 2 is True or False. As a result, it leaves the equation in the symbolic form
x ^ 2 + 3 x == 2.

You can manipulate symbolic equations in Mathematica in many ways. One common goal is to
rearrange the equations so as to "solve" for a particular set of variables.
Mathematics and Algorithms 99

You can manipulate symbolic equations in Mathematica in many ways. One common goal is to
rearrange the equations so as to "solve" for a particular set of variables.

Here is a symbolic equation.

In[1]:= x ^ 2 + 3 x == 2

Out[1]= 3 x + x2 2

You can use the function Reduce to reduce the equation so as to give "solutions" for x. The
result, like the original equation, can be viewed as a logical statement.
In[2]:= Reduce@%, xD
1 1
Out[2]= x -3 - 17 x -3 + 17
2 2

The quadratic equation x ^ 2 + 3 x == 2 can be thought of as an implicit statement about the

value of x. As shown in the example above, you can use the function Reduce to get a more
explicit statement about the value of x. The expression produced by Reduce has the form
x ==r1 x ==r2 . This expression is again a logical statement, which asserts that either x is equal
to r1 , or x is equal to r2 . The values of x that are consistent with this statement are exactly the
same as the ones that are consistent with the original quadratic equation. For many purposes,
however, the form that Reduce gives is much more useful than the original equation.

You can combine and manipulate equations just like other logical statements. You can use
logical connectives such as and && to specify alternative or simultaneous conditions. You can
use functions like LogicalExpand, as well as FullSimplify, to simplify collections of equa-
tions.

For many purposes, you will find it convenient to manipulate equations simply as logical state-
ments. Sometimes, however, you will actually want to use explicit solutions to equations in
other calculations. In such cases, it is convenient to convert equations that are stated in the
form lhs == rhs into transformation rules of the form lhs -> rhs. Once you have the solutions to an
equation in the form of explicit transformation rules, you can substitute the solutions into
expressions by using the . operator.

Reduce produces a logical statement about the values of x corresponding to the roots of the
quadratic equation.
In[3]:= Reduce@x ^ 2 + 3 x == 2, xD
1 1
Out[3]= x -3 - 17 x -3 + 17
2 2

ToRules converts the logical statement into an explicit list of transformation rules.
100 Mathematics and Algorithms

ToRules converts the logical statement into an explicit list of transformation rules.
In[4]:= 8ToRules@%D<
1 1
Out[4]= ::x -3 - 17 >, :x -3 + 17 >>
2 2

You can now use the transformation rules to substitute the solutions for x into expressions
involving x.
In[5]:= x ^ 2 + a x . %
1 2 1 1 2 1
Out[5]= : -3 - 17 + -3 - 17 a, -3 + 17 + -3 + 17 a>
4 2 4 2

The function Solve produces transformation rules for solutions directly.

In[6]:= Solve@x ^ 2 + 3 x == 2, xD
1 1
Out[6]= ::x -3 - 17 >, :x -3 + 17 >>
2 2

Equations in One Variable

The main equations that Solve and related Mathematica functions deal with are polynomial
equations.

It is easy to solve a linear equation in x.

In[1]:= Solve@a x + b == c, xD
-b + c
Out[1]= ::x >>
a

One can also solve quadratic equations just by applying a simple formula.
In[2]:= Solve@x ^ 2 + a x + 2 == 0, xD
1 1
Out[2]= ::x -a - -8 + a2 >, :x -a + -8 + a2 >>
2 2

Mathematica can also find exact solutions to cubic equations. Here is the first solution to a
comparatively simple cubic equation.
In[3]:= Solve@x ^ 3 + 34 x + 1 == 0, xD@@1DD
13 1 13
2 J J-9 + 471 729 NN
2
Out[3]= :x -34 + >
3 J-9 + 471 729 N 323

For cubic and quartic equations the results are often complicated, but for all equations with
degrees up to four Mathematica is always able to give explicit formulas for the solutions.
Mathematics and Algorithms 101

For cubic and quartic equations the results are often complicated, but for all equations with
degrees up to four Mathematica is always able to give explicit formulas for the solutions.

An important feature of these formulas is that they involve only radicals: arithmetic combina-
tions of square roots, cube roots and higher roots.

It is a fundamental mathematical fact, however, that for equations of degree five or higher, it is
no longer possible in general to give explicit formulas for solutions in terms of radicals.

There are some specific equations for which this is still possible, but in the vast majority of
cases it is not.

This constructs a degree six polynomial.

In[4]:= Expand@Product@x ^ 2 - 2 i, 8i, 3<DD

Out[4]= -48 + 44 x2 - 12 x4 + x6

For a polynomial that factors in the way this one does, it is straightforward for Solve to find
the roots.
In[5]:= Solve@% == 0, xD

Out[5]= :8x -2<, 8x 2<, :x - 2 >, :x 2 >, :x - 6 >, :x 6 >>

This constructs a polynomial of degree eight.

In[6]:= Expand@x ^ 2 - 2 . x -> x ^ 2 - 3 . x -> x ^ 2 - 5D

Out[6]= 482 - 440 x2 + 144 x4 - 20 x6 + x8

The polynomial does not factor, but it can be decomposed into nested polynomials, so Solve
can again find explicit formulas for the roots.
In[7]:= Solve@% == 0, xD

Out[7]= ::x - 5- 3- 2 >, :x 5- 3- 2 >, :x - 5+ 3- 2 >, :x 5+ 3- 2 >,

:x - 5- 3+ 2 >, :x 5- 3+ 2 >, :x - 5+ 3+ 2 >, :x 5+ 3+ 2 >>

Root@ f ,kD the kth root of the equation f @xD == 0

Implicit representation for roots.

102 Mathematics and Algorithms

No explicit formulas for the solution to this equation can be given in terms of radicals, so
Mathematica uses an implicit symbolic representation.
In[8]:= Solve@x ^ 5 - x + 11 == 0, xD

Out[8]= 99x RootA11 - 1 + 15 &, 1E=, 9x RootA11 - 1 + 15 &, 2E=,

9x RootA11 - 1 + 15 &, 3E=, 9x RootA11 - 1 + 15 &, 4E=, 9x RootA11 - 1 + 15 &, 5E==

This finds a numerical approximation to each root.

In[9]:= N@%D
Out[9]= 88x -1.66149<, 8x -0.46194 - 1.565 <,
8x -0.46194 + 1.565 <, 8x 1.29268 - 0.903032 <, 8x 1.29268 + 0.903032 <<

If what you want in the end is a numerical solution, it is usually much faster to use NSolve
from the outset.
In[10]:= NSolve@x ^ 5 - x + 11 == 0, xD
Out[10]= 88x -1.66149<, 8x -0.46194 - 1.565 <,
8x -0.46194 + 1.565 <, 8x 1.29268 - 0.903032 <, 8x 1.29268 + 0.903032 <<

Root objects provide an exact, though implicit, representation for the roots of a polynomial.
You can work with them much as you would work with Sqrt@2D or any other expression that
represents an exact numerical quantity.

Here is the Root object representing the first root of the polynomial discussed above.
In[11]:= r = Root@ ^ 5 - + 11 &, 1D

Out[11]= RootA11 - 1 + 15 &, 1E

This is a numerical approximation to its value.

In[12]:= N@rD
Out[12]= -1.66149

Round does an exact computation to find the closest integer to the root.
In[13]:= Round@rD
Out[13]= -2

If you substitute the root into the original polynomial, and then simplify the result, you get
zero.
In[14]:= FullSimplify@x ^ 5 - x + 11 . x -> rD
Out[14]= 0
Mathematics and Algorithms 103

This finds the product of all the roots of the original polynomial.
In[15]:= FullSimplify@Product@Root@11 - + ^ 5 &, kD, 8k, 5<DD
Out[15]= -11

The complex conjugate of the third root is the second root.

In[16]:= Conjugate@Root@11 - + ^ 5 &, 3DD

Out[16]= RootA11 - 1 + 15 &, 2E

If the only symbolic parameter that exists in an equation is the variable that you are solving
for, then all the solutions to the equation will just be numbers. But if there are other symbolic
parameters in the equation, then the solutions will typically be functions of these parameters.

The solution to this equation can again be represented by Root objects, but now each Root
object involves the parameter a.
In[17]:= Solve@x ^ 5 + x + a == 0, xD

Out[17]= 99x RootAa + 1 + 15 &, 1E=, 9x RootAa + 1 + 15 &, 2E=,

9x RootAa + 1 + 15 &, 3E=, 9x RootAa + 1 + 15 &, 4E=, 9x RootAa + 1 + 15 &, 5E==

When a is replaced with 1, the Root objects can be simplified, and some are given as explicit
radicals.
In[18]:= Simplify@% . a -> 1D
1
Out[18]= :9x RootA1 - 12 + 13 &, 1E=, :x - - + 3 >,
2
1
:x + 3 >, 9x RootA1 - 12 + 13 &, 2E=, 9x RootA1 - 12 + 13 &, 3E=>
2

This shows the behavior of the first root as a function of a.

In[19]:= Plot@Root@ ^ 5 + + a &, 1D, 8a, - 2, 2<D
1.0

0.5

Out[19]=
-2 -1 1 2

-0.5

-1.0

This finds the derivative of the first root with respect to a.

In[20]:= D@Root@ ^ 5 + + a &, 1D, aD
1
Out[20]= -
4
1 + 5 RootAa + 1 + 15 &, 1E

If you give Solve any nth -degree polynomial equation, then it will always return exactly n solu-
tions, although some of these may be represented by Root objects. If there are degenerate
104 Mathematics and Algorithms

Solve gives two identical solutions to this equation.

In[21]:= Solve@Hx - 1L ^ 2 == 0, xD
Out[21]= 88x 1<, 8x 1<<

Here are the first four solutions to a tenth-degree equation. The solutions come in pairs.
In[22]:= Take@Solve@Hx ^ 5 - x + 11L ^ 2 == 0, xD, 4D

Out[22]= 99x RootA11 - 1 + 15 &, 1E=, 9x RootA11 - 1 + 15 &, 1E=,

9x RootA11 - 1 + 15 &, 2E=, 9x RootA11 - 1 + 15 &, 2E==

Mathematica also knows how to solve equations which are not explicitly in the form of polynomi-
als.

Here is an equation involving square roots.

In[23]:= Solve@Sqrt@xD + Sqrt@1 + xD == a, xD
1 - 2 a2 + a4
Out[23]= ::x >>
4 a2

And here is one involving logarithms.

In[24]:= Solve@Log@xD + Log@1 - xD == a, xD
1 1
Out[24]= ::x 1- 1 - 4 a >, :x 1+ 1 - 4 a >>
2 2

So long as it can reduce an equation to some kind of polynomial form, Mathematica will always
be able to represent its solution in terms of Root objects. However, with more general equa-
tions, involving say transcendental functions, there is no systematic way to use Root objects, or
even necessarily to find numerical approximations.

Here is a simple transcendental equation for x.

In[25]:= Solve@ArcSin@xD == a, xD
Out[25]= 88x Sin@aD<<
Mathematics and Algorithms 105

There is no solution to this equation in terms of standard functions.

In[26]:= Solve@Cos@xD == x, xD

Solve::tdep :
The equations appear to involve the variables to be solved for in an essentially non-algebraic way.
Out[26]= Solve@Cos@xD x, xD

Mathematica can nevertheless find a numerical solution even in this case.

In[27]:= FindRoot@Cos@xD == x, 8x, 0<D
Out[27]= 8x 0.739085<

Polynomial equations in one variable only ever have a finite number of solutions. But transcen-
dental equations often have an infinite number. Typically the reason for this is that functions
like Sin in effect have infinitely many possible inverses. With the default option setting
InverseFunctions -> True, Solve will nevertheless assume that there is a definite inverse for
any such function. Solve may then be able to return particular solutions in terms of this inverse
function.

Mathematica returns a particular solution in terms of ArcSin, but prints a warning indicating
that other solutions are lost.
In[28]:= Solve@Sin@xD == a, xD

Solve::ifun : Inverse functions are being used by Solve, so some

solutions may not be found; use Reduce for complete solution information.
Out[28]= 88x ArcSin@aD<<

Here the answer comes out in terms of ProductLog .

In[29]:= Solve@Exp@xD + x + 1 == 0, xD

InverseFunction::ifun : Inverse functions are being used. Values may be lost for multivalued inverses.

Solve::ifun : Inverse functions are being used by Solve, so some

solutions may not be found; use Reduce for complete solution information.
1
Out[29]= ::x -1 - ProductLogB F>>

If you ask Solve to solve an equation involving an arbitrary function like f, it will by default try
to construct a formal solution in terms of inverse functions.
106 Mathematics and Algorithms

Solve by default uses a formal inverse for the function f.

In[30]:= Solve@f@xD == a, xD

InverseFunction::ifun : Inverse functions are being used. Values may be lost for multivalued inverses.

Out[30]= 99x fH-1L @aD==

This is the structure of the inverse function.

In[31]:= InputForm@%D
Out[31]//InputForm= {{x -> InverseFunction[f, 1, 1][a]}}

InverseFunction@ f D the inverse function of f

InverseFunction@ f ,k,nD the inverse function of the n-argument function f with
respect to its kth argument

Inverse functions.

This returns an explicit inverse function.

In[32]:= InverseFunction@TanD
Out[32]= ArcTan

Mathematica can do formal operations on inverse functions.

In[33]:= D@InverseFunction@fD@x ^ 2D, xD
2x
Out[33]=
H-1L
f Af Ax2 EE

While Solve can only give specific solutions to an equation, Reduce can give a representation of
a whole solution set. For transcendental equations, it often ends up introducing new parame-
ters, say with values ranging over all possible integers.

This is a complete representation of the solution set.

In[34]:= Reduce@Sin@xD == a, xD
Out[34]= C@1D Integers && Hx p - ArcSin@aD + 2 p C@1D x ArcSin@aD + 2 p C@1DL
Mathematics and Algorithms 107

Here again is a representation of the general solution.

In[35]:= Reduce@Exp@xD + x + 1 == 0, xD
1
Out[35]= C@1D Integers && x -1 - ProductLogBC@1D, F

As discussed at more length in "Equations and Inequalities over Domains", Reduce allows you
to restrict the domains of variables. Sometimes this will let you generate definite solutions to
transcendental equations~or show that they do not exist.

With the domain of x restricted, this yields definite solutions.

In[36]:= Reduce@8Sin@xD == 1 2, Abs@xD < 4<, xD
7p p 5p
Out[36]= x- x x
6 6 6

With x constrained to be real, only one solution is possible.

In[37]:= Reduce@Exp@xD + x + 1 == 0, x, RealsD
1
Out[37]= x -1 - ProductLogB F

Reduce knows there can be no solution here.

In[38]:= Reduce@8Sin@xD == x, x > 1<, xD
Out[38]= False

Counting and Isolating Polynomial Roots

Counting Roots of Polynomials

CountRoots@poly,xD give the number of real roots of the polynomial poly in x

CountRoots@poly,8x,a,b<D give the number of roots of the polynomial poly in x with
ReHaL ReHrL ReHbL ImHaL ImHrL ImHbL

Counting roots of polynomials.

CountRoots accepts polynomials with Gaussian rational coefficients. The root count includes
multiplicities.
108 Mathematics and Algorithms

This gives the number of real roots of Ix2 - 2M Ix2 - 3M Ix2 - 4M.

In[1]:= CountRootsAIx2 - 2M Ix2 - 3M Ix2 - 4M, xE

Out[1]= 6

This counts the roots of Ix2 - 2M Ix2 - 3M Ix2 - 4M in the closed interval @1, 2D.

In[2]:= CountRootsAIx2 - 2M Ix2 - 3M Ix2 - 4M, 8x, 1, 2<E

Out[2]= 3

The roots of Ix2 + 1M x3 in the vertical axis segment between 0 and 2 consist of a triple root at 0
and a single root at .
In[3]:= CountRootsAIx2 + 1M x3 , 8x, 0, 2 <E
Out[3]= 4

This counts 17th -degree roots of unity in the closed unit square.
In[4]:= CountRootsAx17 - 1, 8x, 0, 1 + <E
Out[4]= 5

The coefficients of the polynomial can be Gaussian rationals.

3
In[5]:= CountRootsBx3 - x + - 1, 8x, 0, 1 + <F
4
Out[5]= 1

Isolating Intervals
A set S K, where K is or , is an isolating set for a root a of a polynomial f if a is the only
root of f in S. Isolating roots of a polynomial means finding disjoint isolating sets for all the
roots of the polynomial.

RootIntervals@8poly1 ,poly2 ,<D give a list of disjoint isolating intervals for the real roots of
any of the polyi , together with a list of which polynomials
actually have each successive root
RootIntervals@polyD give disjoint isolating intervals for real roots of a single
polynomial
Mathematics and Algorithms 109

RootIntervalsApolys,ComplexesE give disjoint isolating intervals or rectangles for complex

roots of polys
IsolatingInterval@aD give an isolating interval for the algebraic number a
IsolatingInterval@a,dxD give an isolating interval of width at most dx

Functions for isolating roots of polynomials.

RootIntervals accepts polynomials with rational number coefficients.

For a real root r the returned isolating interval is a pair of rational numbers 8a, b<, such that
either a < r < b or a b r. For a nonreal root r the isolating rectangle returned is a pair of Gaus-
sian rational numbers 8a, b<, such that ReHaL < ReHrL < ReHbL ImHaL < ImHrL < ImHbL and either ImHaL 0
or ImHbL 0.

Here are isolating intervals for the real roots of f .

In[6]:= f = Ix2 - 2M Ix2 - 3M Ix2 - 4M; RootIntervals@fD

3 3 3 3
Out[6]= ::8-2, -2<, :-2, - >, :- , -1>, :1, >, : , 2>, 82, 2<>, 881<, 81<, 81<, 81<, 81<, 81<<>
2 2 2 2

The second list shows which interval contains a root of which polynomial.
In[7]:= RootIntervals@8f + 3, f + 5, f + 7<D
5 9 9 9 9 5
Out[7]= :::- , - >, :- , -1>, 8-1, 0<, 80, 1<, :1, >, : , >>, 881<, 82<, 83<, 83<, 82<, 81<<>
4 8 8 8 8 4

This gives isolating intervals for all complex roots of f + 3.

In[8]:= RootIntervals@f + 3, ComplexesD
7 7 7 7 7 7
Out[8]= ::8-2, -1<, 81, 2<, :-7 - 7 , - >, :-7, - + 7 >, :- - 7 , >, :- , + 7 >>,
4 4 4 2 4 2
881<, 81<, 81<, 81<, 81<, 81<<>

Here are isolating intervals for the third- and fourth-degree roots of unity. The second interval
contains a root common to both polynomials.
In[9]:= RootIntervalsA9x3 - 1, x4 - 1=, ComplexesE
3 3 3 3 3 3 3 3
Out[9]= ::8-1, -1<, 80, 2<, :- - , - - >, :- >, + , - +
4 216 4 4 4 16 2
3 3 3 3 3 3
3 3
:- - , - >, :- + , + >>, 882<, 81, 2<, 81<, 81<, 82<, 82<<>
16 2 8 4 16 4 8 2
110 Mathematics and Algorithms

Here is an isolating interval for a root of a polynomial of degree seven.

In[10]:= IsolatingInterval@Root@ ^ 7 - 11 + 3 &, 5DD
Out[10]= 8-4, 4 <

This gives an isolating interval of width at most 10-10 .

In[11]:= IsolatingInterval@Root@ ^ 7 - 11 + 3 &, 5D, 10 ^ - 10D
866 877 392 461 355 978 878 543 3 467 509 569 843 5 695 662 056 689
Out[11]= :- + , - + >
1 099 511 627 776 274 877 906 944 4 398 046 511 104 4 398 046 511 104

All numbers in the interval have the first ten decimal digits in common.
In[12]:= N@%, 10D
Out[12]= 8-0.788420396 + 1.295043616 , -0.788420396 + 1.295043616 <

Algebraic Numbers

Root@ f ,kD the kth root of the polynomial equation f @xD == 0

The representation of algebraic numbers.

When you enter a Root object, the polynomial that appears in it is automatically reduced to a
minimal form.
In[1]:= Root@24 - 2 + 4 ^ 5 &, 1D

Out[1]= RootA12 - 1 + 2 15 &, 1E

This extracts the pure function which represents the polynomial, and applies it to x.
In[2]:= First@%D@xD

Out[2]= 12 - x + 2 x5

Root objects are the way that Mathematica represents algebraic numbers. Algebraic numbers
have the property that when you perform algebraic operations on them, you always get a single
algebraic number as the result.

Here is the square root of an algebraic number.

In[3]:= Sqrt@Root@2 - + ^ 5 &, 1DD

Out[3]= RootA2 - 1 + 15 &, 1E

RootReduce reduces this to a single Root object.

Mathematics and Algorithms 111

RootReduce reduces this to a single Root object.

In[4]:= RootReduce@%D

Out[4]= RootA2 - 12 + 110 &, 6E

Here is a more complicated expression involving an algebraic number.

In[5]:= Sqrt@2D + Root@2 - + ^ 5 &, 1D ^ 2
2
Out[5]= 2 + RootA2 - 1 + 15 &, 1E

Again this can be reduced to a single Root object, albeit a fairly complicated one.
In[6]:= RootReduce@%D

Out[6]= RootA14 - 72 1 + 25 12 - 144 13 - 88 14 - 8 15 + 62 16 - 14 18 + 110 &, 2E

RootReduce@exprD attempt to reduce expr to a single Root object

ToRadicals@exprD attempt to transform Root objects to explicit radicals

Operations on algebraic numbers.

In this simple case the Root object is automatically expressed in terms of radicals.
In[7]:= Root@ ^ 2 - - 1 &, 1D
1
Out[7]= 1- 5
2

When cubic polynomials are involved, Root objects are not automatically expressed in terms of
radicals.
In[8]:= Root@ ^ 3 - 2 &, 1D

Out[8]= RootA-2 + 13 &, 1E

ToRadicals attempts to express all Root objects in terms of radicals.

In[9]:= ToRadicals@%D

Out[9]= 213

If Solve and ToRadicals do not succeed in expressing the solution to a particular polynomial
equation in terms of radicals, then it is a good guess that this fundamentally cannot be done.
However, you should realize that there are some special cases in which a reduction to radicals
is in principle possible, but Mathematica cannot find it. The simplest example is the equation
x5 + 20 x + 32 = 0, but here the solution in terms of radicals is very complicated. The equation
1 1

x6 - 9 x4 - 4 x3 + 27 x2 - 36 x - 23 is another example, where now x = 2 3 + 3 2 is a solution.

equation in terms of radicals, then it is a good guess that this fundamentally cannot be done.
However, you should realize that there are some special cases in which a reduction to radicals
112 Mathematics and Algorithms

x5 + 20 x + 32 = 0, but here the solution in terms of radicals is very complicated. The equation
1 1

x6 - 9 x4 - 4 x3 + 27 x2 - 36 x - 23 is another example, where now x = 2 3 + 3 2 is a solution.

This gives a Root object involving a degree-six polynomial.

In[10]:= RootReduce@2 ^ H1 3L + Sqrt@3DD

Out[10]= RootA-23 - 36 1 + 27 12 - 4 13 - 9 14 + 16 &, 2E

Even though a simple form in terms of radicals does exist, ToRadicals does not find it.
In[11]:= ToRadicals@%D

Out[11]= RootA-23 - 36 1 + 27 12 - 4 13 - 9 14 + 16 &, 2E

Beyond degree four, most polynomials do not have roots that can be expressed at all in terms
of radicals. However, for degree five it turns out that the roots can always be expressed in
terms of elliptic or hypergeometric functions. The results, however, are typically much too
complicated to be useful in practice.

RootSum@ f , formD the sum of form@xD for all x satisfying the polynomial
equation f @xD == 0
Normal@exprD the form of expr with RootSum replaced by explicit sums of
Root objects

Sums of roots.

This computes the sum of the reciprocals of the roots of 1 + 2 x + x5 .

In[12]:= RootSum@H1 + 2 + ^ 5L &, H1 L &D
Out[12]= -2

Now no explicit result can be given in terms of radicals.

In[13]:= RootSum@H1 + 2 + ^ 5L &, H Log@1 + DL &D

Out[13]= RootSumA1 + 2 1 + 15 &, Log@1 + 1D 1 &E

This expands the RootSum into an explicit sum involving Root objects.
In[14]:= Normal@%D

Out[14]= LogA1 + RootA1 + 2 1 + 15 &, 1EE RootA1 + 2 1 + 15 &, 1E +

LogA1 + RootA1 + 2 1 + 15 &, 2EE RootA1 + 2 1 + 15 &, 2E +
LogA1 + RootA1 + 2 1 + 15 &, 3EE RootA1 + 2 1 + 15 &, 3E +
LogA1 + RootA1 + 2 1 + 15 &, 4EE RootA1 + 2 1 + 15 &, 4E +
LogA1 + RootA1 + 2 1 + 15 &, 5EE RootA1 + 2 1 + 15 &, 5E
Mathematics and Algorithms 113

RootApproximant@xD converts the number x to one of the "simplest" algebraic

numbers that approximates it well
RootApproximant@x,nD finds an algebraic number of degree at most n that approxi -
mates x

This recovers 2 from a numerical approximation.

In[15]:= RootApproximant@N@Sqrt@2DDD

Out[15]= 2

In this case, the result has degree at most 4.

In[16]:= RootApproximant@N@Sqrt@2D + Sqrt@3DD, 4D
2 4
Out[16]= RootA1 - 10 1 + 1 &, 4E

This confirms that the Root expression does correspond to 2 + 3.

In[17]:= RootReduce@Sqrt@2D + Sqrt@3DD
2 4
Out[17]= RootA1 - 10 1 + 1 &, 4E

Simultaneous Equations
You can give Solve a list of simultaneous equations to solve. Solve can find explicit solutions
for a large class of simultaneous polynomial equations.

Here is a simple linear equation with two unknowns.

In[1]:= Solve@8a x + b y == 1, x - y == 2<, 8x, y<D
-1 - 2 b -1 + 2 a
Out[1]= ::x - , y- >>
a+b a+b

Here is a more complicated example. The result is a list of solutions, with each solution consist -
ing of a list of transformation rules for the variables.
In[2]:= Solve@8x ^ 2 + y ^ 2 == 1, x + y == a<, 8x, y<D

1 1 a 2 - a2 1
Out[2]= ::x a- 2 - a2 , y a+ 2 - a2 >, :x + , y a- 2 - a2 >>
2 2 2 2 2
114 Mathematics and Algorithms

You can use the list of solutions with the . operator.

In[3]:= x ^ 3 + y ^ 4 . % . a -> 0.7
Out[3]= 80.846577, 0.901873<

Even when Solve cannot find explicit solutions, it often can "unwind" simultaneous equations
to produce a symbolic result in terms of Root objects.
In[4]:= First@Solve@8x ^ 2 + y ^ 3 == x y, x + y + x y == 1<, 8x, y<DD
1 2 3
Out[4]= :x J1 - RootA1 - 3 1 + 12 + 2 13 + 2 14 + 15 &, 1E - RootA1 - 3 1 + 12 + 2 13 + 2 14 + 15 &, 1E -
2
4
RootA1 - 3 1 + 12 + 2 13 + 2 14 + 15 &, 1E N, y RootA1 - 3 1 + 12 + 2 13 + 2 14 + 15 &, 1E>

You can then use N to get a numerical result.

In[5]:= N@%D
Out[5]= 8x -3.4875, y -1.80402<

The variables that you use in Solve do not need to be single symbols. Often when you set up
large collections of simultaneous equations, you will want to use expressions like a@iD as
variables.

Here is a list of three equations for the a@iD.

In[6]:= Table@2 a@iD + a@i - 1D == a@i + 1D, 8i, 3<D
Out[6]= 8a@0D + 2 a@1D a@2D, a@1D + 2 a@2D a@3D, a@2D + 2 a@3D a@4D<

This solves for some of the a@iD.

In[7]:= Solve@%, 8a@1D, a@2D, a@3D<D
1 1 1
Out[7]= ::a@1D H-5 a@0D + a@4DL, a@2D Ha@0D + a@4DL, a@3D H-a@0D + 5 a@4DL>>
12 6 12

Solve@eqns,8x1 ,x2 ,<D solve eqns for the specific objects xi

Solve@eqnsD try to solve eqns for all the objects that appear in them

Solving simultaneous equations.

If you do not explicitly specify objects to solve for, Solve will try to solve for all the variables.
In[8]:= Solve@8x + y == 1, x - 3 y == 2<D
5 1
Out[8]= ::x , y - >>
4 4
Mathematics and Algorithms 115

Solve@8lhs1 ==rhs1 ,lhs2 ==rhs2 ,<,varsD

Solve@lhs1 ==rhs1 &&lhs2 ==rhs2 &&,varsD
Solve@8lhs1 ,lhs2 ,<==8rhs1 ,rhs2 ,<,varsD

Ways to present simultaneous equations to Solve .

If you construct simultaneous equations from matrices, you typically get equations between
lists of expressions.
In[9]:= 883, 1<, 82, - 5<<.8x, y< == 87, 8<
Out[9]= 83 x + y, 2 x - 5 y< 87, 8<

Solve converts equations involving lists to lists of equations.

In[10]:= Solve@%, 8x, y<D
43 10
Out[10]= ::x , y- >>
17 17

You can use LogicalExpand to do the conversion explicitly.

In[11]:= LogicalExpand@%%D
Out[11]= 2 x - 5 y 8 && 3 x + y 7

In some kinds of computations, it is convenient to work with arrays of coefficients instead of

explicit equations. You can construct such arrays from equations by using CoefficientArrays.

Generic and Non-Generic Solutions

If you have an equation like 2 x == 0, it is perfectly clear that the only possible solution is
x -> 0. However, if you have an equation like a x == 0, things are not so clear. If a is not equal
to zero, then x -> 0 is again the only solution. However, if a is in fact equal to zero, then any
value of x is a solution. You can see this by using Reduce.

Solve implicitly assumes that the parameter a does not have the special value 0.
In[1]:= Solve@a x == 0, xD
Out[1]= 88x 0<<
116 Mathematics and Algorithms

Reduce, on the other hand, gives you all the possibilities, without assuming anything about the
value of a.
In[2]:= Reduce@a x == 0, xD
Out[2]= a 0 x 0

A basic difference between Reduce and Solve is that Reduce gives all the possible solutions to a
set of equations, while Solve gives only the generic ones. Solutions are considered "generic" if
they involve conditions only on the variables that you explicitly solve for, and not on other
parameters in the equations. Reduce and Solve also differ in that Reduce always returns combi-
nations of equations, while Solve gives results in the form of transformation rules.

Solve@eqns,varsD find generic solutions to equations

Reduce@eqns,varsD reduce equations, maintaining all solutions

Solving equations.

This is the solution to an arbitrary linear equation given by Solve .

In[3]:= Solve@a x + b == 0, xD
b
Out[3]= ::x - >>
a

Reduce gives the full version, which includes the possibility a == b == 0. In reading the output,
note that && has higher precedence than .
In[4]:= Reduce@a x + b == 0, xD
b
Out[4]= Hb 0 && a 0L a 0 && x -
a

Here is the full solution to a general quadratic equation. There are three alternatives. If a is
nonzero, then there are two solutions for x, given by the standard quadratic formula. If a is
zero, however, the equation reduces to a linear one. Finally, if a, b and c are all zero, there is
no restriction on x.
In[5]:= Reduce@a x ^ 2 + b x + c == 0, xD

-b - b2 - 4 a c -b + b2 - 4 a c
Out[5]= a 0 && x x
2a 2a
c
a 0 && b 0 && x - Hc 0 && b 0 && a 0L
b
Mathematics and Algorithms 117

When you have several simultaneous equations, Reduce can show you under what conditions
the equations have solutions. Solve shows you whether there are any generic solutions.

This shows there can never be any solution to these equations.

In[6]:= Reduce@8x == 1, x == 2<, xD
Out[6]= False

There is a solution to these equations, but only when a has the special value 1.
In[7]:= Reduce@8x == 1, x == a<, xD
Out[7]= a 1 && x 1

The solution is not generic, and is rejected by Solve .

In[8]:= Solve@8x == 1, x == a<, xD
Out[8]= 8<

But if a is constrained to have value 1, then Solve again returns a solution.

In[9]:= Solve@8x == 1, x == a, a == 1<, xD
Out[9]= 88x 1<<

This equation is true for any value of x.

In[10]:= Reduce@x == x, xD
Out[10]= True

This is the kind of result Solve returns when you give an equation that is always true.
In[11]:= Solve@x == x, xD
Out[11]= 88<<

When you work with systems of linear equations, you can use Solve to get generic solutions,
and Reduce to find out for what values of parameters solutions exist.

Here is a matrix whose i, jth element is i + j.

In[12]:= m = Table@i + j, 8i, 3<, 8j, 3<D
Out[12]= 882, 3, 4<, 83, 4, 5<, 84, 5, 6<<
118 Mathematics and Algorithms

The matrix has determinant zero.

In[13]:= Det@mD
Out[13]= 0

This makes a set of three simultaneous equations.

In[14]:= eqn = m.8x, y, z< == 8a, b, c<
Out[14]= 82 x + 3 y + 4 z, 3 x + 4 y + 5 z, 4 x + 5 y + 6 z< 8a, b, c<

Solve reports that there are no generic solutions.

In[15]:= Solve@eqn, 8x, y, z<D
Out[15]= 8<

Reduce, however, shows that there would be a solution if the parameters satisfied the special
condition a == 2 b - c.
In[16]:= Reduce@eqn, 8x, y, z<D
Out[16]= a 2 b - c && y -6 b + 5 c - 2 x && z 5 b - 4 c + x

For nonlinear equations, the conditions for the existence of solutions can be much more compli-
cated.

Here is a very simple pair of nonlinear equations.

In[17]:= eqn = 8x y == a, x ^ 2 y ^ 2 == b<

Out[17]= 9x y a, x2 y2 b=

Solve shows that the equations have no generic solutions.

In[18]:= Solve@eqn, 8x, y<D
Out[18]= 8<

Reduce gives the complete conditions for a solution to exist.

In[19]:= Reduce@eqn, 8x, y<D
a
Out[19]= Hb 0 && a 0 && x 0L a- b a b && x 0 && y
x
Mathematics and Algorithms 119

Eliminating Variables
When you write down a set of simultaneous equations in Mathematica, you are specifying a
collection of constraints between variables. When you use Solve, you are finding values for
some of the variables in terms of others, subject to the constraints represented by the
equations.

Solve@eqns,vars,elimsD find solutions for vars, eliminating the variables elims

Eliminate@eqns,elimsD rearrange equations to eliminate the variables elims

Eliminating variables.

Here are two equations involving x, y and the "parameters" a and b.

In[1]:= eqn = 8x + y == 6 a + 3 b, y == 9 a + 2 x<
Out[1]= 8x + y 6 a + 3 b, y 9 a + 2 x<

If you solve for both x and y, you get results in terms of a and b.
In[2]:= Solve@eqn, 8x, y<D
Out[2]= 88x -a + b, y 7 a + 2 b<<

Similarly, if you solve for x and a, you get results in terms of y and b.
In[3]:= Solve@eqn, 8x, a<D
1 1
Out[3]= ::x H9 b - yL, a H-2 b + yL>>
7 7

If you only want to solve for x, however, you have to specify whether you want to eliminate y
or a or b. This eliminates y, and so gives the result in terms of a and b.
In[4]:= Solve@eqn, x, yD
Out[4]= 88x -a + b<<

If you eliminate a, then you get a result in terms of y and b.

In[5]:= Solve@eqn, x, aD
1
Out[5]= ::x H9 b - yL>>
7
120 Mathematics and Algorithms

In some cases, you may want to construct explicitly equations in which variables have been
eliminated. You can do this using Eliminate.

This combines the two equations in the list eqn, by eliminating the variable a.
In[6]:= Eliminate@eqn, aD
Out[6]= 9 b - y 7 x

This is what you get if you eliminate y instead of a.

In[7]:= Eliminate@eqn, yD
Out[7]= b - x a

As a more sophisticated example of Eliminate, consider the problem of writing x5 + y5 in terms

of the "symmetric polynomials" x + y and x y.

To solve the problem, we simply have to write f in terms of a and b, eliminating the original
variables x and y.
In[8]:= Eliminate@8f == x ^ 5 + y ^ 5, a == x + y, b == x y<, 8x, y<D
5 3 2
Out[8]= f a - 5 a b + 5 a b

In dealing with sets of equations, it is common to consider some of the objects that appear as
true "variables", and others as "parameters". In some cases, you may need to know for what
values of parameters a particular relation between the variables is always satisfied.

SolveAlways @eqns,varsD solve for the values of parameters for which the eqns are
satisfied for all values of the vars

Solving for parameters that make relations always true.

This finds the values of parameters that make the equation hold for all x.
In[9]:= SolveAlways@a + b x + c x ^ 2 == H1 + xL ^ 2, xD
Out[9]= 88a 1, b 2, c 1<<

This equates two series.

In[10]:= Series@a Cos@xD + b Cos@2 xD + Cos@3 xD, 8x, 0, 3<D == Series@Cosh@xD, 8x, 0, 3<D
9 a x2
Out[10]= H1 + a + bL + - - - 2 b x2 + O@xD4 1 + + O@xD4
2 2 2
Mathematics and Algorithms 121

This finds values of the undetermined coefficients.

In[11]:= SolveAlways@%, xD
10 10
Out[11]= ::a , b- >>
3 3

Relational and Logical Operators

x==y equal (also input as x y)

x!=y unequal (also input as x y)
x>y greater than
x>=y greater than or equal to (also input as x y)
x<y less than
x<=y less than or equal to (also input as x y)
x==y==z all equal
x!=y!=z all unequal (distinct)
x>y>z , etc. strictly decreasing, etc.

Relational operators.

This tests whether 10 is less than 7. The result is False .

In[1]:= 10 < 7
Out[1]= False

Not all of these numbers are unequal, so this gives False .

In[2]:= 3 != 2 != 3
Out[2]= False

You can mix < and <=.

In[3]:= 3 < 5 <= 6
Out[3]= True

Since both of the quantities involved are numeric, Mathematica can determine that this is true.
In[4]:= Pi ^ E < E ^ Pi
Out[4]= True

Mathematica does not know whether this is true or false.

122 Mathematics and Algorithms

Mathematica does not know whether this is true or false.

In[5]:= x>y
Out[5]= x>y

!p not (also input as p)

p&&q&& and (also input as p fl q fl )
pq or (also input as p fi q fi )
Xor@p,q,D exclusive or (also input as p q )
Nand@p,q,D and Nor@p,q,D
nand and nor (also input as and )
If@p,then,elseD give then if p is True , and else if p is False
LogicalExpand@exprD expand out logical expressions

Logical operations.

Both tests give True , so the result is True .

In[6]:= 7 > 4 && 2 != 3
Out[6]= True

You should remember that the logical operations ==, && and are all double characters in
Mathematica. If you have used a programming language such as C, you will be familiar with
this notation.

Mathematica does not know whether this is true or false.

In[7]:= p && q
Out[7]= p && q

Mathematica leaves this expression unchanged.

In[8]:= Hp qL && ! Hr sL
Out[8]= Hp qL && ! Hr sL

You can use LogicalExpand to expand out the terms.

In[9]:= LogicalExpand@%D
Out[9]= Hp && ! r && ! sL Hq && ! r && ! sL

Solving Logical Combinations of Equations

Mathematics and Algorithms 123

Solving Logical Combinations of Equations

When you give a list of equations to Solve, it assumes that you want all the equations to be
satisfied simultaneously. It is also possible to give Solve more complicated logical combinations
of equations.

Solve assumes that the equations x + y == 1 and x - y == 2 are simultaneously valid.

In[1]:= Solve@8x + y == 1, x - y == 2<, 8x, y<D
3 1
Out[1]= ::x , y - >>
2 2

Here is an alternative form, using the logical connective && explicitly.

In[2]:= Solve@x + y == 1 && x - y == 2, 8x, y<D
3 1
Out[2]= ::x , y - >>
2 2

This specifies that either x + y == 1 or x - y == 2. Solve gives two solutions for x, correspond-
ing to these two possibilities.
In[3]:= Solve@x + y == 1 x - y == 2, xD
Out[3]= 88x 1 - y<, 8x 2 + y<<

Solve gives three solutions to this equation.

In[4]:= Solve@x ^ 3 == x, xD
Out[4]= 88x -1<, 8x 0<, 8x 1<<

If you explicitly include the assertion that x != 0, one of the previous solutions is suppressed.
In[5]:= Solve@x ^ 3 == x && x != 0, xD
Out[5]= 88x -1<, 8x 1<<

Here is a slightly more complicated example. Note that the precedence of is lower than the
precedence of &&, so the equation is interpreted as Hx ^ 3 == x && x != 1L x ^ 2 == 2, not
x ^ 3 == x && Hx != 1 x ^ 2 == 2L.
In[6]:= Solve@x ^ 3 == x && x != 1 x ^ 2 == 2, xD

Out[6]= :8x -1<, 8x 0<, :x - 2 >, :x 2 >>

When you use Solve, the final results you get are in the form of transformation rules. If you
use Reduce or Eliminate, on the other hand, then your results are logical statements, which
you can manipulate further.
124 Mathematics and Algorithms

This gives a logical statement representing the solutions of the equation x ^ 2 == x.

In[7]:= Reduce@x ^ 2 == x, xD
Out[7]= x 0 x 1

This finds values of x which satisfy x ^ 5 == x but do not satisfy the statement representing the
solutions of x ^ 2 == x.
In[8]:= Reduce@x ^ 5 == x && ! %, xD
Out[8]= x -1 x - x

The logical statements produced by Reduce can be thought of as representations of the solution
set for your equations. The logical connectives &&, and so on then correspond to operations
on these sets.

eqns1 eqns2 union of solution sets

eqns1 &&eqns2 intersection of solution sets
!eqns complement of a solution set
Implies@eqns1 ,eqns2 D the part of eqns1 that contains eqns2

Operations on solution sets.

You may often find it convenient to use special notations for logical connectives, as discussed in
"Operators".

The input uses special notations for Implies and Or .

In[9]:= Reduce@x ^ 2 == 1 fl Hx == 1 x == - 1L, xD
Out[9]= True

Inequalities
Just as the equation x ^ 2 + 3 x == 2 asserts that x ^ 2 + 3 x is equal to 2, so also the inequality
x ^ 2 + 3 x > 2 asserts that x ^ 2 + 3 x is greater than 2. In Mathematica, Reduce works not only on
equations, but also on inequalities.
Mathematics and Algorithms 125

Reduce@8ineq1 ,ineq2 ,<,xD reduce a collection of inequalities in x

Manipulating univariate inequalities.

This pair of inequalities reduces to a single inequality.

In[1]:= Reduce@80 < x < 2, 1 < x < 4<, xD
Out[1]= 1 < x < 2

These inequalities can never simultaneously be satisfied.

In[2]:= Reduce@8x < 1, x > 3<, xD
Out[2]= False

When applied to an equation, Reduce@eqn, xD tries to get a result consisting of simple equations
for x of the form x ==r1 , . When applied to an inequality, Reduce@ineq, xD does the exactly
analogous thing, and tries to get a result consisting of simple inequalities for x of the form
l1 < x <r1 , .

This reduces a quadratic equation to two simple equations for x.

In[3]:= Reduce@x ^ 2 + 3 x == 2, xD
1 1
Out[3]= x -3 - 17 x -3 + 17
2 2

This reduces a quadratic inequality to two simple inequalities for x.

In[4]:= Reduce@x ^ 2 + 3 x > 2, xD
1 1
Out[4]= x < -3 - 17 x > -3 + 17
2 2

You can think of the result generated by Reduce@ineq, xD as representing a series of intervals,
described by inequalities. Since the graph of a polynomial of degree n can go up and down as
many as n times, a polynomial inequality of degree n can give rise to as many as n 2 + 1 distinct
intervals.

This inequality yields three distinct intervals.

In[5]:= Reduce@Hx - 1L Hx - 2L Hx - 3L Hx - 4L > 0, xD
Out[5]= x < 1 2 < x < 3 x > 4
126 Mathematics and Algorithms

The ends of the intervals are at roots and poles.

In[6]:= Reduce@1 < Hx ^ 2 + 3 xL Hx + 1L < 2, xD

Out[6]= -1 - 2 < x < -2 -1 + 2 <x<1

Solving this inequality requires introducing ProductLog .

In[7]:= Reduce@x - 2 < Log@xD < x, xD

Reduce::ztest : Unable to decide whether numeric quantities

1
:-2 - Log@-ProductLog@-Power@2DDD - ProductLogB- F, -2 - Log@-ProductLog@-1, -Power@2DD
2
1
D - ProductLogB-1, - F> are equal to zero. Assuming they are.
2
1 1
Out[7]= -ProductLogB- F < x < -ProductLogB-1, - F
2 2

Transcendental functions like sin HxL have graphs that go up and down infinitely many times, so
that infinitely many intervals can be generated.

The second inequality allows only finitely many intervals.

In[8]:= Reduce@8Sin@xD > 0, 0 < x < 20<, xD
Out[8]= 0 < x < p 2 p < x < 3 p 4 p < x < 5 p 6 p < x < 20

This is how Reduce represents infinitely many intervals.

In[9]:= Reduce@8Sin@xD > 0, 0 < x<, xD
Out[9]= C@1D Integers && H0 < x < p HC@1D 1 && 2 p C@1D < x < p + 2 p C@1DLL

Fairly simple inputs can give fairly complicated results.

In[10]:= Reduce@8Sin@xD ^ 2 + Sin@3 xD > 0, x ^ 2 + 2 < 20<, xD
1 1
Out[10]= -3 2 < x < -p 2 ArcTanB -4 - 7 F < x < 2 ArcTanB -4 + 7 F
3 3
p p 1
0<x< < x < p 2 p + 2 ArcTanB -4 - 7 F<x<3 2
2 2 3

If you have inequalities that involve <= as well as <, there may be isolated points where the
inequalities can be satisfied. Reduce represents such points by giving equations.
Mathematics and Algorithms 127

This inequality can be satisfied at just two isolated points.

In[11]:= Reduce@Hx ^ 2 - 3 x + 1L ^ 2 <= 0, xD
1 1
Out[11]= x 3- 5 x 3+ 5
2 2

This yields both intervals and isolated points.

In[12]:= Reduce@8Max@Sin@2 xD, Cos@3 xDD <= 0, 0 < x < 10<, xD
p 5p 3p 11 p 5p 17 p
Out[12]= x x p x x x3p
2 6 2 6 2 6

ReduceA8ineq1 ,ineq2 ,<, reduce a collection of inequalities in several variables

9x1 , x2 , }]

Multivariate inequalities.

For inequalities involving several variables, Reduce in effect yields nested collections of interval
specifications, in which later variables have bounds that depend on earlier variables.

This represents the unit disk as nested inequalities for x and y.

In[13]:= Reduce@x ^ 2 + y ^ 2 < 1, 8x, y<D

Out[13]= -1 < x < 1 && - 1 - x2 < y < 1 - x2

In geometrical terms, any linear inequality divides space into two halves. Lists of linear inequali-
ties thus define polyhedra, sometimes bounded, sometimes not. Reduce represents such polyhe-
dra in terms of nested inequalities. The corners of the polyhedra always appear among the
endpoints of these inequalities.

This defines a triangular region in the plane.

In[14]:= Reduce@8x > 0, y > 0, x + y < 1<, 8x, y<D
Out[14]= 0 < x < 1 && 0 < y < 1 - x

Even a single triangle may need to be described as two components.

In[15]:= Reduce@8x > y - 1, y > 0, x + y < 1<, 8x, y<D
Out[15]= H-1 < x 0 && 0 < y < 1 + xL H0 < x < 1 && 0 < y < 1 - xL
128 Mathematics and Algorithms

Lists of inequalities in general represent regions of overlap between geometrical objects. Often
the description of these can be quite complicated.

This represents the part of the unit disk on one side of a line.
In[16]:= Reduce@8x ^ 2 + y ^ 2 < 1, x + 3 y > 2<, 8x, y<D
1 1 2-x
Out[16]= 2-3 6 <x< 2+3 6 && <y< 1 - x2
10 10 3

Here is the intersection between two disks.

In[17]:= Reduce@8Hx - 1L ^ 2 + y ^ 2 < 2, x ^ 2 + y ^ 2 < 2<, 8x, y<D
1 1
Out[17]= 1- 2 <x && - 1 + 2 x - x2 < y < 1 + 2 x - x2 <x< 2 && - 2 - x2 < y < 2 - x2
2 2

If the disks are too far apart, there is no intersection.

In[18]:= Reduce@8Hx - 4L ^ 2 + y ^ 2 < 2, x ^ 2 + y ^ 2 < 2<, 8x, y<D
Out[18]= False

Here is an example involving a transcendental inequality.

In[19]:= Reduce@8Sin@x yD > 1 2, x ^ 2 + y ^ 2 < 3 2<, 8x, y<D

3 1 1 1 3 - 2 x2 p
Out[19]= - + 81 - 4 p2 <x<- 9- 81 - 4 p2 && - <y<
4 12 2 3 2 6x

1 1 3 1 p 3 - 2 x2
9- 81 - 4 p2 <x< + 81 - 4 p2 && <y<
2 3 4 12 6x 2

If you have inequalities that involve parameters, Reduce automatically handles the different
cases that can occur, just as it does for equations.

The form of the intervals depends on the value of a.

In[20]:= Reduce@Hx - 1L Hx - aL > 0, xD
Out[20]= Ha 1 && Hx < a x > 1LL Ha > 1 && Hx < 1 x > aLL
Mathematics and Algorithms 129

One gets a hyperbolic or an elliptical region, depending on the value of a.

In[21]:= Reduce@x ^ 2 + a y ^ 2 < 1, 8x, y<D

1 - x2 1 - x2
Out[21]= y Reals && a < 0 && x -1 && y < - y >
a a

1 - x2 1 - x2
-1 < x < 1 x 1 && y < - y >
a a

1 - x2 1 - x2
Ha 0 && -1 < x < 1L a > 0 && -1 < x < 1 && - <y<
a a

Reduce tries to give you a complete description of the region defined by a set of inequalities.
Sometimes, however, you may just want to find individual instances of values of variables that
satisfy the inequalities. You can do this using FindInstance.

FindInstance@ineqs,8x1 ,x2 ,<D try to find an instance of the xi satisfying ineqs

FindInstance@ineqs,vars,nD try to find n instances

Finding individual points that satisfy inequalities.

This finds a specific instance that satisfies the inequalities.

In[22]:= FindInstance@8Sin@x yD > 1 2, x ^ 2 + y ^ 2 < 3 2<, 8x, y<D
88 543
Out[22]= ::x - , y- >>
151 566

This shows that there is no way to satisfy the inequalities.

In[23]:= FindInstance@8Sin@x yD > 1 2, x ^ 2 + y ^ 2 < 1 4<, 8x, y<D
Out[23]= 8<

FindInstance is in some ways an analog for inequalities of Solve for equations. For like Solve,
it returns a list of rules giving specific values for variables. But while for equations these values
can generically give an accurate representation of all solutions, for inequalities they can only
correspond to isolated sample points within the regions described by the inequalities.

Every time you call FindInstance with specific input, it will give the same output. And when
there are instances that correspond to special, limiting, points of some kind, it will preferentially
return these. But in general, the distribution of instances returned by FindInstance will typi-
Every time you call FindInstance with specific input, it will give the same output. And when
there are instances that correspond to special, limiting, points of some kind, it will preferentially
130 Mathematics and Algorithms

cally seem somewhat random. Each instance is, however, in effect a constructive proof that the
inequalities you have given can in fact be satisfied.

If you ask for one point in the unit disk, FindInstance gives the origin.
In[24]:= FindInstance@x ^ 2 + y ^ 2 <= 1, 8x, y<D
Out[24]= 88x 0, y 0<<

This finds 500 points in the unit disk.

In[25]:= FindInstance@x ^ 2 + y ^ 2 <= 1, 8x, y<, 500D;

Their distribution seems somewhat random.

In[26]:= ListPlot@8x, y< . %, AspectRatio -> AutomaticD
1.0

0.5

Out[26]=
1.0 0.5 0.5 1.0

0.5

1.0

Equations and Inequalities over Domains

Mathematica normally assumes that variables which appear in equations can stand for arbitrary
complex numbers. But when you use Reduce, you can explicitly tell Mathematica that the vari-
ables stand for objects in more restricted domains.

Reduce@expr,vars,domD reduce eqns over the domain dom

Complexes complex numbers

Reals real numbers

Integers integers

Solving over domains.

Reduce by default assumes that x can be complex, and gives all five complex solutions.
In[1]:= Reduce@x ^ 6 - x ^ 4 - 4 x ^ 2 + 4 == 0, xD

Out[1]= x -1 x 1 x - 2 x - 2 x 2 x 2

But here it assumes that x is real, and gives only the real solutions.
Mathematics and Algorithms 131

But here it assumes that x is real, and gives only the real solutions.
In[2]:= Reduce@x ^ 6 - x ^ 4 - 4 x ^ 2 + 4 == 0, x, RealsD

Out[2]= x -1 x 1 x - 2 x 2

And here it assumes that x is an integer, and gives only the integer solutions.
In[3]:= Reduce@x ^ 6 - x ^ 4 - 4 x ^ 2 + 4 == 0, x, IntegersD
Out[3]= x -1 x 1

A single polynomial equation in one variable will always have a finite set of discrete solutions.
And in such a case one can think of Reduce@eqns, vars, domD as just filtering the solutions by
selecting the ones that happen to lie in the domain dom.

But as soon as there are more variables, things can become more complicated, with solutions
to equations corresponding to parametric curves or surfaces in which the values of some vari-
ables can depend on the values of others. Often this dependence can be described by some
collection of equations or inequalities, but the form of these can change significantly when one
goes from one domain to another.

This gives solutions over the complex numbers as simple formulas.

In[4]:= Reduce@x ^ 2 + y ^ 2 == 1, 8x, y<D

Out[4]= y - 1 - x2 y 1 - x2

To represent solutions over the reals requires introducing an inequality.

In[5]:= Reduce@x ^ 2 + y ^ 2 == 1, 8x, y<, RealsD

Out[5]= -1 x 1 && y - 1 - x2 y 1 - x2

Over the integers, the solution can be represented as equations for discrete points.
In[6]:= Reduce@x ^ 2 + y ^ 2 == 1, 8x, y<, IntegersD
Out[6]= Hx -1 && y 0L Hx 0 && y -1L Hx 0 && y 1L Hx 1 && y 0L

If your input involves only equations, then Reduce will by default assume that all variables are
complex. But if your input involves inequalities, then Reduce will assume that any algebraic
variables appearing in them are real, since inequalities can only compare real quantities.
132 Mathematics and Algorithms

Since the variables appear in an inequality, they are assumed to be real.

In[7]:= Reduce@8x + y + z == 1, x ^ 2 + y ^ 2 + z ^ 2 < 1<, 8x, y, z<D
1 1-x 1 1-x 1
Out[7]= - < x < 1 && - 1 + 2 x - 3 x2 < y < + 1 + 2 x - 3 x2 && z 1 - x - y
3 2 2 2 2

Complexes polynomial != 0, xi == Root@D

Reals Root@D < xi < Root@D, xi == Root@D
Integers arbitrarily complicated

Schematic building blocks for solutions to polynomial equations and inequalities.

For systems of polynomials over real and complex domains, the solutions always consist of a
finite number of components, within which the values of variables are given by algebraic num-
bers or functions.

Here the components are distinguished by equations and inequations on x.

In[8]:= Reduce@x y ^ 3 + y == 1, 8x, y<, ComplexesD
Out[8]= Hx 0 && y 1L
Ix 0 && Iy RootA-1 + 1 + x 13 &, 1E y RootA-1 + 1 + x 13 &, 2E y RootA-1 + 1 + x 13 &, 3EMM

And here the components are distinguished by inequalities on x.

In[9]:= Reduce@x y ^ 3 + y == 1, 8x, y<, RealsD
4 4 3 4
Out[9]= x<- && y RootA-1 + 1 + x 13 &, 1E x - && y -3 y - < x < 0 &&
27 27 2 27
Iy RootA-1 + 1 + x 13 &, 1E y RootA-1 + 1 + x 13 &, 2E y RootA-1 + 1 + x 13 &, 3EM

Ix 0 && y RootA-1 + 1 + x 13 &, 1EM

While in principle Reduce can always find the complete solution to any collection of polynomial
equations and inequalities with real or complex variables, the results are often very compli-
cated, with the number of components typically growing exponentially as the number of vari-
ables increases.

With 3 variables, the solution here already involves 8 components.

In[10]:= Reduce@x ^ 2 == y ^ 2 == z ^ 2 == 1, 8x, y, z<D
Out[10]= Hz -1 && y -1 && x -1L Hz -1 && y -1 && x 1L
Hz -1 && y 1 && x -1L Hz -1 && y 1 && x 1L Hz 1 && y -1 && x -1L
Hz 1 && y -1 && x 1L Hz 1 && y 1 && x -1L Hz 1 && y 1 && x 1L

As soon as one introduces functions like Sin or Exp, even equations in single real or complex
variables can have solutions with an infinite number of components. Reduce labels these compo-
nents by introducing additional parameters. By default, the nth parameter in a given solution
will be named C@nD. In general you can specify that it should be named f @nD by giving the
Mathematics and Algorithms 133

The components here are labeled by the integer parameter c1 .

In[11]:= Reduce@Exp@xD == 2, x, GeneratedParameters -> HSubscript@c, D &LD
Out[11]= c1 Integers && x Log@2D + 2 p c1

Reduce can handle equations not only over real and complex variables, but also over integers.
Solving such Diophantine equations can often be a very difficult problem.

Describing the solution to this equation over the reals is straightforward.

In[12]:= Reduce@x y == 8, 8x, y<, RealsD
8
Out[12]= Hx < 0 x > 0L && y
x

The solution over the integers involves the divisors of 8.

In[13]:= Reduce@x y == 8, 8x, y<, IntegersD
Out[13]= Hx -8 && y -1L Hx -4 && y -2L Hx -2 && y -4L Hx -1 && y -8L
Hx 1 && y 8L Hx 2 && y 4L Hx 4 && y 2L Hx 8 && y 1L

Solving an equation like this effectively requires factoring a large number.

In[14]:= Reduce@8x y == 7 777 777, x > y > 0<, 8x, y<, IntegersD
Out[14]= Hx 4649 && y 1673L Hx 32 543 && y 239L Hx 1 111 111 && y 7L Hx 7 777 777 && y 1L

Reduce can solve any system of linear equations or inequalities over the integers. With m linear
equations in n variables, n - m parameters typically need to be introduced. But with inequalities,
a much larger number of parameters may be needed.

Three parameters are needed here, even though there are only two variables.
In[15]:= Reduce@83 x - 2 y > 1, x > 0, y > 0<, 8x, y<, IntegersD
Out[15]= HC@1D C@2D C@3DL Integers && C@1D 0 && C@2D 0 &&
C@3D 0 && HHx 2 + 2 C@1D + C@2D + C@3D && y 2 + 3 C@1D + C@2DL
Hx 2 + 2 C@1D + C@2D + C@3D && y 1 + 3 C@1D + C@2DLL

With two variables, Reduce can solve any quadratic equation over the integers. The result can
be a Fibonacci-like sequence, represented in terms of powers of quadratic irrationals.

Here is the solution to a Pell equation.

134 Mathematics and Algorithms

Here is the solution to a Pell equation.

In[16]:= Reduce@8x ^ 2 == 13 y ^ 2 + 1, x > 0, y > 0<, 8x, y<, IntegersD
1 C@1D C@1D
Out[16]= C@1D Integers && C@1D 1 && x 649 - 180 13 + 649 + 180 13 &&
2
C@1D C@1D
J649 - 180 13 N - J649 + 180 13 N
y-
2 13

The actual values for specific C@1D as integers, as they should be.
In[17]:= FullSimplify@% . Table@8C@1D -> i<, 8i, 4<DD
Out[17]= 8x 649 && y 180, x 842 401 && y 233 640,
x 1 093 435 849 && y 303 264 540, x 1 419 278 889 601 && y 393 637 139 280<

Reduce can handle many specific classes of equations over the integers.

Here Reduce finds the solution to a Thue equation.

In[18]:= Reduce@x ^ 3 - 4 x y ^ 2 + y ^ 3 == 1, 8x, y<, IntegersD
Out[18]= Hx -2 && y 1L Hx 0 && y 1L Hx 1 && y 0L
Hx 1 && y 4L Hx 2 && y 1L Hx 508 && y 273L

Changing the right-hand side to 3, the equation now has no solution.

In[19]:= Reduce@x ^ 3 - 4 x y ^ 2 + y ^ 3 == 3, 8x, y<, IntegersD
Out[19]= False

Equations over the integers sometimes have seemingly quite random collections of solutions.
And even small changes in equations can often lead them to have no solutions at all.

For polynomial equations over real and complex numbers, there is a definite decision procedure
for determining whether or not any solution exists. But for polynomial equations over the inte-
gers, the unsolvability of Hilbert's tenth problem demonstrates that there can never be any
such general procedure.

For specific classes of equations, however, procedures can be found, and indeed many are
implemented in Reduce. But handling different classes of equations can often seem to require
whole different branches of number theory, and quite different kinds of computations. And in
fact it is known that there are universal integer polynomial equations, for which filling in some
variables can make solutions for other variables correspond to the output of absolutely any
possible program. This then means that for such equations there can never in general be any
closed-form solution built from fixed elements like algebraic functions.

If one includes functions like Sin, then even for equations involving real and complex numbers
the same issues can arise.
Mathematics and Algorithms 135

If one includes functions like Sin, then even for equations involving real and complex numbers
the same issues can arise.

Reduce here effectively has to solve an equation over the integers.

In[20]:= Reduce@Sin@Pi xD ^ 2 + Sin@Pi yD ^ 2 + Hx ^ 2 + y ^ 2 - 25L ^ 2 == 0, 8x, y<, RealsD
Out[20]= Hx -5 && y 0L Hx -4 && Hy -3 y 3LL Hx -3 && Hy -4 y 4LL
Hx 0 && Hy -5 y 5LL Hx 3 && Hy -4 y 4LL Hx 4 && Hy -3 y 3LL Hx 5 && y 0L

ReduceAeqns,vars,Modulus->nE find solutions modulo n

Handling equations involving integers modulo n.

Since there are only ever a finite number of possible solutions for integer equations modulo n,
Reduce can systematically find them.

This finds all solutions modulo 4.

In[21]:= Reduce@x ^ 5 == y ^ 4 + x y + 1, 8x, y<, Modulus -> 4D
Out[21]= Hx 1 && y 0L Hx 1 && y 3L Hx 2 && y 1L Hx 2 && y 3L Hx 3 && y 2L Hx 3 && y 3L

Reduce can also handle equations that involve several different moduli.

Here is an equation involving two different moduli.

In[22]:= Reduce@Mod@2 x + 1, 5D == Mod@x, 7D && 0 < x < 50, xD
Out[22]= x 4 x 7 x 15 x 23 x 31 x 39 x 42

Reduce@expr,vars,domD specify a default domain for all variables

Reduce@8expr1 ,,x1 dom1 ,<,varsD explicitly specify individual domains for variables

Different ways to specify domains for variables.

This assumes that x is an integer, but y is a real.

In[23]:= Reduce@8x ^ 2 + 2 y ^ 2 == 1, x Integers, y Reals<, 8x, y<D
1 1
Out[23]= Hx -1 && y 0L x 0 && y - y Hx 1 && y 0L
2 2

Reduce normally treats complex variables as single objects. But in dealing with functions that
are not analytic or have branch cuts, it sometimes has to break them into pairs of real variables
Re@zD and Im@zD.
136 Mathematics and Algorithms

The result involves separate real and imaginary parts.

In[24]:= Reduce@Abs@zD == 1, zD

Out[24]= -1 Re@zD 1 && Im@zD - 1 - Re@zD2 Im@zD 1 - Re@zD2

Here again there is a separate condition on the imaginary part.

In[25]:= Reduce@Log@zD == a, 8a, z<D
a
Out[25]= -p < Im@aD p && z

Reduce by default assumes that variables that appear algebraically in inequalities are real. But
you can override this by explicitly specifying Complexes as the default domain. It is often useful
in such cases to be able to specify that certain variables are still real.

Reduce by default assumes that x is a real.

In[26]:= Reduce@x ^ 2 < 1, xD
Out[26]= -1 < x < 1

This forces Reduce to consider the case where x can be complex.

In[27]:= Reduce@x ^ 2 < 1, x, ComplexesD
Out[27]= H-1 < Re@xD < 0 && Im@xD 0L Re@xD 0 H0 < Re@xD < 1 && Im@xD 0L

Since x does not appear algebraically, Reduce immediately assumes that it can be complex.
In[28]:= Reduce@Abs@xD < 1, xD

Out[28]= -1 < Re@xD < 1 && - 1 - Re@xD2 < Im@xD < 1 - Re@xD2

Here x is a real, but y can be complex.

In[29]:= Reduce@8Abs@yD < Abs@xD, x Reals<, 8x, y<D

Out[29]= x < 0 && - x2 < Re@yD < x2 && - x2 - Re@yD2 < Im@yD < x2 - Re@yD2

x > 0 && - x2 < Re@yD < x2 && - x2 - Re@yD2 < Im@yD < x2 - Re@yD2

FindInstance@expr,8x1 ,x2 ,<,domD try to find an instance of the xi in dom satisfying expr
FindInstance@expr,vars,dom,nD try to find n instances
Complexes the domain of complex numbers
Mathematics and Algorithms 137

Reals the domain of real numbers

Integers the domain of integers

Booleans the domain of Booleans (True and False )

Finding particular solutions in domains.

Reduce always returns a complete representation of the solution to a system of equations or

inequalities. Sometimes, however, you may just want to find particular sample solutions. You
can do this using FindInstance.

If FindInstance@expr, vars, domD returns 8< then this means that Mathematica has effectively
proved that expr cannot be satisfied for any values of variables in the specified domain. When
expr can be satisfied, FindInstance will normally pick quite arbitrarily among values that do
this, as discussed for inequalities in "Inequalities: Manipulating Equations and Inequalities".

Particularly for integer equations, FindInstance can often find particular solutions to equations
even when Reduce cannot find a complete solution. In such cases it usually returns one of the
smallest solutions to the equations.

This finds the smallest integer point on an elliptic curve.

In[30]:= FindInstance@8x ^ 2 == y ^ 3 + 12, x > 0, y > 0<, 8x, y<, IntegersD
Out[30]= 88x 47, y 13<<

One feature of FindInstance is that it also works with Boolean expressions whose variables
can have values True or False. You can use FindInstance to determine whether a particular
expression is satisfiable, so that there is some choice of truth values for its variables that
makes the expression True.

This expression cannot be satisfied for any choice of p and q.

In[31]:= FindInstance@p && ! Hp ! qL, 8p, q<, BooleansD
Out[31]= 8<

But this can.

In[32]:= FindInstance@p && ! H! p ! qL, 8p, q<, BooleansD
Out[32]= 88p True, q True<<
138 Mathematics and Algorithms

The Representation of Solution Sets

Any combination of equations or inequalities can be thought of as implicitly defining a region in
some kind of space. The fundamental function of Reduce is to turn this type of implicit descrip-
tion into an explicit one.

An implicit description in terms of equations or inequalities is sufficient if one just wants to test
whether a point specified by values of variables is in the region. But to understand the structure
of the region, or to generate points in it, one typically needs a more explicit description, of the
kind obtained from Reduce.

Here are inequalities that implicitly define a semicircular region.

In[1]:= semi = x > 0 && x ^ 2 + y ^ 2 < 1

Out[1]= x > 0 && x2 + y2 < 1

This shows that the point H1 2, 1 2L lies in the region.

In[2]:= semi . 8x -> 1 2, y -> 1 2<
Out[2]= True

Reduce gives a more explicit representation of the region.

In[3]:= Reduce@semi, 8x, y<D

Out[3]= 0 < x < 1 && - 1 - x2 < y < 1 - x2

If we pick a value for x consistent with the first inequality, we then immediately get an explicit
inequality for y.
In[4]:= % . x -> 1 2

3 3
Out[4]= - <y<
2 2

Reduce@expr, 8x1 , x2 , <D is set up to describe regions by first giving fixed conditions for x1 ,
then giving conditions for x2 that depend on x1 , then conditions for x3 that depend on x1 and x2 ,
and so on. This structure has the feature that it allows one to pick points by successively choos-
ing values for each of the xi in turn~in much the same way as when one uses iterators in func-
tions like Table.
Mathematics and Algorithms 139

This gives a representation for the region in which one first picks a value for y, then x.
In[5]:= Reduce@semi, 8y, x<D

Out[5]= -1 < y < 1 && 0 < x < 1 - y2

In some simple cases the region defined by a system of equations or inequalities will end up
having only one component. In such cases, the output from Reduce will be of the form
e1 && e2 where each of the ei is an equation or inequality involving variables up to xi .

In most cases, however, there will be several components, represented by output containing
forms such as u1 u2 . Reduce typically tries to minimize the number of components used
in describing a region. But in some cases multiple parametrizations may be needed to cover a
single connected component, and each one of these will appear as a separate component in the
output from Reduce.

In representing solution sets, it is common to find that several components can be described
together by using forms such as && Hu1 u2 L && . Reduce by default does this so as to
return its results as compactly as possible. You can use LogicalExpand to generate an
expanded form in which each component appears separately.

In generating the most compact results, Reduce sometimes ends up making conditions on later
variables xi depend on more of the earlier xi than is strictly necessary. You can force Reduce to
generate results in which a particular xi only has minimal dependence on earlier xi by giving the
option Backsubstitution -> True. Usually this will lead to much larger output, although some-
times it may be easier to interpret.

By default, Reduce expresses the condition on y in terms of x.

In[6]:= Reduce@x ^ 2 + y == 4 && x ^ 3 - 4 y == 8, 8x, y<D

Out[6]= x 2 x -3 - 3 x -3 + 3 && y 4 - x2

Backsubstituting allows conditions for y to be given without involving x.

In[7]:= Reduce@x ^ 2 + y == 4 && x ^ 3 - 4 y == 8, 8x, y<, Backsubstitution -> TrueD

Out[7]= Hx 2 && y 0L x - -3 + 3 && y -2 - + 3 3 x 3 + 3 && y 2 + 3 3

140 Mathematics and Algorithms

CylindricalDecomposition@expr,8x1 ,x2 ,<D

generate the cylindrical algebraic decomposition of the
region defined by expr
GenericCylindricalDecomposition@expr,8x1 ,x2 ,<D
find the full-dimensional part of the decomposition of the
region defined by expr, together with any hypersurfaces
containing the rest of the region
SemialgebraicComponentInstances@expr,8x1 ,x2 ,<D
give at least one point in each connected component of the
region defined by expr

Cylindrical algebraic decomposition.

For polynomial equations or inequalities over the reals, the structure of the result returned by
Reduce is typically a cylindrical algebraic decomposition or CAD. Sometimes Reduce can yield a
simpler form. But in all cases you can get the complete CAD by using
CylindricalDecomposition. For systems containing inequalities only,
GenericCylindricalDecomposition gives you "most" of the solution set and is often faster.

Here is the cylindrical algebraic decomposition of a region bounded by a hyperbola.

In[8]:= CylindricalDecomposition@x ^ 2 - y ^ 2 >= 1, 8x, y<D

Out[8]= x < -1 && - -1 + x2 y -1 + x2 Hx -1 && y 0L

Hx 1 && y 0L x > 1 && - -1 + x2 y -1 + x2

This gives the two-dimensional part of the solution set along with a curve containing the
boundary.
In[9]:= GenericCylindricalDecomposition@x ^ 2 - y ^ 2 >= 1, 8x, y<D

Out[9]= : x < -1 && - -1 + x2 < y < -1 + x2 x > 1 && - -1 + x2 < y < -1 + x2 , 1 - x2 + y2 0>

This finds solutions from both parts of the solution set.

In[10]:= SemialgebraicComponentInstances@x ^ 2 - y ^ 2 >= 1, 8x, y<D

Out[10]= :8x -2, y -1<, 8x -2, y 1<, 8x -1, y 0<, 8x 1, y 0<,

8x 2, y -1<, 8x 2, y 1<, :x - 2 , y -1>, :x - 2 , y 0>,

:x - 2 , y 1>, :x 2 , y -1>, :x 2 , y 0>, :x 2 , y 1>>

Mathematics and Algorithms 141

The results include a few points from each piece of the solution set.
In[11]:= Show@
8RegionPlot@x ^ 2 - y ^ 2 >= 1, 8x, - 3, 3<, 8y, - 3, 3<D, Graphics@Point@8x, y<D . %D<D
3

Out[11]=
0

-1

-2

-3
-3 -2 -1 0 1 2 3

Quantifiers
In a statement like x ^ 4 + x ^ 2 > 0, Mathematica treats the variable x as having a definite,
though unspecified, value. Sometimes, however, it is useful to be able to make statements
about whole collections of possible values for x. You can do this using quantifiers.

ForAll@x,exprD expr holds for all values of x

ForAll@8x1 ,x2 ,<,exprD expr holds for all values of all the xi
ForAll@8x1 ,x2 ,<,cond,exprD expr holds for all xi satisfying cond
Exists@x,exprD there exists a value of x for which expr holds
Exists@8x1 ,x2 ,<,exprD there exist values of the xi for which expr holds

Exists@8x1 ,<,cond,exprD there exist values of the xi satisfying cond for which expr
holds

The structure of quantifiers.

You can work with quantifiers in Mathematica much as you work with equations, inequalities or
logical connectives. In most cases, the quantifiers will not immediately be changed by evalua-
tion. But they can be simplified or reduced by functions like FullSimplify and Reduce.

This asserts that an x exists that makes the inequality true. The output here is just a formatted
version of the input.
In[1]:= Exists@x, x ^ 4 + x ^ 2 > 0D

Out[1]= $x x2 + x4 > 0
142 Mathematics and Algorithms

FullSimplify establishes that the assertion is true.

In[2]:= FullSimplify@%D
Out[2]= True

This gives False , since the inequality fails when x is zero.

In[3]:= FullSimplify@ForAll@x, x ^ 4 + x ^ 2 > 0DD
Out[3]= False

Mathematica supports a version of the standard notation for quantifiers used in predicate logic
and pure mathematics. You can input " as @ForAllD or fa, and you can input $ as @ExistsD
or ex. To make the notation precise, however, Mathematica makes the quantified variable a
subscript. The conditions on the variable can also be given in the subscript, separated by a
comma.

"x expr ForAll@x,exprD

"8x1 ,x2 ,< expr ForAll@8x1 ,x2 ,<,exprD
"x,cond expr ForAll@x,cond,exprD
$x expr Exists@x,exprD
$8x1 ,x2 ,< expr Exists@8x1 ,x2 ,<,exprD
$x,cond expr Exists@x,cond,exprD

Notation for quantifiers.

Given a statement that involves quantifiers, there are certain important cases where it is possi-
ble to resolve it into an equivalent statement in which the quantifiers have been eliminated.
Somewhat like solving an equation, such quantifier elimination turns an implicit statement
about what is true for all x or for some x into an explicit statement about the conditions under
which this holds.

Resolve@exprD attempt to eliminate quantifiers from expr

Resolve@expr,domD attempt to eliminate quantifiers with all variables assumed
to be in domain dom

Quantifier elimination.
Mathematics and Algorithms 143

This shows that an x exists that makes the equation true.

In[4]:= Resolve@Exists@x, x ^ 2 == x ^ 3DD
Out[4]= True

This shows that the equations can only be satisfied if c obeys a certain condition.
In[5]:= Resolve@Exists@x, x ^ 2 == c && x ^ 3 == c + 1DD

Out[5]= -1 - 2 c - c2 + c3 0

Resolve can always eliminate quantifiers from any collection of polynomial equations and
inequations over complex numbers, and from any collection of polynomial equations and inequal -
ities over real numbers. It can also eliminate quantifiers from Boolean expressions.

This finds the conditions for a quadratic form over the reals to be positive.
In[6]:= Resolve@ForAll@x, a x ^ 2 + b x + c > 0D, RealsD

Out[6]= Ia > 0 && -a b2 + 4 a2 c > 0M Ha 0 && b 0 && c > 0L Ia 0 && b 0 && c > 0 && -a b2 + 4 a2 c > 0M

This shows that there is a way of assigning truth values to p and q that makes the expression
true.
In[7]:= Resolve@Exists@8p, q<, p q && ! qD, BooleansD
Out[7]= True

You can also use quantifiers with Reduce. If you give Reduce a collection of equations or inequali-
ties, then it will try to produce a detailed representation of the complete solution set. But some-
times you may want to address a more global question, such as whether the solution set covers
all values of x, or whether it covers none of these values. Quantifiers provide a convenient way
to specify such questions.

This gives the complete structure of the solution set.

In[8]:= Reduce@x ^ 2 + x + c == 0, 8c, x<, RealsD
1 1 1 1 1 1 1
Out[8]= c< && x - - 1 - 4 c x - + 1-4c c && x -
4 2 2 2 2 4 2

This instead just gives the condition for a solution to exist.

In[9]:= Reduce@Exists@x, x ^ 2 + x + c == 0D, 8c<, RealsD
1
Out[9]= c
4

It is possible to formulate a great many mathematical questions in terms of quantifiers.

144 Mathematics and Algorithms

It is possible to formulate a great many mathematical questions in terms of quantifiers.

This finds the conditions for a circle to be contained within an arbitrary conic section.
In[10]:= Reduce@ForAll@8x, y<, x ^ 2 + y ^ 2 < 1, a x ^ 2 + b y ^ 2 < cD, 8a, b, c<, RealsD
Out[10]= Ha 0 && HHb 0 && c > 0L Hb > 0 && c bLLL Ha > 0 && HHb < a && c aL Hb a && c bLLL

This finds the conditions for a line to intersect a circle.

In[11]:= Reduce@Exists@8x, y<, x ^ 2 + y ^ 2 < 1, r x + s y == 1D, 8r, s<, RealsD

Out[11]= r < -1 -1 r 1 && s < - 1 - r2 s > 1 - r2 r > 1

This defines q to be a general monic quartic.

In[12]:= q@x_D := x ^ 4 + b x ^ 3 + c x ^ 2 + d x + e

This finds the condition for all pairs of roots to the quartic to be equal.
In[13]:= Reduce@ForAll@8x, y<, q@xD == 0 && q@yD == 0, x == yD, 8b, c, d, e<D
3 b2 b3 b4
Out[13]= c && d && e Hb 0 && c 0 && d 0 && e 0L
8 16 256

Although quantifier elimination over the integers is in general a computationally impossible

problem, Mathematica can do it in specific cases.

This shows that 2 cannot be a rational number.

In[14]:= Resolve@Exists@8x, y<, x ^ 2 == 2 y ^ 2 && y > 0D, IntegersD
Out[14]= False

9 4 is, though.
In[15]:= Resolve@Exists@8x, y<, 4 x ^ 2 == 9 y ^ 2 && y > 0D, IntegersD
Out[15]= True
Mathematics and Algorithms 145

Minimization and Maximization

Minimize@expr,8x1 ,x2 ,<D minimize expr

Minimize@8expr,cons<,8x1 ,x2 ,<D minimize expr subject to the constraints cons
Maximize@expr,8x1 ,x2 ,<D maximize expr
Maximize@8expr,cons<,8x1 ,x2 ,<D maximize expr subject to the constraints cons

Minimization and maximization.

Minimize and Maximize yield lists giving the value attained at the minimum or maximum,
together with rules specifying where the minimum or maximum occurs.

This finds the minimum of a quadratic function.

In[1]:= Minimize@x ^ 2 - 3 x + 6, xD
15 3
Out[1]= : , :x >>
4 2

Applying the rule for x gives the value at the minimum.

In[2]:= x ^ 2 - 3 x + 6 . Last@%D
15
Out[2]=
4

This maximizes with respect to x and y.

In[3]:= Maximize@5 x y - x ^ 4 - y ^ 4, 8x, y<D

25 5 5
Out[3]= : , :x - , y- >>
8 2 2

Minimize@expr, xD minimizes expr allowing x to range over all possible values from - to +.
Minimize@8expr, cons<, xD minimizes expr subject to the constraints cons being satisfied. The
constraints can consist of any combination of equations and inequalities.

This finds the minimum subject to the constraint x 3.

In[4]:= Minimize@8x ^ 2 - 3 x + 6, x >= 3<, xD
Out[4]= 86, 8x 3<<
146 Mathematics and Algorithms

This finds the maximum within the unit circle.

In[5]:= Maximize@85 x y - x ^ 4 - y ^ 4, x ^ 2 + y ^ 2 <= 1<, 8x, y<D
1 1
Out[5]= :2, :x - , y- >>
2 2

This finds the maximum within an ellipse. The result is fairly complicated.
In[6]:= Maximize@85 x y - x ^ 4 - y ^ 4, x ^ 2 + 2 y ^ 2 <= 1<, 8x, y<D

Out[6]= 9-RootA-811 219 + 320 160 1 + 274 624 12 - 170 240 13 + 25 600 14 &, 1E,
9x RootA25 - 102 12 + 122 14 - 70 16 + 50 18 &, 2E,
y RootA25 - 264 12 + 848 14 - 1040 16 + 800 18 &, 1E==

This finds the maximum along a line.

In[7]:= Maximize@85 x y - x ^ 4 - y ^ 4, x + y == 1<, 8x, y<D
9 1 1
Out[7]= : , :x , y >>
8 2 2

Minimize and Maximize can solve any linear programming problem in which both the objective
function expr and the constraints cons involve the variables xi only linearly.

Here is a typical linear programming problem.

In[8]:= Minimize@8x + 3 y, x - 3 y <= 7 && x + 2 y >= 10<, 8x, y<D
53 44 3
Out[8]= : , :x , y >>
5 5 5

They can also in principle solve any polynomial programming problem in which the objective
function and the constraints involve arbitrary polynomial functions of the variables. There are
many important geometrical and other problems that can be formulated in this way.

This solves the simple geometrical problem of maximizing the area of a rectangle with fixed
perimeter.
In[9]:= Maximize@8x y, x + y == 1<, 8x, y<D
1 1 1
Out[9]= : , :x , y >>
4 2 2

This finds the maximal volume of a cuboid that fits inside the unit sphere.
In[10]:= Maximize@88 x y z, x ^ 2 + y ^ 2 + z ^ 2 <= 1<, 8x, y, z<D
8 1 1 1
Out[10]= : , :x , y- , z- >>
3 3 3 3 3

An important feature of Minimize and Maximize is that they always find global minima and
maxima. Often functions will have various local minima and maxima at which derivatives van-
Mathematics and Algorithms 147

Here is a function with many local maxima and minima.

In[11]:= Plot@x + 2 Sin@xD, 8x, - 10, 10<D
10

Out[11]=
-10 -5 5 10

-5

-10

Maximize finds the global maximum.

In[12]:= Maximize@8x + 2 Sin@xD, - 10 <= x <= 10<, xD
8p 8p
Out[12]= : 3 + , :x >>
3 3

If you give functions that are unbounded, Minimize and Maximize will return - and + as the
minima and maxima. And if you give constraints that can never be satisfied, they will return +
and - as the minima and maxima, and Indeterminate as the values of variables.

One subtle issue is that Minimize and Maximize allow both nonstrict inequalities of the form
x <= v, and strict ones of the form x < v. With nonstrict inequalities there is no problem with a
minimum or maximum lying exactly on the boundary x -> v. But with strict inequalities, a mini-
mum or maximum must in principle be at least infinitesimally inside the boundary.

With a strict inequality, Mathematica prints a warning, then returns the point on the boundary.
In[13]:= Minimize@8x ^ 2 - 3 x + 6, x > 3<, xD

Minimize::wksol : Warning: There is no minimum in the

region described by the contraints; returning a result on the boundary.
Out[13]= 86, 8x 3<<

Minimize and Maximize normally assume that all variables you give are real. But by giving a
constraint such as x Integers you can specify that a variable must in fact be an integer.
148 Mathematics and Algorithms

This does maximization only over integer values of x and y.

In[14]:= Maximize@8x y, x ^ 2 + y ^ 2 < 120 && Hx yL Integers<, 8x, y<D
Out[14]= 856, 8x -8, y -7<<

Minimize and Maximize can compute maxima and minima of linear functions over the integers
in bounded polyhedra. This is known as integer linear programming.

This does maximization over integer values of x and y in a triangle.

In[15]:= Maximize@85 + 3 y + 7 x, x >= 0 && y >= 0 && 3 x + 4 y <= 100 && Hx yL Integers<, 8x, y<D
Out[15]= 8236, 8x 33, y 0<<

Minimize and Maximize can produce exact symbolic results for polynomial optimization prob-
lems with parameters.

This finds the minimum of a general quadratic polynomial.

In[16]:= Minimize@a x ^ 2 + b x + c, xD

Out[16]=

MinValue@8 f ,cons<,8x,y,<D give the minimum value of f subject to the constraints cons
MaxValue@8 f ,cons<,8x,y,<D give the maximum value of f subject to the constraints cons
ArgMin@8 f ,cons<,8x,y,<D give a position at which f is minimized subject to the
constraints cons
ArgMax@8 f ,cons<,8x,y,<D give a position at which f is maximized subject to the
constraints cons

Computing values and positions of minima and maxima.

Maximize gives both the value and the position of a maximum.

In[17]:= Maximize@8x + 2 y, x ^ 2 + y ^ 2 1<, 8x, y<D
4 2
Out[17]= : 5 , :x - + 5 , y >>
5 5
Mathematics and Algorithms 149

Use MaxValue if you only need the maximum value.

In[18]:= MaxValue@8x + 2 y, x ^ 2 + y ^ 2 1<, 8x, y<D

Out[18]= 5

For strict polynomial inequality constraints computing only the maximum value may be much
faster.
In[19]:= TimeConstrained@
Maximize@8- x ^ 2 + 2 x y - z - 1, x ^ 2 y < z ^ 3 && x - z ^ 2 > y ^ 2 + 2<, 8x, y, z<D, 300D
Out[19]= $Aborted

In[20]:= MaxValue@8- x ^ 2 + 2 x y - z - 1, x ^ 2 y < z ^ 3 && x - z ^ 2 > y ^ 2 + 2<, 8x, y, z<D Timing

Out[20]= 90.312, -1 - RootA-6 674 484 057 677 824 + 27 190 416 613 703 680 1 -
9 845 871 213 297 967 104 12 + 30 310 812 947 042 320 384 13 - 38 968 344 650 849 575 680 14 +
27 943 648 095 748 511 616 15 - 13 622 697 129 083 140 957 16 + 5 905 344 357 450 294 480 17 -
2 872 859 681 127 251 424 18 + 1 484 592 492 626 145 792 19 - 567 863 224 101 551 360 110 +
100 879 538 475 737 088 111 + 303 891 741 605 888 112 - 2 224 545 911 472 128 113 +
70 301 735 976 960 114 + 25 686 756 556 800 115 + 1 786 706 395 136 116 + 73 014 444 032 117 &, 1E=

ArgMax gives a position at which the maximum value is attained.

In[21]:= ArgMax@8x + 2 y, x ^ 2 + y ^ 2 1<, 8x, y<D
4 2
Out[21]= :- + 5 , >
5 5

Linear Algebra

Constructing Matrices

Table@ f ,8i,m<,8 j,n<D build an mn matrix where f is a function of i and j that

gives the value of the i, jth entry

Array@ f ,8m,n<D build an mn matrix whose i, jth entry is f @i, jD

ConstantArray@a,8m,n<D build an mn matrix with all entries equal to a
DiagonalMatrix@listD generate a diagonal matrix with the elements of list on the
diagonal
IdentityMatrix@nD generate an nn identity matrix
Normal@SparseArray @88i1 , j1 <-> make a matrix with nonzero values vk at positions 8ik , jk <
v1 ,8i2 , j2 <->v2 ,<,8m,n<DD

Functions for constructing matrices.

This generates a 22 matrix whose i, jth entry is a@i, jD.

150 Mathematics and Algorithms

This generates a 22 matrix whose i, jth entry is a@i, jD.

In[1]:= Table@a@i, jD, 8i, 2<, 8j, 2<D
Out[1]= 88a@1, 1D, a@1, 2D<, 8a@2, 1D, a@2, 2D<<

Here is another way to produce the same matrix.

In[2]:= Array@a, 82, 2<D
Out[2]= 88a@1, 1D, a@1, 2D<, 8a@2, 1D, a@2, 2D<<

This creates a 32 matrix of zeros.

In[3]:= ConstantArray@0, 83, 2<D
Out[3]= 880, 0<, 80, 0<, 80, 0<<

DiagonalMatrix makes a matrix with zeros everywhere except on the leading diagonal.
In[4]:= DiagonalMatrix@8a, b, c<D
Out[4]= 88a, 0, 0<, 80, b, 0<, 80, 0, c<<

IdentityMatrix@nD produces an nn identity matrix.

In[5]:= IdentityMatrix@3D
Out[5]= 881, 0, 0<, 80, 1, 0<, 80, 0, 1<<

This makes a 34 matrix with two nonzero values filled in.

In[6]:= Normal@SparseArray@882, 3< -> a, 83, 2< -> b<, 83, 4<DD
Out[6]= 880, 0, 0, 0<, 80, 0, a, 0<, 80, b, 0, 0<<

MatrixForm prints the matrix in a two-dimensional form.

In[7]:= MatrixForm@%D
0 0 0 0
Out[7]//MatrixForm= 0 0 a 0
0 b 0 0

Table@0,8m<,8n<D a matrix of zeros

Table@If@i>= j,1,0D,8i,m<,8 j,n<D a lower-triangular matrix
RandomReal@80,1<,8m,n<D a matrix with random numerical entries

Constructing special types of matrices.

Table evaluates If@i j, a ++, 0D separately for each element, to give a matrix with
sequentially increasing entries in the lower-triangular part.
Mathematics and Algorithms 151

Table evaluates If@i j, a ++, 0D separately for each element, to give a matrix with
sequentially increasing entries in the lower-triangular part.
In[8]:= a = 1; Table@If@i j, a ++, 0D, 8i, 3<, 8j, 3<D
Out[8]= 881, 0, 0<, 82, 3, 0<, 84, 5, 6<<

SparseArray @8<,8n,n<D a zero matrix

SparseArray @8i _,i _<->1,8n,n<D an n n identity matrix
SparseArray @8i _, j a lower-triangular matrix
_< ; i >= j->1,8n,n<D

Constructing special types of matrices with SparseArray .

This sets up a general lower-triangular matrix.

In[9]:= SparseArray@8i_, j_< ; i >= j -> f@i, jD, 83, 3<D MatrixForm
f@1, 1D 0 0
Out[9]//MatrixForm= f@2, 1D f@2, 2D 0
f@3, 1D f@3, 2D f@3, 3D

Getting and Setting Pieces of Matrices

m@@i, jDD the i, jth entry

m@@iDD the ith row

mAAAll,iEE the ith column

Take@m,8i0 ,i1 <,8 j0 , j1 <D the submatrix with rows i0 through i1 and columns j0
through j1
m@@i0 ;;i1 , j0 ;; j1 DD the submatrix with rows i0 through i1 and columns j0
through j1
mAA9i1 ,,ir }, { j1 , , js }]] the rs submatrix with elements having row indices ik and
column indices jk

TrAm,ListE elements on the diagonal

ArrayRules@mD positions of nonzero elements

Ways to get pieces of matrices.

Matrices in Mathematica are represented as lists of lists. You can use all the standard Mathemat-
ica list-manipulation operations on matrices.
152 Mathematics and Algorithms

Here is a sample 33 matrix.

In[1]:= t = Array@a, 83, 3<D
Out[1]= 88a@1, 1D, a@1, 2D, a@1, 3D<, 8a@2, 1D, a@2, 2D, a@2, 3D<, 8a@3, 1D, a@3, 2D, a@3, 3D<<

This picks out the second row of the matrix.

In[2]:= t@@2DD
Out[2]= 8a@2, 1D, a@2, 2D, a@2, 3D<

Here is the second column of the matrix.

In[3]:= t@@All, 2DD
Out[3]= 8a@1, 2D, a@2, 2D, a@3, 2D<

This picks out a submatrix.

In[4]:= Take@t, 81, 2<, 82, 3<D
Out[4]= 88a@1, 2D, a@1, 3D<, 8a@2, 2D, a@2, 3D<<

m=88a11 ,a12 ,<,8a21 ,a22 ,<,< assign m to be a matrix

m@@i, jDD=a reset element 8i, j< to be a
m@@iDD=a reset all elements in row i to be a
m@@iDD=8a1 ,a2 ,< reset elements in row i to be 8a1 , a2 , <
m@@i0 ;;i1 DD=8v1 ,v2 ,< reset rows i0 through i1 to be vectors 8v1 , v2 , <

mAAAll, jEE=a reset all elements in column j to be a

mAAAll, jEE=8a1 ,a2 ,< reset elements in column j to be 8a1 , a2 , <

m@@i0 ;;i1 , j0 ;; j1 DD= reset the submatrix with rows i0 through i1 and columns j0
88a11 ,a12 ,<,8a21 ,a22 ,<,< through j1 to new values

Resetting parts of matrices.

Here is a 33 matrix.
In[5]:= m = 88a, b, c<, 8d, e, f<, 8g, h, i<<
Out[5]= 88a, b, c<, 8d, e, f<, 8g, h, i<<

This resets the 2, 2 element to be x, then shows the whole matrix.

In[6]:= m@@2, 2DD = x; m
Out[6]= 88a, b, c<, 8d, x, f<, 8g, h, i<<

This resets all elements in the second column to be z.

Mathematics and Algorithms 153

This resets all elements in the second column to be z.

In[7]:= m@@All, 2DD = z; m
Out[7]= 88a, z, c<, 8d, z, f<, 8g, z, i<<

This separately resets the three elements in the second column.

In[8]:= m@@All, 2DD = 8i, j, k<; m
Out[8]= 88a, i, c<, 8d, j, f<, 8g, k, i<<

This increments all the values in the second column.

In[9]:= m@@All, 2DD ++; m
Out[9]= 88a, 1 + i, c<, 8d, 1 + j, f<, 8g, 1 + k, i<<

A range of indices can be specified by using ;; (Span).

This resets the first two rows to be new vectors.

In[10]:= m@@1 ;; 2DD = 88u, v, w<, 8x, y, z<<; m
Out[10]= 88u, v, w<, 8x, y, z<, 8g, 1 + k, i<<

This resets elements in the first and third columns of each row.
In[11]:= m@@All, 1 ;; 3 ;; 2DD = 881, 2<, 83, 4<, 85, 6<<; m
Out[11]= 881, v, 2<, 83, y, 4<, 85, 1 + k, 6<<

This resets elements in the first and third columns of rows 2 through 3.
In[12]:= m@@2 ;; 3, 1 ;; 3 ;; 2DD = 88a, b<, 8c, d<<; m
Out[12]= 881, v, 2<, 8a, y, b<, 8c, 1 + k, d<<

Scalars, Vectors and Matrices

Mathematica represents matrices and vectors using lists. Anything that is not a list Mathematica
considers as a scalar.

A vector in Mathematica consists of a list of scalars. A matrix consists of a list of vectors, repre-
senting each of its rows. In order to be a valid matrix, all the rows must be the same length, so
that the elements of the matrix effectively form a rectangular array.
154 Mathematics and Algorithms

VectorQ@exprD give True if expr has the form of a vector, and False
otherwise
MatrixQ@exprD give True if expr has the form of a matrix, and False
otherwise
Dimensions@exprD a list of the dimensions of a vector or matrix

Functions for testing the structure of vectors and matrices.

The list 8a, b, c< has the form of a vector.

In[1]:= VectorQ@8a, b, c<D
Out[1]= True

Anything that is not manifestly a list is treated as a scalar, so applying VectorQ gives False .
In[2]:= VectorQ@x + yD
Out[2]= False

This is a 23 matrix.
In[3]:= Dimensions@88a, b, c<, 8ap, bp, cp<<D
Out[3]= 82, 3<

For a vector, Dimensions gives a list with a single element equal to the result from Length.
In[4]:= Dimensions@8a, b, c<D
Out[4]= 83<

This object does not count as a matrix because its rows are of different lengths.
In[5]:= MatrixQ@88a, b, c<, 8ap, bp<<D
Out[5]= False

Operations on Scalars, Vectors and Matrices

Most mathematical functions in Mathematica are set up to apply themselves separately to each
element in a list. This is true in particular of all functions that carry the attribute Listable.

A consequence is that most mathematical functions are applied element by element to matrices
and vectors.
Mathematics and Algorithms 155

The Log applies itself separately to each element in the vector.

In[1]:= Log@8a, b, c<D
Out[1]= 8Log@aD, Log@bD, Log@cD<

The same is true for a matrix, or, for that matter, for any nested list.
In[2]:= Log@88a, b<, 8c, d<<D
Out[2]= 88Log@aD, Log@bD<, 8Log@cD, Log@dD<<

The differentiation function D also applies separately to each element in a list.

In[3]:= D@8x, x ^ 2, x ^ 3<, xD
2
Out[3]= 91, 2 x, 3 x =

The sum of two vectors is carried out element by element.

In[4]:= 8a, b< + 8ap, bp<
Out[4]= 8a + ap, b + bp<

If you try to add two vectors with different lengths, you get an error.
In[5]:= 8a, b, c< + 8ap, bp<

Thread::tdlen : Objects of unequal length in 8a, b, c< + 8ap, bp< cannot be combined.

Out[5]= 8ap, bp< + 8a, b, c<

This adds the scalar 1 to each element of the vector.

In[6]:= 1 + 8a, b<
Out[6]= 81 + a, 1 + b<

Any object that is not manifestly a list is treated as a scalar. Here c is treated as a scalar, and
added separately to each element in the vector.
In[7]:= 8a, b< + c
Out[7]= 8a + c, b + c<

This multiplies each element in the vector by the scalar k.

In[8]:= k 8a, b<
Out[8]= 8a k, b k<

It is important to realize that Mathematica treats an object as a vector in a particular operation

only if the object is explicitly a list at the time when the operation is done. If the object is not
explicitly a list, Mathematica always treats it as a scalar. This means that you can get different
results, depending on whether you assign a particular object to be a list before or after you do
156 Mathematics and Algorithms

It is important to realize that Mathematica treats an object as a vector in a particular operation

The object p is treated as a scalar, and added separately to each element in the vector.
In[9]:= 8a, b< + p
Out[9]= 8a + p, b + p<

This is what happens if you now replace p by the list 8c, d<.
In[10]:= % . p -> 8c, d<
Out[10]= 88a + c, a + d<, 8b + c, b + d<<

You would have got a different result if you had replaced p by 8c, d< before you did the first
operation.
In[11]:= 8a, b< + 8c, d<
Out[11]= 8a + c, b + d<

Multiplying Vectors and Matrices

c v , c m , etc. multiply each element by a scalar

u.v , v.m , m.v , m1 .m2 , etc. vector and matrix multiplication
Cross@u,vD vector cross product (also input as u v)
OuterATimes,t,uE outer product

KroneckerProduct@m1 ,m2 ,D Kronecker product

Different kinds of vector and matrix multiplication.

This multiplies each element of the vector by the scalar k.

In[1]:= k 8a, b, c<
Out[1]= 8a k, b k, c k<
Mathematics and Algorithms 157

The "dot" operator gives the scalar product of two vectors.

In[2]:= 8a, b, c<.8ap, bp, cp<
Out[2]= a ap + b bp + c cp

You can also use dot to multiply a matrix by a vector.

In[3]:= 88a, b<, 8c, d<<.8x, y<
Out[3]= 8a x + b y, c x + d y<

Dot is also the notation for matrix multiplication in Mathematica.

In[4]:= 88a, b<, 8c, d<<.881, 2<, 83, 4<<
Out[4]= 88a + 3 b, 2 a + 4 b<, 8c + 3 d, 2 c + 4 d<<

It is important to realize that you can use "dot" for both left- and right-multiplication of vectors
by matrices. Mathematica makes no distinction between "row" and "column" vectors. Dot car-
ries out whatever operation is possible. (In formal terms, a.b contracts the last index of the
tensor a with the first index of b.)

Here are definitions for a matrix m and a vector v.

In[5]:= m = 88a, b<, 8c, d<<; v = 8x, y<
Out[5]= 8x, y<

This left-multiplies the vector v by m. The object v is effectively treated as a column vector in
this case.
In[6]:= m.v
Out[6]= 8a x + b y, c x + d y<

You can also use dot to right-multiply v by m. Now v is effectively treated as a row vector.
In[7]:= v.m
Out[7]= 8a x + c y, b x + d y<

You can multiply m by v on both sides, to get a scalar.

In[8]:= v.m.v
Out[8]= x Ha x + c yL + y Hb x + d yL

For some purposes, you may need to represent vectors and matrices symbolically, without
explicitly giving their elements. You can use dot to represent multiplication of such symbolic
objects.
158 Mathematics and Algorithms

For some purposes, you may need to represent vectors and matrices symbolically, without
explicitly giving their elements. You can use dot to represent multiplication of such symbolic
objects.

Dot effectively acts here as a noncommutative form of multiplication.

In[9]:= a.b.a
Out[9]= a.b.a

It is, nevertheless, associative.

In[10]:= Ha.bL.Ha.bL
Out[10]= a.b.a.b

Dot products of sums are not automatically expanded out.

In[11]:= Ha + bL.c.Hd + eL
Out[11]= Ha + bL.c.Hd + eL

You can apply the distributive law in this case using the function Distribute , as discussed in
"Structural Operations".
In[12]:= Distribute@%D
Out[12]= a.c.d + a.c.e + b.c.d + b.c.e

The "dot" operator gives "inner products" of vectors, matrices, and so on. In more advanced
calculations, you may also need to construct outer or Kronecker products of vectors and matri-
ces. You can use the general function Outer or KroneckerProduct to do this.

The outer product of two vectors is a matrix.

In[13]:= Outer@Times, 8a, b<, 8c, d<D
Out[13]= 88a c, a d<, 8b c, b d<<

The outer product of a matrix and a vector is a rank three tensor.

In[14]:= Outer@Times, 881, 2<, 83, 4<<, 8x, y, z<D
Out[14]= 888x, y, z<, 82 x, 2 y, 2 z<<, 883 x, 3 y, 3 z<, 84 x, 4 y, 4 z<<<
Mathematics and Algorithms 159

Outer products are discussed in more detail in "Tensors".

The Kronecker product of a matrix and a vector is a matrix.

In[15]:= KroneckerProduct@881, 2<, 83, 4<<, 8x, y, z<D
Out[15]= 88x, y, z, 2 x, 2 y, 2 z<, 83 x, 3 y, 3 z, 4 x, 4 y, 4 z<<

The Kronecker product of a pair of 22 matrices is a 44 matrix.

In[16]:= KroneckerProduct@881, 2<, 83, 4<<, 88a, b<, 8c, d<<D
Out[16]= 88a, b, 2 a, 2 b<, 8c, d, 2 c, 2 d<, 83 a, 3 b, 4 a, 4 b<, 83 c, 3 d, 4 c, 4 d<<

Vector Operations

v@@iDD or Part@v,iD give the ith element in the vector v

c v scalar multiplication of c times the vector v
u.v dot product of two vectors
Norm@vD give the norm of v
Normalize@vD give a unit vector in the direction of v
Standardize @vD shift v to have zero mean and unit sample variance
Standardize @v, f1 D shift v by f1 @vD and scale to have unit sample variance

Basic vector operations.

This is a vector in three dimensions.

In[1]:= v = 81, 3, 2<
Out[1]= 81, 3, 2<

This gives a vector u in the direction opposite to v with twice the magnitude.
In[2]:= u = -2 v
Out[2]= 8-2, -6, -4<

This reassigns the first component of u to be its negative.

In[3]:= u@@1DD = - u@@1DD; u
Out[3]= 82, -6, -4<
160 Mathematics and Algorithms

This gives the dot product of u and v.

In[4]:= u.v
Out[4]= -24

This is the norm of v.

In[5]:= Norm@vD

Out[5]= 14

This is the unit vector in the same direction as v.

In[6]:= Normalize@vD

1 3 2
Out[6]= : , , >
14 14 7

This verifies that the norm is 1.

In[7]:= Norm@%D
Out[7]= 1

Transform v to have zero mean and unit sample variance.

In[8]:= Standardize@vD
Out[8]= 8-1, 1, 0<

This shows the transformed values have mean 0 and variance 1.

In[9]:= 8Mean@%D, Variance@%D<
Out[9]= 80, 1<

Two vectors are orthogonal if their dot product is zero. A set of vectors is orthonormal if they
are all unit vectors and are pairwise orthogonal.

Projection@u,vD give the orthogonal projection of u onto v

OrthogonalizeA9v1 ,v2, =E generate an orthonormal set from the given list of vectors

Orthogonal vector operations.

Mathematics and Algorithms 161

This gives the projection of u onto v.

In[10]:= p = Projection@u, vD
12 36 24
Out[10]= :- , - , - >
7 7 7

p is a scalar multiple of v.
In[11]:= pv
12 12 12
Out[11]= :- , - , - >
7 7 7

u - p is orthogonal to v.
In[12]:= Hu - pL.v
Out[12]= 0

Starting from the set of vectors 8u, v<, this finds an orthonormal set of two vectors.
In[13]:= Orthogonalize@8u, v<D

1 3 2 13 3 2
Out[13]= :: , - , - >, : , , >>
14 14 7 14 182 91

When one of the vectors is linearly dependent on the vectors preceding it, the corresponding
position in the result will be a zero vector.
In[14]:= Orthogonalize@8v, p, u<D

1 3 2 13 3 2
Out[14]= :: , , >, 80, 0, 0<, : , - , - >>
14 14 7 14 182 91

Matrix Inversion

Inverse@mD find the inverse of a square matrix

Matrix inversion.

Here is a simple 22 matrix.

In[1]:= m = 88a, b<, 8c, d<<
Out[1]= 88a, b<, 8c, d<<

This gives the inverse of m. In producing this formula, Mathematica implicitly assumes that the
determinant a d - b c is nonzero.
162 Mathematics and Algorithms

This gives the inverse of m. In producing this formula, Mathematica implicitly assumes that the
determinant a d - b c is nonzero.
In[2]:= Inverse@mD
d b c a
Out[2]= :: , - >, :- , >>
-b c + a d -b c + a d -b c + a d -b c + a d

Multiplying the inverse by the original matrix should give the identity matrix.
In[3]:= %.m
bc ad bc ad
Out[3]= ::- + , 0>, :0, - + >>
-b c + a d -b c + a d -b c + a d -b c + a d

You have to use Together to clear the denominators, and get back a standard identity matrix.
In[4]:= Together@%D
Out[4]= 881, 0<, 80, 1<<

Here is a matrix of rational numbers.

In[5]:= hb = Table@1 Hi + jL, 8i, 4<, 8j, 4<D
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Out[5]= :: , , , >, : , , , >, : , , , >, : , , , >>
2 3 4 5 3 4 5 6 4 5 6 7 5 6 7 8

Mathematica finds the exact inverse of the matrix.

In[6]:= Inverse@hbD
Out[6]= 88200, -1200, 2100, -1120<, 8-1200, 8100, -15 120, 8400<,
82100, -15 120, 29 400, -16 800<, 8-1120, 8400, -16 800, 9800<<

Multiplying by the original matrix gives the identity matrix.

In[7]:= %.hb
Out[7]= 881, 0, 0, 0<, 80, 1, 0, 0<, 80, 0, 1, 0<, 80, 0, 0, 1<<

If you try to invert a singular matrix, Mathematica prints a warning message, and returns the
input unchanged.
In[8]:= Inverse@881, 2<, 81, 2<<D

Inverse::sing : Matrix 881, 2<, 81, 2<< is singular.

Out[8]= Inverse@881, 2<, 81, 2<<D

Mathematics and Algorithms 163

If you give a matrix with exact symbolic or numerical entries, Mathematica gives the exact
inverse. If, on the other hand, some of the entries in your matrix are approximate real num-
bers, then Mathematica finds an approximate numerical result.

Here is a matrix containing approximate real numbers.

In[9]:= m = 881.2, 5.7<, 81.3, 5.6<<
Out[9]= 881.2, 5.7<, 81.3, 5.6<<

This finds the numerical inverse.

In[10]:= Inverse@%D
Out[10]= 88-8.11594, 8.26087<, 81.88406, -1.73913<<

Multiplying by the original matrix gives you an identity matrix with small round-off errors.
In[11]:= %.m
-15 -16
Out[11]= 991., 1.66187 10 =, 93.27429 10 , 1.==

You can get rid of small off-diagonal terms using Chop .

In[12]:= Chop@%D
Out[12]= 881., 0<, 80, 1.<<

When you try to invert a matrix with exact numerical entries, Mathematica can always tell
whether or not the matrix is singular. When you invert an approximate numerical matrix, Mathe-
matica can usually not tell for certain whether or not the matrix is singular: all it can tell is, for
example, that the determinant is small compared to the entries of the matrix. When Mathemat-
ica suspects that you are trying to invert a singular numerical matrix, it prints a warning.

Mathematica prints a warning if you invert a numerical matrix that it suspects is singular.
In[13]:= Inverse@881., 2.<, 81., 2.<<D

Inverse::sing : Matrix 881., 2.<, 81., 2.<< is singular.

Out[13]= Inverse@881., 2.<, 81., 2.<<D

164 Mathematics and Algorithms

This matrix is singular, but the warning is different, and the result is useless.
In[14]:= Inverse@N@881, 2, 3<, 84, 5, 6<, 87, 8, 9<<DD

Inverse::luc : Result for Inverse of badly conditioned matrix

881., 2., 3.<, 84., 5., 6.<, 87., 8., 9.<< may contain significant numerical errors.
15 15 15
Out[14]= 993.15221 10 , -6.30442 10 , 3.15221 10 =,
9-6.30442 1015 , 1.26088 1016 , -6.30442 1015 =, 93.15221 1015 , -6.30442 1015 , 3.15221 1015 ==

If you work with high-precision approximate numbers, Mathematica will keep track of the
precision of matrix inverses that you generate.

This generates a 66 numerical matrix with entries of 20-digit precision.

In[15]:= m = N@Table@GCD@i, jD + 1, 8i, 6<, 8j, 6<D, 20D;

This takes the matrix, multiplies it by its inverse, and shows the first row of the result.
In[16]:= Hm.Inverse@mDL@@1DD
-19 -19 -20 -20 -20
Out[16]= 91.000000000000000000, 0. 10 , 0. 10 , 0. 10 , 0. 10 , 0. 10 =

This generates a 20-digit numerical approximation to a 66 Hilbert matrix. Hilbert matrices are
notoriously hard to invert numerically.
In[17]:= m = N@Table@1 Hi + j - 1L, 8i, 6<, 8j, 6<D, 20D;

The result is still correct, but the zeros now have lower accuracy.
In[18]:= Hm.Inverse@mDL@@1DD
-15 -14 -14 -14 -14
Out[18]= 91.000000000000000, 0. 10 , 0. 10 , 0. 10 , 0. 10 , 0. 10 =

Inverse works only on square matrices. "Advanced Matrix Operations" discusses the function
PseudoInverse, which can also be used with nonsquare matrices.

Basic Matrix Operations

Transpose@mD transpose m

ConjugateTranspose@mD conjugate transpose m (Hermitian conjugate)

Inverse@mD matrix inverse
Det@mD determinant
Minors@mD matrix of minors
Mathematics and Algorithms 165

Minors@m,kD kth minors

Tr@mD trace
MatrixRank@mD rank of matrix

Some basic matrix operations.

Transposing a matrix interchanges the rows and columns in the matrix. If you transpose an
mn matrix, you get an nm matrix as the result.

Transposing a 23 matrix gives a 32 result.

In[1]:= Transpose@88a, b, c<, 8ap, bp, cp<<D
Out[1]= 88a, ap<, 8b, bp<, 8c, cp<<

Det@mD gives the determinant of a square matrix m. Minors@mD is the matrix whose Hi, jLth
element gives the determinant of the submatrix obtained by deleting the Hn - i + 1Lth row and the
Hn - j + 1Lth column of m. The Hi, jLth cofactor of m is H-1Li+ j times the Hn - i + 1, n - j + 1Lth element of
the matrix of minors.

Minors@m, kD gives the determinants of the kk submatrices obtained by picking each possible
set of k rows and k columns from m. Note that you can apply Minors to rectangular, as well as
square, matrices.

Here is the determinant of a simple 22 matrix.

In[2]:= Det@88a, b<, 8c, d<<D
Out[2]= -b c + a d

This generates a 33 matrix, whose i, jth entry is a@i, jD.

In[3]:= m = Array@a, 83, 3<D
Out[3]= 88a@1, 1D, a@1, 2D, a@1, 3D<, 8a@2, 1D, a@2, 2D, a@2, 3D<, 8a@3, 1D, a@3, 2D, a@3, 3D<<

Here is the determinant of m.

In[4]:= Det@mD
Out[4]= -a@1, 3D a@2, 2D a@3, 1D + a@1, 2D a@2, 3D a@3, 1D + a@1, 3D a@2, 1D a@3, 2D -
a@1, 1D a@2, 3D a@3, 2D - a@1, 2D a@2, 1D a@3, 3D + a@1, 1D a@2, 2D a@3, 3D
166 Mathematics and Algorithms

The trace or spur of a matrix Tr@mD is the sum of the terms on the leading diagonal.

This finds the trace of a simple 22 matrix.

In[5]:= Tr@88a, b<, 8c, d<<D
Out[5]= a + d

The rank of a matrix is the number of linearly independent rows or columns.

This finds the rank of a matrix.

In[6]:= MatrixRank@881, 2<, 81, 2<<D
Out[6]= 1

MatrixPower @m,nD nth matrix power

MatrixExp@mD matrix exponential

Powers and exponentials of matrices.

Here is a 22 matrix.
In[7]:= m = 880.4, 0.6<, 80.525, 0.475<<
Out[7]= 880.4, 0.6<, 80.525, 0.475<<

This gives the third matrix power of m.

In[8]:= MatrixPower@m, 3D
Out[8]= 880.465625, 0.534375<, 80.467578, 0.532422<<

It is equivalent to multiplying three copies of the matrix.

In[9]:= m.m.m
Out[9]= 880.465625, 0.534375<, 80.467578, 0.532422<<

Here is the millionth matrix power.

In[10]:= MatrixPower@m, 10 ^ 6D
Out[10]= 880.466667, 0.533333<, 80.466667, 0.533333<<
Mathematics and Algorithms 167

The matrix exponential of a matrix m is

k=0 m k !, where m indicates a matrix power.
k k

This gives the matrix exponential of m.

In[11]:= MatrixExp@mD
Out[11]= 881.7392, 0.979085<, 80.8567, 1.86158<<

Here is an approximation to the exponential of m, based on a power series approximation.

In[12]:= Sum@MatrixPower@m, kD k !, 8k, 0, 5<D
Out[12]= 881.73844, 0.978224<, 80.855946, 1.86072<<

Solving Linear Systems

Many calculations involve solving systems of linear equations. In many cases, you will find it
convenient to write down the equations explicitly, and then solve them using Solve.

In some cases, however, you may prefer to convert the system of linear equations into a matrix
equation, and then apply matrix manipulation operations to solve it. This approach is often
useful when the system of equations arises as part of a general algorithm, and you do not know
in advance how many variables will be involved.

A system of linear equations can be stated in matrix form as m.x = b, where x is the vector of
variables.

Note that if your system of equations is sparse, so that most of the entries in the matrix m are
zero, then it is best to represent the matrix as a SparseArray object. As discussed in "Sparse
Arrays: Linear Algebra", you can convert from symbolic equations to SparseArray objects
using CoefficientArrays. All the functions described here work on SparseArray objects as
well as ordinary matrices.

LinearSolve @m,bD a vector x which solves the matrix equation m.x == b

NullSpace@mD a list of linearly independent vectors whose linear combina-
tions span all solutions to the matrix equation m.x == 0
MatrixRank@mD the number of linearly independent rows or columns of m
RowReduce@mD a simplified form of m obtained by making linear combina-
tions of rows

Solving and analyzing linear systems.

Here is a 2x2 matrix.

168 Mathematics and Algorithms

Here is a 2x2 matrix.

In[1]:= m = 881, 5<, 82, 1<<
Out[1]= 881, 5<, 82, 1<<

This gives two linear equations.

In[2]:= m.8x, y< == 8a, b<
Out[2]= 8x + 5 y, 2 x + y< 8a, b<

You can use Solve directly to solve these equations.

In[3]:= Solve@%, 8x, y<D
1 1
Out[3]= ::x H-a + 5 bL, y H2 a - bL>>
9 9

You can also get the vector of solutions by calling LinearSolve . The result is equivalent to the
one you get from Solve .
In[4]:= LinearSolve@m, 8a, b<D
1 1
Out[4]= : H-a + 5 bL, H2 a - bL>
9 9

Another way to solve the equations is to invert the matrix m, and then multiply 8a, b< by the
inverse. This is not as efficient as using LinearSolve .
In[5]:= [email protected], b<
a 5b 2a b
Out[5]= :- + , - >
9 9 9 9

RowReduce performs a version of Gaussian elimination and can also be used to solve the
equations.
In[6]:= RowReduce@881, 5, a<, 82, 1, b<<D
1 1
Out[6]= ::1, 0, H-a + 5 bL>, :0, 1, H2 a - bL>>
9 9

If you have a square matrix m with a nonzero determinant, then you can always find a unique
solution to the matrix equation m.x = b for any b. If, however, the matrix m has determinant
zero, then there may be either no vector, or an infinite number of vectors x which satisfy m.x = b
for a particular b. This occurs when the linear equations embodied in m are not independent.

When m has determinant zero, it is nevertheless always possible to find nonzero vectors x that
satisfy m.x = 0. The set of vectors x satisfying this equation form the null space or kernel of the
matrix m. Any of these vectors can be expressed as a linear combination of a particular set of
basis vectors, which can be obtained using NullSpace@mD.
Mathematics and Algorithms 169

Here is a simple matrix, corresponding to two identical linear equations.

In[7]:= m = 881, 2<, 81, 2<<
Out[7]= 881, 2<, 81, 2<<

The matrix has determinant zero.

In[8]:= Det@mD
Out[8]= 0

LinearSolve cannot find a solution to the equation m.x b in this case.

In[9]:= LinearSolve@m, 8a, b<D

LinearSolve::nosol : Linear equation encountered that has no solution.

Out[9]= LinearSolve@881, 2<, 81, 2<<, 8a, b<D

There is a single basis vector for the null space of m.

In[10]:= NullSpace@mD
Out[10]= 88-2, 1<<

Multiplying the basis vector for the null space by m gives the zero vector.
In[11]:= m.%@@1DD
Out[11]= 80, 0<

There is only 1 linearly independent row in m.

In[12]:= MatrixRank@mD
Out[12]= 1

NullSpace and MatrixRank have to determine whether particular combinations of matrix

elements are zero. For approximate numerical matrices, the Tolerance option can be used to
specify how close to zero is considered good enough. For exact symbolic matrices, you may
sometimes need to specify something like ZeroTest -> HFullSimplify@D == 0 &L to force
more to be done to test whether symbolic expressions are zero.

Here is a simple symbolic matrix with determinant zero.

170 Mathematics and Algorithms

Here is a simple symbolic matrix with determinant zero.

In[13]:= m = 88a, b, c<, 82 a, 2 b, 2 c<, 83 a, 3 b, 3 c<<
Out[13]= 88a, b, c<, 82 a, 2 b, 2 c<, 83 a, 3 b, 3 c<<

The basis for the null space of m contains two vectors.

In[14]:= NullSpace@mD
c b
Out[14]= ::- , 0, 1>, :- , 1, 0>>
a a

Multiplying m by any linear combination of these vectors gives zero.

In[15]:= [email protected] %@@1DD + y %@@2DDLD
Out[15]= 80, 0, 0<

An important feature of functions like LinearSolve and NullSpace is that they work with
rectangular, as well as square, matrices.

When you represent a system of linear equations by a matrix equation of the form m.x = b, the
number of columns in m gives the number of variables, and the number of rows gives the num-
ber of equations. There are a number of cases.

Underdetermined number of equations less than the number of variables; no

solutions or many solutions may exist
Overdetermined number of equations more than the number of variables;
solutions may or may not exist
Nonsingular number of independent equations equal to the number of
variables, and determinant nonzero; a unique solution
exists
Consistent at least one solution exists
Inconsistent no solutions exist

Classes of linear systems represented by rectangular matrices.

This asks for the solution to the inconsistent set of equations x = 1 and x = 0.
In[16]:= LinearSolve@881<, 81<<, 81, 0<D

LinearSolve::nosol : Linear equation encountered that has no solution.

Out[16]= LinearSolve@881<, 81<<, 81, 0<D

Mathematics and Algorithms 171

This matrix represents two equations, for three variables.

In[17]:= m = 881, 3, 4<, 82, 1, 3<<
Out[17]= 881, 3, 4<, 82, 1, 3<<

LinearSolve gives one of the possible solutions to this underdetermined set of equations.
In[18]:= v = LinearSolve@m, 81, 1<D
2 1
Out[18]= : , , 0>
5 5

When a matrix represents an underdetermined system of equations, the matrix has a nontrivial
null space. In this case, the null space is spanned by a single vector.
In[19]:= NullSpace@mD
Out[19]= 88-1, -1, 1<<

If you take the solution you get from LinearSolve , and add any linear combination of the
basis vectors for the null space, you still get a solution.
In[20]:= m.Hv + 4 %@@1DDL
Out[20]= 81, 1<

The number of independent equations is the rank of the matrix MatrixRank@mD. The number of
redundant equations is Length@NullSpace@mDD. Note that the sum of these quantities is always
equal to the number of columns in m.

LinearSolve @mD generate a function for solving equations of the form m.x = b

Generating LinearSolveFunction objects.

In some applications, you will want to solve equations of the form m.x = b many times with the
same m, but different b. You can do this efficiently in Mathematica by using LinearSolve @mD to
create a single LinearSolveFunction that you can apply to as many vectors as you want.

This creates a LinearSolveFunction .

In[21]:= f = LinearSolve@881, 4<, 82, 3<<D
Out[21]= LinearSolveFunction@82, 2<, <>D
172 Mathematics and Algorithms

You can apply this to a vector.

In[22]:= f@85, 7<D
13 3
Out[22]= : , >
5 5

You get the same result by giving the vector as an explicit second argument to LinearSolve .
In[23]:= LinearSolve@881, 4<, 82, 3<<, 85, 7<D
13 3
Out[23]= : , >
5 5

But you can apply f to any vector you want.

In[24]:= f@8- 5, 9<D
51 19
Out[24]= : , - >
5 5

LeastSquares@m,bD give a vector x that solves the least-squares problem

m.x == b

Solving least-squares problems.

This linear system is inconsistent.

In[25]:= LinearSolve@881, 2<, 83, 4<, 85, 6<<, 8- 1, 0, 2<D

LinearSolve::nosol : Linear equation encountered that has no solution.

Out[25]= LinearSolve@881, 2<, 83, 4<, 85, 6<<, 8-1, 0, 2<D

LeastSquares finds a vector x that minimizes m.x - b in the least-squares sense.

In[26]:= LeastSquares@881, 2<, 83, 4<, 85, 6<<, 8- 1, 0, 2<D
8 23
Out[26]= : , - >
3 12

Eigenvalues and Eigenvectors

Eigenvalues @mD a list of the eigenvalues of m

Eigenvectors@mD a list of the eigenvectors of m
Mathematics and Algorithms 173

Eigensystem @mD a list of the form 8eigenvalues, eigenvectors<

Eigenvalues AN@mDE , etc. numerical eigenvalues

Eigenvalues AN@m,pDE , etc. numerical eigenvalues, starting with p-digit precision

CharacteristicPolynomial@m,xD
the characteristic polynomial of m

Eigenvalues and eigenvectors.

The eigenvalues of a matrix m are the values li for which one can find nonzero vectors vi such
that m.vi = li vi . The eigenvectors are the vectors vi .

The characteristic polynomial CharacteristicPolynomial@m, xD for an n n matrix is given by

Det@m - x IdentityMatrix@nDD. The eigenvalues are the roots of this polynomial.

Finding the eigenvalues of an n n matrix in general involves solving an nth -degree polynomial
equation. For n 5, therefore, the results cannot in general be expressed purely in terms of
explicit radicals. Root objects can nevertheless always be used, although except for fairly
sparse or otherwise simple matrices the expressions obtained are often unmanageably complex.

Even for a matrix as simple as this, the explicit form of the eigenvalues is quite complicated.
In[1]:= Eigenvalues@88a, b<, 8- b, 2 a<<D
1 1
Out[1]= : 3a- a2 - 4 b 2 , 3a+ a2 - 4 b 2 >
2 2

If you give a matrix of approximate real numbers, Mathematica will find the approximate numeri -
cal eigenvalues and eigenvectors.

Here is a 22 numerical matrix.

In[2]:= m = 882.3, 4.5<, 86.7, - 1.2<<
Out[2]= 882.3, 4.5<, 86.7, -1.2<<

The matrix has two eigenvalues, in this case both real.

In[3]:= Eigenvalues@mD
Out[3]= 86.31303, -5.21303<

Here are the two eigenvectors of m.

In[4]:= Eigenvectors@mD
Out[4]= 880.746335, 0.66557<, 8-0.513839, 0.857886<<

Eigensystem computes the eigenvalues and eigenvectors at the same time. The assignment
sets vals to the list of eigenvalues, and vecs to the list of eigenvectors.
174 Mathematics and Algorithms

Eigensystem computes the eigenvalues and eigenvectors at the same time. The assignment
sets vals to the list of eigenvalues, and vecs to the list of eigenvectors.
In[5]:= 8vals, vecs< = Eigensystem@mD
Out[5]= 886.31303, -5.21303<, 880.746335, 0.66557<, 8-0.513839, 0.857886<<<

This verifies that the first eigenvalue and eigenvector satisfy the appropriate condition.
In[6]:= m.vecs@@1DD == vals@@1DD vecs@@1DD
Out[6]= True

This finds the eigenvalues of a random 44 matrix. For nonsymmetric matrices, the eigenvalues
can have imaginary parts.
In[7]:= Eigenvalues@Table@RandomReal@D, 84<, 84<DD
Out[7]= 82.30022, 0.319764 + 0.547199 , 0.319764 - 0.547199 , 0.449291<

The function Eigenvalues always gives you a list of n eigenvalues for an nn matrix. The eigen-
values correspond to the roots of the characteristic polynomial for the matrix, and may not
necessarily be distinct. Eigenvectors, on the other hand, gives a list of eigenvectors which are
guaranteed to be independent. If the number of such eigenvectors is less than n, then
Eigenvectors appends zero vectors to the list it returns, so that the total length of the list is
always n.

Here is a 33 matrix.
In[8]:= mz = 880, 1, 0<, 80, 0, 1<, 80, 0, 0<<
Out[8]= 880, 1, 0<, 80, 0, 1<, 80, 0, 0<<

The matrix has three eigenvalues, all equal to zero.

In[9]:= Eigenvalues@mzD
Out[9]= 80, 0, 0<

There is, however, only one independent eigenvector for the matrix. Eigenvectors appends
two zero vectors to give a total of three vectors in this case.
In[10]:= Eigenvectors@mzD
Out[10]= 881, 0, 0<, 80, 0, 0<, 80, 0, 0<<
Mathematics and Algorithms 175

This gives the characteristic polynomial of the matrix.

In[11]:= CharacteristicPolynomial@mz, xD
3
Out[11]= -x

Eigenvalues @m,kD the largest k eigenvalues of m

Eigenvectors@m,kD the corresponding eigenvectors of m
Eigensystem @m,kD the largest k eigenvalues with corresponding eigenvectors
Eigenvalues @m,-kD the smallest k eigenvalues of m
Eigenvectors@m,-kD the corresponding eigenvectors of m
Eigensystem @m,-kD the smallest k eigenvalues with corresponding eigenvectors

Finding largest and smallest eigenvalues.

Eigenvalues sorts numeric eigenvalues so that the ones with large absolute value come first.
In many situations, you may be interested only in the largest or smallest eigenvalues of a
matrix. You can get these efficiently using Eigenvalues @m, kD and Eigenvalues @m, - kD.

This computes the exact eigenvalues of an integer matrix.

In[12]:= Eigenvalues@881, 2<, 83, 4<<D
1 1
Out[12]= : 5+ 33 , 5- 33 >
2 2

The eigenvalues are sorted in decreasing order of size.

In[13]:= N@%D
Out[13]= 85.37228, -0.372281<

This gives the three eigenvalues with largest absolute value.

In[14]:= Eigenvalues@Table@N@Tan@i jDD, 8i, 10<, 8j, 10<D, 3D
Out[14]= 810.044, 2.94396 + 6.03728 , 2.94396 - 6.03728 <

Eigenvalues @8m,a<D the generalized eigenvalues of m with respect to a

Eigenvectors@8m,a<D the generalized eigenvectors of m with respect to a
Eigensystem @8m,a<D the generalized eigensystem of m with respect to a
CharacteristicPolynomial@8m,a<,xD
the generalized characteristic polynomial of m with respect
to a

Generalized eigenvalues, eigenvectors, and characteristic polynomial.

The generalized eigenvalues for a matrix m with respect to a matrix a are defined to be those li
176 Mathematics and Algorithms

The generalized eigenvalues for a matrix m with respect to a matrix a are defined to be those li
for which m.vi = li a.vi .

The generalized eigenvalues correspond to zeros of the generalized characteristic polynomial

Det@m - x aD.

Note that while ordinary matrix eigenvalues always have definite values, some generalized
eigenvalues will always be Indeterminate if the generalized characteristic polynomial vanishes,
which happens if m and a share a null space. Note also that generalized eigenvalues can be
infinite.

These two matrices share a one-dimensional null space, so one generalized eigenvalue is
Indeterminate .
In[15]:= [email protected], 0<, 80, 0<<, 882, 0<, 81, 0<<<D
Out[15]= 80., Indeterminate<

This gives a generalized characteristic polynomial.

In[16]:= [email protected], 0<, 80, 1<<, 882, 0<, 81, 0<<<, xD
Out[16]= 1.5 - 2. x

Advanced Matrix Operations

SingularValueList@mD the list of nonzero singular values of m

SingularValueList@m,kD the k largest singular values of m
SingularValueList@8m,a<D the generalized singular values of m with respect to a
Norm@m,pD the p-norm of m
Norm@m,"Frobenius"D the Frobenius norm of m

Finding singular values and norms of matrices.

The singular values of a matrix m are the square roots of the eigenvalues of m.m* , where *
denotes Hermitian transpose. The number of such singular values is the smaller dimension of
the matrix. SingularValueList sorts the singular values from largest to smallest. Very small
singular values are usually numerically meaningless. With the option setting Tolerance -> t,
SingularValueList drops singular values that are less than a fraction t of the largest singular
value. For approximate numerical matrices, the tolerance is by default slightly greater than zero.

If you multiply the vector for each point in a unit sphere in n-dimensional space by an mn
matrix m, then you get an m-dimensional ellipsoid, whose principal axes have lengths given by
the singular values of m.
Mathematics and Algorithms 177

The 2-norm of a matrix Norm@m, 2D is the largest principal axis of the ellipsoid, equal to the
largest singular value of the matrix. This is also the maximum 2-norm length of m.v for any
possible unit vector v.

The p-norm of a matrix Norm@m, pD is in general the maximum p-norm length of m.v that can be
attained. The cases most often considered are p = 1, p = 2 and p = . Also sometimes considered
is the Frobenius norm Norm@m, "Frobenius"D, which is the square root of the trace of m.m* .

LUDecomposition@mD the LU decomposition

CholeskyDecomposition@mD the Cholesky decomposition

Decomposing square matrices into triangular forms.

When you create a LinearSolveFunction using LinearSolve @mD, this often works by decom-
posing the matrix m into triangular forms, and sometimes it is useful to be able to get such
forms explicitly.

LU decomposition effectively factors any square matrix into a product of lower- and upper-
triangular matrices. Cholesky decomposition effectively factors any Hermitian positive-definite
matrix into a product of a lower-triangular matrix and its Hermitian conjugate, which can be
viewed as the analog of finding a square root of a matrix.

PseudoInverse@mD the pseudoinverse

QRDecomposition@mD the QR decomposition
SingularValueDecomposition@mD the singular value decomposition
SingularValueDecomposition@ the generalized singular value decomposition
8m,a<D

Orthogonal decompositions of matrices.

The standard definition for the inverse of a matrix fails if the matrix is not square or is singular.
The pseudoinverse mH-1L of a matrix m can however still be defined. It is set up to minimize the
sum of the squares of all entries in m.mH-1L - I, where I is the identity matrix. The pseudoinverse
is sometimes known as the generalized inverse, or the Moore|Penrose inverse. It is particularly
used for problems related to least-squares fitting.

QR decomposition writes any matrix m as a product q* .r, where q is an orthonormal matrix, *

denotes Hermitian transpose, and r is a triangular matrix, in which all entries below the leading
diagonal are zero.
178 Mathematics and Algorithms

QR decomposition writes any matrix m as a product q* .r, where q is an orthonormal matrix, *

denotes Hermitian transpose, and r is a triangular matrix, in which all entries below the leading
diagonal are zero.

Singular value decomposition, or SVD, is an underlying element in many numerical matrix

algorithms. The basic idea is to write any matrix m in the form u.s.v* , where s is a matrix with
the singular values of m on its diagonal, u and v are orthonormal matrices, and v* is the Hermi-
tian transpose of v.

JordanDecomposition@mD the Jordan decomposition

SchurDecomposition@mD the Schur decomposition
SchurDecomposition@8m,a<D the generalized Schur decomposition
HessenbergDecomposition@mD the Hessenberg decomposition

Functions related to eigenvalue problems.

Most square matrices can be reduced to a diagonal matrix of eigenvalues by applying a matrix
of their eigenvectors as a similarity transformation. But even when there are not enough eigen-
vectors to do this, one can still reduce a matrix to a Jordan form in which there are both eigen-
values and Jordan blocks on the diagonal. Jordan decomposition in general writes any square
matrix in the form s. j.s-1 .

Numerically more stable is the Schur decomposition, which writes any square matrix m in the
form q.t.q* , where q is an orthonormal matrix, and t is block upper-triangular. Also related is the
Hessenberg decomposition, which writes a square matrix m in the form p.h.p* , where p is an
orthonormal matrix, and h can have nonzero elements down to the diagonal below the leading
diagonal.

Tensors
Tensors are mathematical objects that give generalizations of vectors and matrices. In Mathe-
matica, a tensor is represented as a set of lists, nested to a certain number of levels. The
nesting level is the rank of the tensor.
Mathematics and Algorithms 179

rank 0 scalar
rank 1 vector
rank 2 matrix
rank k rank k tensor

Interpretations of nested lists.

A tensor of rank k is essentially a k-dimensional table of values. To be a true rank k tensor, it

must be possible to arrange the elements in the table in a k-dimensional cuboidal array. There
can be no holes or protrusions in the cuboid.

The indices that specify a particular element in the tensor correspond to the coordinates in the
cuboid. The dimensions of the tensor correspond to the side lengths of the cuboid.

One simple way that a rank k tensor can arise is in giving a table of values for a function of k
variables. In physics, the tensors that occur typically have indices which run over the possible
directions in space or spacetime. Notice, however, that there is no built-in notion of covariant
and contravariant tensor indices in Mathematica: you have to set these up explicitly using
metric tensors.

Table@ f ,8i1 ,n1 <,8i2 ,n2 <,,8ik ,nk <D

create an n1 n2 nk tensor whose elements are the
values of f
Array@a,8n1 ,n2 ,,nk <D create an n1 n2 nk tensor with elements given by
applying a to each set of indices
ArrayQ@t,nD test whether t is a tensor of rank n
Dimensions@tD give a list of the dimensions of a tensor
ArrayDepth@tD find the rank of a tensor
MatrixForm@tD print with the elements of t arranged in a two-dimensional
array

Functions for creating and testing the structure of tensors.

Here is a 232 tensor.

In[1]:= t = Table@i1 + i2 i3, 8i1, 2<, 8i2, 3<, 8i3, 2<D
Out[1]= 8882, 3<, 83, 5<, 84, 7<<, 883, 4<, 84, 6<, 85, 8<<<
180 Mathematics and Algorithms

This is another way to produce the same tensor.

In[2]:= Array@H1 + 2 3L &, 82, 3, 2<D
Out[2]= 8882, 3<, 83, 5<, 84, 7<<, 883, 4<, 84, 6<, 85, 8<<<

MatrixForm displays the elements of the tensor in a two-dimensional array. You can think of
the array as being a 23 matrix of column vectors.
In[3]:= MatrixForm@tD
2 3 4
3 5 7
Out[3]//MatrixForm=
3 4 5
4 6 8

Dimensions gives the dimensions of the tensor.

In[4]:= Dimensions@tD
Out[4]= 82, 3, 2<

Here is the 111 element of the tensor.

In[5]:= t@@1, 1, 1DD
Out[5]= 2

ArrayDepth gives the rank of the tensor.

In[6]:= ArrayDepth@tD
Out[6]= 3

The rank of a tensor is equal to the number of indices needed to specify each element. You can
pick out subtensors by using a smaller number of indices.

Transpose@tD transpose the first two indices in a tensor

Transpose@t,8p1 ,p2 ,<D transpose the indices in a tensor so that the kth becomes
the pk th

Tr@t, f D form the generalized trace of the tensor t

Outer@ f ,t1 ,t2 D form the generalized outer product of the tensors t1 and t2
with "multiplication operator" f
t1 .t2 form the dot product of t1 and t2 (last index of t1 contracted
with first index of t2 )
Inner@ f ,t1 ,t2 ,gD form the generalized inner product, with "multiplication
operator" f and "addition operator" g

Tensor manipulation operations.

You can think of a rank k tensor as having k "slots" into which you insert indices. Applying
Transpose is effectively a way of reordering these slots. If you think of the elements of a tensor
Mathematics and Algorithms 181

In the most general case, Transpose allows you to specify an arbitrary reordering to apply to
the indices of a tensor. The function Transpose@T, 8p1 , p2 , , pk <D gives you a new tensor T
such that the value of T i1 i2 ik is given by Ti p ip ip .
1 2 k

If you originally had an n p1 n p2 n pk tensor, then by applying Transpose, you will get an

n1 n2 nk tensor.

Here is a matrix that you can also think of as a 23 tensor.

In[7]:= m = 88a, b, c<, 8ap, bp, cp<<
Out[7]= 88a, b, c<, 8ap, bp, cp<<

Applying Transpose gives you a 32 tensor. Transpose effectively interchanges the two
"slots" for tensor indices.
In[8]:= mt = Transpose@mD
Out[8]= 88a, ap<, 8b, bp<, 8c, cp<<

The element m@@2, 3DD in the original tensor becomes the element m@@3, 2DD in the trans-
posed tensor.
In[9]:= 8m@@2, 3DD, mt@@3, 2DD<
Out[9]= 8cp, cp<

This produces a 2312 tensor.

In[10]:= t = Array@a, 82, 3, 1, 2<D
Out[10]= 8888a@1, 1, 1, 1D, a@1, 1, 1, 2D<<, 88a@1, 2, 1, 1D, a@1, 2, 1, 2D<<,
88a@1, 3, 1, 1D, a@1, 3, 1, 2D<<<, 888a@2, 1, 1, 1D, a@2, 1, 1, 2D<<,
88a@2, 2, 1, 1D, a@2, 2, 1, 2D<<, 88a@2, 3, 1, 1D, a@2, 3, 1, 2D<<<<

This transposes the first two levels of t.

In[11]:= tt1 = Transpose@tD
Out[11]= 8888a@1, 1, 1, 1D, a@1, 1, 1, 2D<<, 88a@2, 1, 1, 1D, a@2, 1, 1, 2D<<<,
888a@1, 2, 1, 1D, a@1, 2, 1, 2D<<, 88a@2, 2, 1, 1D, a@2, 2, 1, 2D<<<,
888a@1, 3, 1, 1D, a@1, 3, 1, 2D<<, 88a@2, 3, 1, 1D, a@2, 3, 1, 2D<<<<
182 Mathematics and Algorithms

The result is a 3212 tensor.

In[12]:= Dimensions@tt1D
Out[12]= 83, 2, 1, 2<

If you have a tensor that contains lists of the same length at different levels, then you can use
Transpose to effectively collapse different levels.

This collapses all three levels, giving a list of the elements on the "main diagonal".
In[13]:= Transpose@Array@a, 83, 3, 3<D, 81, 1, 1<D
Out[13]= 8a@1, 1, 1D, a@2, 2, 2D, a@3, 3, 3D<

This collapses only the first two levels.

In[14]:= Transpose@Array@a, 82, 2, 2<D, 81, 1<D
Out[14]= 88a@1, 1, 1D, a@1, 1, 2D<, 8a@2, 2, 1D, a@2, 2, 2D<<

You can also use Tr to extract diagonal elements of a tensor.

This forms the ordinary trace of a rank 3 tensor.

In[15]:= Tr@Array@a, 83, 3, 3<DD
Out[15]= a@1, 1, 1D + a@2, 2, 2D + a@3, 3, 3D

Here is a generalized trace, with elements combined into a list.

In[16]:= Tr@Array@a, 83, 3, 3<D, ListD
Out[16]= 8a@1, 1, 1D, a@2, 2, 2D, a@3, 3, 3D<

This combines diagonal elements only down to level 2.

In[17]:= Tr@Array@a, 83, 3, 3<D, List, 2D
Out[17]= 88a@1, 1, 1D, a@1, 1, 2D, a@1, 1, 3D<,
8a@2, 2, 1D, a@2, 2, 2D, a@2, 2, 3D<, 8a@3, 3, 1D, a@3, 3, 2D, a@3, 3, 3D<<

Outer products, and their generalizations, are a way of building higher-rank tensors from lower-
rank ones. Outer products are also sometimes known as direct, tensor or Kronecker products.

From a structural point of view, the tensor you get from Outer@ f , t, uD has a copy of the
structure of u inserted at the "position" of each element in t. The elements in the resulting
structure are obtained by combining elements of t and u using the function f .
Mathematics and Algorithms 183

This gives the "outer f" of two vectors. The result is a matrix.
In[18]:= Outer@f, 8a, b<, 8ap, bp<D
Out[18]= 88f@a, apD, f@a, bpD<, 8f@b, apD, f@b, bpD<<

If you take the "outer f" of a length 3 vector with a length 2 vector, you get a 32 matrix.
In[19]:= Outer@f, 8a, b, c<, 8ap, bp<D
Out[19]= 88f@a, apD, f@a, bpD<, 8f@b, apD, f@b, bpD<, 8f@c, apD, f@c, bpD<<

The result of taking the "outer f" of a 22 matrix and a length 3 vector is a 223 tensor.
In[20]:= Outer@f, 88m11, m12<, 8m21, m22<<, 8a, b, c<D
Out[20]= 888f@m11, aD, f@m11, bD, f@m11, cD<, 8f@m12, aD, f@m12, bD, f@m12, cD<<,
88f@m21, aD, f@m21, bD, f@m21, cD<, 8f@m22, aD, f@m22, bD, f@m22, cD<<<

Here are the dimensions of the tensor.

In[21]:= Dimensions@%D
Out[21]= 82, 2, 3<

If you take the generalized outer product of an m1 m2 mr tensor and an n1 n2 ns tensor,

you get an m1 mr n1 ns tensor. If the original tensors have ranks r and s, your result will
be a rank r + s tensor.

In terms of indices, the result of applying Outer to two tensors Ti1 i2 ir and U j1 j2 js is the tensor

Vi1 i2 ir j1 j2 js with elements f @Ti1 i2 ir , U j1 j2 js D.

In doing standard tensor calculations, the most common function f to use in Outer is Times,
corresponding to the standard outer product.

Particularly in doing combinatorial calculations, however, it is often convenient to take f to be

List. Using Outer, you can then get combinations of all possible elements in one tensor, with
all possible elements in the other.

In constructing Outer@ f , t, uD you effectively insert a copy of u at every point in t. To form

Inner@ f , t, uD, you effectively combine and collapse the last dimension of t and the first dimen-
sion of u. The idea is to take an m1 m2 mr tensor and an n1 n2 ns tensor, with mr = n1 , and
get an m1 m2 mr-1 n2 ns tensor as the result.

The simplest examples are with vectors. If you apply Inner to two vectors of equal length, you
get a scalar. Inner@ f , v1 , v2 , gD gives a generalization of the usual scalar product, with f
playing the role of multiplication, and g playing the role of addition.
184 Mathematics and Algorithms

This gives a generalization of the standard scalar product of two vectors.

In[22]:= Inner@f, 8a, b, c<, 8ap, bp, cp<, gD
Out[22]= g@f@a, apD, f@b, bpD, f@c, cpDD

This gives a generalization of a matrix product.

In[23]:= Inner@f, 881, 2<, 83, 4<<, 88a, b<, 8c, d<<, gD
Out[23]= 88g@f@1, aD, f@2, cDD, g@f@1, bD, f@2, dDD<, 8g@f@3, aD, f@4, cDD, g@f@3, bD, f@4, dDD<<

Here is a 322 tensor.

In[24]:= a = Array@1 &, 83, 2, 2<D
Out[24]= 8881, 1<, 81, 1<<, 881, 1<, 81, 1<<, 881, 1<, 81, 1<<<

Here is a 231 tensor.

In[25]:= b = Array@2 &, 82, 3, 1<D
Out[25]= 8882<, 82<, 82<<, 882<, 82<, 82<<<

This gives a 3231 tensor.

In[26]:= a.b
Out[26]= 88884<, 84<, 84<<, 884<, 84<, 84<<<,
8884<, 84<, 84<<, 884<, 84<, 84<<<, 8884<, 84<, 84<<, 884<, 84<, 84<<<<

Here are the dimensions of the result.

In[27]:= Dimensions@%D
Out[27]= 83, 2, 3, 1<

You can think of Inner as performing a "contraction" of the last index of one tensor with the
first index of another. If you want to perform contractions across other pairs of indices, you can
do so by first transposing the appropriate indices into the first or last position, then applying
Inner, and then transposing the result back.

In many applications of tensors, you need to insert signs to implement antisymmetry. The
function Signature@8i1 , i2 , <D, which gives the signature of a permutation, is often useful for
this purpose.
Mathematics and Algorithms 185

Outer@ f ,t1 ,t2 ,D form a generalized outer product by combining the lowest-
level elements of t1 , t2 ,
Outer@ f ,t1 ,t2 ,,nD treat only sublists at level n as separate elements
Outer@ f ,t1 ,t2 ,,n1 ,n2 ,D treat only sublists at level ni in ti as separate elements

Inner@ f ,t1 ,t2 ,gD form a generalized inner product using the lowest-level
elements of t1
Inner@ f ,t1 ,t2 ,g,nD contract index n of the first tensor with the first index of
the second tensor

Treating only certain sublists in tensors as separate elements.

Here every single symbol is treated as a separate element.

In[28]:= Outer@f, 88i, j<, 8k, l<<, 8x, y<D
Out[28]= 888f@i, xD, f@i, yD<, 8f@j, xD, f@j, yD<<, 88f@k, xD, f@k, yD<, 8f@l, xD, f@l, yD<<<

But here only sublists at level 1 are treated as separate elements.

In[29]:= Outer@f, 88i, j<, 8k, l<<, 8x, y<, 1D
Out[29]= 88f@8i, j<, xD, f@8i, j<, yD<, 8f@8k, l<, xD, f@8k, l<, yD<<

ArrayFlatten@t,rD create a flat rank r tensor from a rank r tensor of rank r

tensors
ArrayFlatten@tD flatten a matrix of matrices (equivalent to
ArrayFlatten@t, 2D)

Flattening block tensors.

Here is a block matrix (a matrix of matrices that can be viewed as blocks that fit edge to edge
within a larger matrix).
In[30]:= TableForm@88 881, 2<, 84, 5<<, 883<, 86<< <, 8 887, 8<<, 889<< <<D
1 2 3
Out[30]//TableForm= 4 5 6
7 8 9

Here is the matrix formed by piecing the blocks together.

In[31]:= TableForm@ArrayFlatten@%DD
1 2 3
Out[31]//TableForm= 4 5 6
7 8 9

Sparse Arrays: Linear Algebra

186 Mathematics and Algorithms

Sparse Arrays: Linear Algebra

Many large-scale applications of linear algebra involve matrices that have many elements, but
comparatively few that are nonzero. You can represent such sparse matrices efficiently in
Mathematica using SparseArray objects, as discussed in "Sparse Arrays: Manipulating Lists".
SparseArray objects work by having lists of rules that specify where nonzero values appear.

SparseArray @listD a SparseArray version of an ordinary list

SparseArray @88i1 , j1 <->v1 ,8i2 , j2 <->v2 ,<,8m,n<D
an mn sparse array with element 8ik , jk < having value vk

SparseArray @88i1 , j1 <,8i2 , j2 <,<->8v1 ,v2 ,<,8m,n<D

the same sparse array
Normal@arrayD the ordinary list corresponding to a SparseArray

Specifying sparse arrays.

As discussed in "Sparse Arrays: Manipulating Lists", you can use patterns to specify collections
of elements in sparse arrays. You can also have sparse arrays that correspond to tensors of any
rank.

This makes a 5050 sparse numerical matrix, with 148 nonzero elements.
In[1]:= m = SparseArray@8830, _< -> 11.5, 8_, 30< -> 21.5, 8i_, i_< -> i<, 850, 50<D
Out[1]= SparseArray@<148>, 850, 50<D

This shows a visual representation of the matrix elements.

In[2]:= ArrayPlot@mD

Out[2]=

Here are the four largest eigenvalues of the matrix.

Mathematics and Algorithms 187

Here are the four largest eigenvalues of the matrix.

In[3]:= Eigenvalues@m, 4D
Out[3]= 8129.846, -92.6878, 49.7867, 48.7478<

Dot gives a SparseArray result.

In[4]:= m.m
Out[4]= SparseArray@<2500>, 850, 50<D

You can extract parts just like in an ordinary array.

In[5]:= %@@20, 20DD
Out[5]= 647.25

You can apply most standard structural operations directly to SparseArray objects, just as you
would to ordinary lists. When the results are sparse, they typically return SparseArray objects.

Dimensions@mD the dimensions of an array

ArrayRules@mD the rules for nonzero elements in an array
m@@i, jDD element i, j

m@@iDD the ith row

mAAAll, jEE the ith column

m@@i, jDD=v reset element i, j

A few structural operations that can be done directly on SparseArray objects.

This gives the first column of m. It has only 2 nonzero elements.

In[6]:= m@@All, 1DD
Out[6]= SparseArray@<2>, 850<D

This adds 3 to each element in the first column of m.

In[7]:= m@@All, 1DD = 3 + m@@All, 1DD
Out[7]= SparseArray@<2>, 850<, 3D

Now all the elements in the first column are nonzero.

In[8]:= m@@All, 1DD
Out[8]= SparseArray@<50>, 850<D

This gives the rules for the nonzero elements on the second row.
188 Mathematics and Algorithms

This gives the rules for the nonzero elements on the second row.
In[9]:= ArrayRules@m@@2DDD
Out[9]= 881< 3, 82< 2, 830< 21.5, 8_< 0<

SparseArray @rulesD generate a sparse array from rules

CoefficientArrays@8eqns1 ,eqns2 ,<,8x1 ,x2 ,<D
get arrays of coefficients from equations
Import@" file.mtx"D import a sparse array from a file

Typical ways to get sparse arrays.

This generates a tridiagonal random matrix.

In[10]:= SparseArray@8i_, j_< ; Abs@i - jD <= 1 :> RandomReal@D, 8100, 100<D
Out[10]= SparseArray@<298>, 8100, 100<D

Even the tenth power of the matrix is still fairly sparse.

In[11]:= MatrixPower@%, 10D
Out[11]= SparseArray@<1990>, 8100, 100<D

This extracts the coefficients as sparse arrays.

In[12]:= s = CoefficientArrays@8c + x - z == 0, x + 2 y + z == 0<, 8x, y, z<D
Out[12]= 8SparseArray@<1>, 82<D, SparseArray@<5>, 82, 3<D<

Here are the corresponding ordinary arrays.

In[13]:= Normal@%D
Out[13]= 88c, 0<, 881, 0, -1<, 81, 2, 1<<<

This reproduces the original forms.

In[14]:= s@@1DD + s@@2DD.8x, y, z<
Out[14]= 8c + x - z, x + 2 y + z<

CoefficientArrays can handle general polynomial equations.

In[15]:= s = CoefficientArrays@8c + x ^ 2 - z == 0, x ^ 2 + 2 y + z ^ 2 == 0<, 8x, y, z<D
Out[15]= 8SparseArray@<1>, 82<D, SparseArray@<2>, 82, 3<D, SparseArray@<3>, 82, 3, 3<D<
Mathematics and Algorithms 189

The coefficients of the quadratic part are given in a rank 3 tensor.

In[16]:= Normal@%D
Out[16]= 88c, 0<, 880, 0, -1<, 80, 2, 0<<,
8881, 0, 0<, 80, 0, 0<, 80, 0, 0<<, 881, 0, 0<, 80, 0, 0<, 80, 0, 1<<<<

This reproduces the original forms.

In[17]:= s@@1DD + s@@2DD.8x, y, z< + s@@3DD.8x, y, z<.8x, y, z<

Out[17]= 9c + x2 - z, x2 + 2 y + z2 =

For machine-precision numerical sparse matrices, Mathematica supports standard file formats
such as Matrix Market (.mtx) and Harwell|Boeing. You can import and export matrices in these
formats using Import and Export.

Series, Limits and Residues

Sums and Products

xi
This constructs the sum 7i=1 .
i

In[1]:= Sum@x ^ i i, 8i, 1, 7<D

x2 x3 x4 x5 x6 x7
Out[1]= x+ + + + + +
2 3 4 5 6 7

You can leave out the lower limit if it is equal to 1.

In[2]:= Sum@x ^ i i, 8i, 7<D
x2 x3 x4 x5 x6 x7
Out[2]= x+ + + + + +
2 3 4 5 6 7

This makes i increase in steps of 2, so that only odd-numbered values are included.
In[3]:= Sum@x ^ i i, 8i, 1, 5, 2<D
x3 x5
Out[3]= x+ +
3 5
190 Mathematics and Algorithms

Products work just like sums.

In[4]:= Product@x + i, 8i, 1, 4<D
Out[4]= H1 + xL H2 + xL H3 + xL H4 + xL

imax
Sum @ f ,8i,imin ,imax <D the sum i=i f
min

Sum @ f ,8i,imin ,imax ,di<D the sum with i increasing in steps of di

Sum @ f ,8i,imin ,imax <,8 j, jmin , jmax <D
imax jmax
the nested sum i=i j= j f
min min

imax
Product@ f ,8i,imin ,imax <D the product i=i f
min

Sums and products.

This sum is computed symbolically as a function of n.

In[5]:= Sum@i ^ 2, 8i, 1, n<D
1
Out[5]= n H1 + nL H1 + 2 nL
6

Mathematica can also give an exact result for this infinite sum.
In[6]:= Sum@1 i ^ 4, 8i, 1, Infinity<D
p4
Out[6]=
90

As with integrals, simple sums can lead to complicated results.

In[7]:= Sum@x ^ Hi Hi + 1LL, 8i, 1, Infinity<D

-2 x14 + EllipticTheta@2, 0, xD
Out[7]=
2 x14

This sum cannot be evaluated exactly using standard mathematical functions.

In[8]:= Sum@1 Hi ! + H2 iL !L, 8i, 1, Infinity<D
1
Out[8]=
i=1 i ! + H2 iL !

You can nevertheless find a numerical approximation to the result.

In[9]:= N@%D
Out[9]= 0.373197

Mathematica also has a notation for multiple sums and products.

Sum @ f , 8i, imin , imax <, 8 j, jmin , jmax <D represents a sum over i and j, which would be written in
Mathematics and Algorithms 191

Mathematica also has a notation for multiple sums and products.

Sum @ f , 8i, imin , imax <, 8 j, jmin , jmax <D represents a sum over i and j, which would be written in
i j
standard mathematical notation as i=i
max
min
j= jmin f . Notice that in Mathematica notation, as in
max

standard mathematical notation, the range of the outermost variable is given first.

3 i
This is the multiple sum i=1 j=1 xi y j . Notice that the outermost sum over i is given first, just
as in the mathematical notation.
In[10]:= Sum@x ^ i y ^ j, 8i, 1, 3<, 8j, 1, i<D

Out[10]= x y + x2 y + x3 y + x2 y2 + x3 y2 + x3 y3

The way the ranges of variables are specified in Sum and Product is an example of the rather
general iterator notation that Mathematica uses. You will see this notation again when we
discuss generating tables and lists using Table ("Making Tables of Values"), and when we
describe Do loops ("Repetitive Operations").

8imax < iterate imax times, without incrementing any variables

8i,imax < i goes from 1 to imax in steps of 1

8i,imin ,imax < i goes from imin to imax in steps of 1
8i,imin ,imax ,di< i goes from imin to imax in steps of di
8i,imin ,imax <,8 j, jmin , jmax <, i goes from imin to imax , and for each such value, j goes
from jmin to jmax , etc.

Mathematica iterator notation.

Power Series
The mathematical operations we have discussed so far are exact. Given precise input, their
results are exact formulas.

In many situations, however, you do not need an exact result. It may be quite sufficient, for
example, to find an approximate formula that is valid, say, when the quantity x is small.

This gives a power series approximation to H1 + xLn for x close to 0, up to terms of order x3 .
In[1]:= Series@H1 + xL ^ n, 8x, 0, 3<D
1 1
Out[1]= 1 + n x + H-1 + nL n x2 + H-2 + nL H-1 + nL n x3 + O@xD4
2 6

Mathematica knows the power series expansions for many mathematical functions.
192 Mathematics and Algorithms

Mathematica knows the power series expansions for many mathematical functions.
In[2]:= Series@Exp@- a tD H1 + Sin@2 tDL, 8t, 0, 4<D
a2 4 a3 1
Out[2]= 1 + H2 - aL t + -2 a + t2 + - + a2 - t3 + I32 a - 8 a3 + a4 M t4 + O@tD5
2 3 6 24

If you give it a function that it does not know, Series writes out the power series in terms of
derivatives.
In[3]:= Series@1 + f@tD, 8t, 0, 3<D
1 1

Out[3]= H1 + f@0DL + f @0D t + f @0D t2 + fH3L @0D t3 + O@tD4
2 6

Power series are approximate formulas that play much the same role with respect to algebraic
expressions as approximate numbers play with respect to numerical expressions. Mathematica
allows you to perform operations on power series, in all cases maintaining the appropriate order
or "degree of precision" for the resulting power series.

Here is a simple power series, accurate to order x5 .

In[4]:= Series@Exp@xD, 8x, 0, 5<D
x2 x3 x4 x5
Out[4]= 1 + x + + + + + O@xD6
2 6 24 120

When you do operations on a power series, the result is computed only to the appropriate order
in x.
In[5]:= % ^ 2 H1 + %L
13 x2 35 x3 97 x4 55 x5
Out[5]= 2 + 5 x + + + + + O@xD6
2 6 24 24

This turns the power series back into an ordinary expression.

In[6]:= Normal@%D
13 x2 35 x3 97 x4 55 x5
Out[6]= 2 + 5 x + + + +
2 6 24 24

Now the square is computed exactly.

In[7]:= %^2
2
13 x2 35 x3 97 x4 55 x5
Out[7]= 2+5x+ + + +
2 6 24 24
Mathematics and Algorithms 193

Applying Expand gives a result with 11 terms.

In[8]:= Expand@%D
265 x3 467 x4 1505 x5 7883 x6 1385 x7 24 809 x8 5335 x9 3025 x10
2
Out[8]= 4 + 20 x + 51 x + + + + + + + +
3 4 12 72 18 576 288 576

Series@expr,8x,x0 ,n<D find the power series expansion of expr about the point
x = x0 to at most nth order
Normal@seriesD truncate a power series to give an ordinary expression

Power series operations.

Making Power Series Expansions

Series@expr,8x,x0 ,n<D find the power series expansion of expr about the point
x = x0 to order at most Hx - x0 Ln
SeriesAexpr,9x,x0 ,nx =,9y,y0 ,ny =E
find series expansions with respect to y then x

Functions for creating power series.

Here is the power series expansion for expHxL about the point x = 0 to order x4 .
In[1]:= Series@Exp@xD, 8x, 0, 4<D
x2 x3 x4
Out[1]= 1 + x + + + + O@xD5
2 6 24

Here is the series expansion of expHxL about the point x = 1.

In[2]:= Series@Exp@xD, 8x, 1, 4<D
1 1 1
Out[2]= + Hx - 1L + Hx - 1L2 + Hx - 1L3 + Hx - 1L4 + O@x - 1D5
2 6 24

If Mathematica does not know the series expansion of a particular function, it writes the result
symbolically in terms of derivatives.
In[3]:= Series@f@xD, 8x, 0, 3<D
1 1

Out[3]= f@0D + f @0D x + f @0D x2 + fH3L @0D x3 + O@xD4
2 6

In mathematical terms, Series can be viewed as a way of constructing Taylor series for func-
tions.
194 Mathematics and Algorithms

In mathematical terms, Series can be viewed as a way of constructing Taylor series for func-
tions.

The standard formula for the Taylor series expansion about the point x = x0 of a function gHxL
Ix-x0 Mk
with kth derivative gHkL HxL is gHxL = HkL
k=0 g Hx0 L k!
. Whenever this formula applies, it gives the

same results as Series. (For common functions, Series nevertheless internally uses somewhat
more efficient algorithms.)

Series can also generate some power series that involve fractional and negative powers, not
directly covered by the standard Taylor series formula.

Here is a power series that contains negative powers of x.

In[4]:= Series@Exp@xD x ^ 2, 8x, 0, 4<D
1 1 1 x x2 x3 x4
Out[4]= + + + + + + + O@xD5
x2 x 2 6 24 120 720

Here is a power series involving fractional powers of x.

In[5]:= Series@Exp@Sqrt@xDD, 8x, 0, 2<D

x x32 x2
Out[5]= 1 + x + + + + O@xD52
2 6 24

Series can also handle series that involve logarithmic terms.

In[6]:= Series@Exp@2 xD Log@xD, 8x, 0, 2<D
2 3
Out[6]= Log@xD + 2 Log@xD x + 2 Log@xD x + O@xD

There are, of course, mathematical functions for which no standard power series exist. Mathe-
matica recognizes many such cases.

1
Series sees that expJ x N has an essential singularity at x = 0, and does not produce a power
series.
In[7]:= Series@Exp@1 xD, 8x, 0, 2<D
1

Out[7]= x
Mathematics and Algorithms 195

1
Series can nevertheless give you the power series for expJ x N about the point x = .
In[8]:= Series@Exp@1 xD, 8x, Infinity, 3<D
2 3
1 1 1 1 1 1 4
Out[8]= 1 + + + + OB F
x 2 x 6 x x

Especially when negative powers occur, there is some subtlety in exactly how many terms of a
particular power series the function Series will generate.

One way to understand what happens is to think of the analogy between power series taken to
a certain order, and real numbers taken to a certain precision. Power series are "approximate
formulas" in much the same sense as finite-precision real numbers are approximate numbers.

The procedure that Series follows in constructing a power series is largely analogous to the
procedure that N follows in constructing a real-number approximation. Both functions effectively
start by replacing the smallest pieces of your expression by finite-order, or finite-precision,
approximations, and then evaluating the resulting expression. If there are, for example, cancella-
tions, this procedure may give a final result whose order or precision is less than the order or
precision that you originally asked for. Like N, however, Series has some ability to retry its
computations so as to get results to the order you ask for. In cases where it does not succeed,
you can usually still get results to a particular order by asking for a higher order than you need.

Series compensates for cancellations in this computation, and succeeds in giving you a result
to order x3 .
In[9]:= Series@Sin@xD x ^ 2, 8x, 0, 3<D
1 x x3
Out[9]= - + + O@xD4
x 6 120

When you make a power series expansion in a variable x, Mathematica assumes that all objects
that do not explicitly contain x are in fact independent of x. Series thus does partial derivatives
(effectively using D) to build up Taylor series.

Both a and n are assumed to be independent of x.

In[10]:= Series@Ha + xL ^ n, 8x, 0, 2<D
1
-1+n
n
Out[10]= a + a nx+ a-2+n H-1 + nL n x2 + O@xD3
2
196 Mathematics and Algorithms

a@xD is now given as an explicit function of x.

In[11]:= Series@Ha@xD + xL ^ n, 8x, 0, 2<D
1 1
-1+n
n
Out[11]= a@0D + n a@0D H1 + a @0DL x + H-1 + nL n a@0D-2+n H1 + a @0DL2 + n a@0D-1+n a @0D x2 + O@xD3
2 2

You can use Series to generate power series in a sequence of different variables. Series works
like Integrate, Sum and so on, and expands first with respect to the last variable you specify.

Series performs a series expansion successively with respect to each variable. The result in
this case is a series in x, whose coefficients are series in y.
In[12]:= Series@Exp@x yD, 8x, 0, 3<, 8y, 0, 3<D
y2 y3
4
Out[12]= 1 + Iy + O@yD M x + + O@yD4 x2 + + O@yD4 x3 + O@xD4
2 6

The Representation of Power Series

Power series are represented in Mathematica as SeriesData objects.

The power series is printed out as a sum of terms, ending with O@xD raised to a power.
In[1]:= Series@Cos@xD, 8x, 0, 4<D
x2 x4
Out[1]= 1 - + + O@xD5
2 24

Internally, however, the series is stored as a SeriesData object.

In[2]:= InputForm@%D
Out[2]//InputForm= SeriesData[x, 0, {1, 0, -1/2, 0, 1/24}, 0, 5, 1]

By using SeriesData objects, rather than ordinary expressions, to represent power series,
Mathematica can keep track of the order and expansion point, and do operations on the power
series appropriately. You should not normally need to know the internal structure of
SeriesData objects.

You can recognize a power series that is printed out in standard output form by the presence of
an O@xD term. This term mimics the standard mathematical notation OHxL, and represents omit-
ted terms of order x. For various reasons of consistency, Mathematica uses the notation O@xD ^ n
for omitted terms of order xn , corresponding to the mathematical notation OHxLn , rather than the
slightly more familiar, though equivalent, form OHxn L.

Any time that an object like O@xD appears in a sum of terms, Mathematica will in fact convert
the whole sum into a power series.
Mathematics and Algorithms 197

Any time that an object like O@xD appears in a sum of terms, Mathematica will in fact convert
the whole sum into a power series.

The presence of O@xD makes Mathematica convert the whole sum to a power series.
In[3]:= a x + Exp@xD + O@xD ^ 3
x2
Out[3]= 1 + H1 + aL x + + O@xD3
2

Series objects can involve fractional powers.

In[4]:= Series@Sqrt@1 - x ^ 3D, 8x, 1, 5<D

1 Hx - 1L52 Hx - 1L72 11 Hx - 1L92

Out[4]= 3 x-1 + 3 Hx - 1L32 + +- + + O@x - 1D112
2 8 3 16 3 384 3

Here is the series' internal representation.

In[5]:= % InputForm
Out[5]//InputForm= SeriesData[x, 1, {I*Sqrt[3], 0, (I/2)*Sqrt[3], 0, (I/8)/Sqrt[3], 0, (-I/16)/Sqrt[3], 0, ((11*I)/384)/Sqrt[
2]

Series can involve logarithmic terms.

In[6]:= Series@x ^ x, 8x, 0, 4<D
1 1 1
Out[6]= 1 + Log@xD x + Log@xD2 x2 + Log@xD3 x3 + Log@xD4 x4 + O@xD5
2 6 24

The logarithmic factors appear explicitly inside the SeriesData coefficient list.
In[7]:= % InputForm
Out[7]//InputForm= SeriesData[x, 0, {1, Log[x], Log[x]^2/2, Log[x]^3/6, Log[x]^4/24}, 0, 5, 1]

Operations on Power Series

Mathematica allows you to perform many operations on power series. In all cases, Mathematica
gives results only to as many terms as can be justified from the accuracy of your input.

Here is a power series accurate to fourth order in x.

In[1]:= Series@Exp@xD, 8x, 0, 4<D
x2 x3 x4
Out[1]= 1 + x + + + + O@xD5
2 6 24

When you square the power series, you get another power series, also accurate to fourth order.
198 Mathematics and Algorithms

When you square the power series, you get another power series, also accurate to fourth order.
In[2]:= %^2
4 x3 2 x4
2
Out[2]= 1 + 2 x + 2 x + + + O@xD5
3 3

Taking the logarithm gives you the result 2 x, but only to order x4 .
In[3]:= Log@%D
5
Out[3]= 2 x + O@xD

Mathematica keeps track of the orders of power series in much the same way as it keeps track
of the precision of approximate real numbers. Just as with numerical calculations, there are
operations on power series which can increase, or decrease, the precision (or order) of your
results.

Here is a power series accurate to order x10 .

In[4]:= Series@Cos@xD, 8x, 0, 10<D
x2 x4 x6 x8 x10
Out[4]= 1 - + - + - + O@xD11
2 24 720 40 320 3 628 800

This gives a power series that is accurate only to order x6 .

In[5]:= 1 H1 - %L
2 1 x2 x4 x6
Out[5]= + + + + + O@xD7
x2 6 120 3024 86 400

Mathematica also allows you to do calculus with power series.

Here is a power series for tan HxL.

In[6]:= Series@Tan@xD, 8x, 0, 10<D
x3 2 x5 17 x7 62 x9
Out[6]= x + + + + + O@xD11
3 15 315 2835

Here is its derivative with respect to x.

In[7]:= D@%, xD
2 x4 17 x6 62 x8
2
Out[7]= 1 + x + + + + O@xD10
3 45 315
Mathematics and Algorithms 199

Integrating with respect to x gives back the original power series.

In[8]:= Integrate@%, xD
x3 2 x5 17 x7 62 x9
Out[8]= x + + + + + O@xD11
3 15 315 2835

When you perform an operation that involves both a normal expression and a power series,
Mathematica "absorbs" the normal expression into the power series whenever possible.

The 1 is automatically absorbed into the power series.

In[9]:= 1 + Series@Exp@xD, 8x, 0, 4<D
x2 x3 x4
Out[9]= 2 + x + + + + O@xD5
2 6 24

The x ^ 2 is also absorbed into the power series.

In[10]:= % + x^2
3 x2 x3 x4
Out[10]= 2 + x + + + + O@xD5
2 6 24

If you add Sin@xD, Mathematica generates the appropriate power series for Sin@xD, and
combines it with the power series you have.
In[11]:= % + Sin@xD
3 x2 x4
Out[11]= 2 + 2 x + + + O@xD5
2 24

Mathematica also absorbs expressions that multiply power series. The symbol a is assumed to
be independent of x.
In[12]:= Ha + xL % ^ 2
29 a
2 3
Out[12]= 4 a + H4 + 8 aL x + H8 + 10 aL x + H10 + 6 aL x + 6 + x4 + O@xD5
12

Mathematica knows how to apply a wide variety of functions to power series. However, if you
apply an arbitrary function to a power series, it is impossible for Mathematica to give you any-
thing but a symbolic result.

Mathematica does not know how to apply the function f to a power series, so it just leaves the
symbolic result.
In[13]:= f@Series@Exp@xD, 8x, 0, 3<DD
x2 x3
Out[13]= fB1 + x + + + O@xD4 F
2 6
200 Mathematics and Algorithms

Composition and Inversion of Power Series

When you manipulate power series, it is sometimes convenient to think of the series as repre-
senting functions, which you can, for example, compose or invert.

ComposeSeries@series1 ,series2 ,D compose power series

InverseSeries@series,xD invert a power series

Composition and inversion of power series.

Here is the power series for expHxL to order x5 .

In[1]:= Series@Exp@xD, 8x, 0, 5<D
x2 x3 x4 x5
Out[1]= 1 + x + + + + + O@xD6
2 6 24 120

This replaces the variable x in the power series for expHxL by a power series for sinHxL.
In[2]:= ComposeSeries@%, Series@Sin@xD, 8x, 0, 5<DD
x2 x4 x5
Out[2]= 1 + x + - - + O@xD6
2 8 15

The result is the power series for expHsinHxLL.

In[3]:= Series@Exp@Sin@xDD, 8x, 0, 5<D
x2 x4 x5
Out[3]= 1 + x + - - + O@xD6
2 8 15

If you have a power series for a function f HyL, then it is often possible to get a power series
approximation to the solution for y in the equation f HyL = x. This power series effectively gives
the inverse function f -1 HxL such that f I f -1 HxLM = x. The operation of finding the power series for an

inverse function is sometimes known as reversion of power series.

Here is the series for sinHyL.

In[4]:= Series@Sin@yD, 8y, 0, 5<D
y3 y5
Out[4]= y - + + O@yD6
6 120
Mathematics and Algorithms 201

Inverting the series gives the series for sin-1 HxL.

In[5]:= InverseSeries@%, xD
x3 3 x5
Out[5]= x + + + O@xD6
6 40

This agrees with the direct series for sin-1 HxL.

In[6]:= Series@ArcSin@xD, 8x, 0, 5<D
x3 3 x5
Out[6]= x + + + O@xD6
6 40

Composing the series with its inverse gives the identity function.
In[7]:= ComposeSeries@%, %%%D
6
Out[7]= y + O@yD

Converting Power Series to Normal Expressions

Normal@exprD convert a power series to a normal expression

Converting power series to normal expressions.

Power series in Mathematica are represented in a special internal form, which keeps track of
such attributes as their expansion order.

For some purposes, you may want to convert power series to normal expressions. From a
mathematical point of view, this corresponds to truncating the power series, and assuming that
all higher-order terms are zero.

This generates a power series, with four terms.

In[1]:= t = Series@ArcTan@xD, 8x, 0, 8<D
x3 x5 x7
Out[1]= x - + - + O@xD9
3 5 7

Squaring the power series gives you another power series, with the appropriate number of
terms.
In[2]:= t^2
2 x4 23 x6 44 x8
2
Out[2]= x - + - + O@xD10
3 45 105

Normal truncates the power series, giving a normal expression.

202 Mathematics and Algorithms

Normal truncates the power series, giving a normal expression.

In[3]:= Normal@%D
2 x4 23 x6 44 x8
2
Out[3]= x - + -
3 45 105

You can now apply standard algebraic operations.

In[4]:= Factor@%D
1
Out[4]= - x2 I-315 + 210 x2 - 161 x4 + 132 x6 M
315

SeriesCoefficient@series,nD give the coefficient of the nth -order term in a power series

Extracting coefficients of terms in power series.

This gives the coefficient of x7 in the original power series.

In[5]:= SeriesCoefficient@t, 7D
1
Out[5]= -
7

2
This gives the coefficient for the term xn in the Taylor expansion of the function ex about zero.
In[6]:= SeriesCoefficient@E ^ x ^ 2, 8x, 0, n<D
KroneckerDelta@Mod@n, 2DD
Out[6]=
n
!
2

Solving Equations Involving Power Series

LogicalExpand@series1 ==series2 D give the equations obtained by equating corresponding

coefficients in the power series
Solve@series1 ==series2 ,8a1 ,a2 ,<D solve for coefficients in power series

Solving equations involving power series.

Here is a power series.

In[1]:= y = 1 + Sum@a@iD x ^ i, 8i, 3<D + O@xD ^ 4
2 3 4
Out[1]= 1 + a@1D x + a@2D x + a@3D x + O@xD
Mathematics and Algorithms 203

This gives an equation involving the power series.

In[2]:= D@y, xD ^ 2 - y == x
2 2 2 3
Out[2]= I-1 + a@1D M + H-a@1D + 4 a@1D a@2DL x + I-a@2D + 4 a@2D + 6 a@1D a@3DM x + O@xD x

LogicalExpand generates a sequence of equations for each power of x.

In[3]:= LogicalExpand@%D
2 2
Out[3]= -1 + a@1D 0 && -1 - a@1D + 4 a@1D a@2D 0 && -a@2D + 4 a@2D + 6 a@1D a@3D 0

This solves the equations for the coefficients a@iD. You can also feed equations involving power
series directly to Solve .
In[4]:= Solve@%D
1 1
Out[4]= ::a@3D - , a@1D 1, a@2D >, 8a@3D 0, a@1D -1, a@2D 0<>
12 2

Some equations involving power series can also be solved using the InverseSeries function
discussed in "Composition and Inversion of Power Series".

Summation of Series

Sum @expr,8n,nmin ,nmax <D find the sum of expr as n goes from nmin to nmax

Evaluating sums.

Mathematica recognizes this as the power series expansion of ex .

In[1]:= Sum@x ^ n n !, 8n, 0, Infinity<D
x
Out[1]=

This sum comes out in terms of a Bessel function.

In[2]:= Sum@x ^ n Hn ! ^ 2L, 8n, 0, Infinity<D

Out[2]= BesselIB0, 2 x F

Here is another sum that can be done in terms of common special functions.
In[3]:= Sum@n ! x ^ n H2 nL !, 8n, 1, Infinity<D

1 x
Out[3]= x4 p x ErfB F
2 2

Generalized hypergeometric functions are not uncommon in sums.

204 Mathematics and Algorithms

Generalized hypergeometric functions are not uncommon in sums.

In[4]:= Sum@x ^ n Hn ! ^ 4L, 8n, 0, Infinity<D
Out[4]= HypergeometricPFQ@8<, 81, 1, 1<, xD

There are many analogies between sums and integrals. And just as it is possible to have indefi-
nite integrals, so indefinite sums can be set up by using symbolic variables as upper limits.

This is effectively an indefinite sum.

In[5]:= Sum@k, 8k, 0, n<D
1
Out[5]= n H1 + nL
2

This sum comes out in terms of incomplete gamma functions.

In[6]:= Sum@x ^ k k !, 8k, 0, n<D
x H1 + nL Gamma@1 + n, xD
Out[6]=
Gamma@2 + nD

This sum involves polygamma functions.

In[7]:= Sum@1 Hk + 1L ^ 4, 8k, 0, n<D
p4 1
Out[7]= - PolyGamma@3, 2 + nD
90 6

Taking the difference between results for successive values of n gives back the original sum-
mand.
In[8]:= FullSimplify@% - H% . n -> n - 1LD
1
Out[8]=
H1 + nL4

Mathematica can do essentially all sums that are found in books of tables. Just as with indefi-
nite integrals, indefinite sums of expressions involving simple functions tend to give answers
that involve more complicated functions. Definite sums, like definite integrals, often, however,
come out in terms of simpler functions.

This indefinite sum gives a quite complicated result.

In[9]:= Sum@Binomial@2 k, kD 3 ^ H2 kL, 8k, 0, n<D
9 -1-n 3 3 4
3 J N GammaB + nF Hypergeometric2F1B1, + n, 2 + n, F
4 2 2 9
Out[9]= -
5 p Gamma@2 + nD

The definite form is much simpler.

Mathematics and Algorithms 205

The definite form is much simpler.

In[10]:= Sum@Binomial@2 k, kD 3 ^ H2 kL, 8k, 0, Infinity<D
3
Out[10]=
5

Here is a slightly more complicated definite sum.

In[11]:= Sum@PolyGamma@kD k ^ 2, 8k, 1, Infinity<D
1
Out[11]= I-EulerGamma p2 + 6 Zeta@3DM
6

Solving Recurrence Equations

If you represent the nth term in a sequence as a@nD, you can use a recurrence equation to specify
how it is related to other terms in the sequence.

RSolve takes recurrence equations and solves them to get explicit formulas for a@nD.

This solves a simple recurrence equation.

In[1]:= RSolve@8a@nD == 2 a@n - 1D, a@1D == 1<, a@nD, nD
-1+n
Out[1]= 99a@nD 2 ==

This takes the solution and makes an explicit table of the first ten a@nD.
In[2]:= Table@a@nD . First@%D, 8n, 10<D
Out[2]= 81, 2, 4, 8, 16, 32, 64, 128, 256, 512<

RSolve@eqn,a@nD,nD solve a recurrence equation

Solving a recurrence equation.

This solves a recurrence equation for a geometric series.

In[3]:= RSolve@8a@nD == r a@n - 1D + 1, a@1D == 1<, a@nD, nD
-1 + rn
Out[3]= ::a@nD >>
-1 + r
206 Mathematics and Algorithms

This gives the same result.

In[4]:= RSolve@8a@n + 1D == r a@nD + 1, a@1D == 1<, a@nD, nD
-1 + rn
Out[4]= ::a@nD >>
-1 + r

This gives an algebraic solution to a recurrence equation.

In[5]:= RSolve@8a@nD == 4 a@n - 1D + a@n - 2D<, a@nD, nD
n n
Out[5]= ::a@nD 2 - 5 C@1D + 2 + 5 C@2D>>

This solves the Fibonacci recurrence equation.

In[6]:= RSolve@8a@nD == a@n - 1D + a@n - 2D, a@1D == a@2D == 1<, a@nD, nD
Out[6]= 88a@nD Fibonacci@nD<<

RSolve can be thought of as a discrete analog of DSolve. Many of the same functions gener-
ated in solving differential equations also appear in finding symbolic solutions to recurrence
equations.

This generates a gamma function, which generalizes the factorial.

In[7]:= RSolve@8a@nD == n a@n - 1D, a@1D == 1<, a@nD, nD
Out[7]= 88a@nD Gamma@1 + nD<<

This second-order recurrence equation comes out in terms of Bessel functions.

In[8]:= RSolve@8a@n + 1D == n a@nD + a@n - 1D, a@1D == 0, a@2D == 1<, a@nD, nD
BesselI@n, -2D BesselK@1, 2D + BesselI@1, 2D BesselK@n, 2D
Out[8]= ::a@nD >>
BesselI@2, 2D BesselK@1, 2D + BesselI@1, 2D BesselK@2, 2D

RSolve does not require you to specify explicit values for terms such as a@1D. Like DSolve, it
automatically introduces undetermined constants C@iD to give a general solution.

This gives a general solution with one undetermined constant.

In[9]:= RSolve@a@nD == n a@n - 1D, a@nD, nD
Out[9]= 88a@nD C@1D Gamma@1 + nD<<
Mathematics and Algorithms 207

RSolve can solve equations that do not depend only linearly on a@nD. For nonlinear equations,
however, there are sometimes several distinct solutions that must be given. Just as for differen-
tial equations, it is a difficult matter to find symbolic solutions to recurrence equations, and
standard mathematical functions only cover a limited set of cases.

Here is the general solution to a nonlinear recurrence equation.

In[10]:= RSolve@8a@nD == a@n + 1D a@n - 1D<, a@nD, nD
np np
C@1D CosB F+C@2D SinB F
Out[10]= ::a@nD 3 3 >>

This gives two distinct solutions.

In[11]:= RSolve@a@nD == Ha@n + 1D a@n - 1DL ^ 2, a@nD, nD
2p
C@2D CosBn ArcTanB 15 FF+C@1D SinBn ArcTanB 15 FF +C@2D CosBn ArcTanB 15 FF+C@1D SinBn ArcTanB 15 FF
Out[11]= ::a@nD >, :a@nD 3 >>

RSolve can solve not only ordinary difference equations in which the arguments of a differ by
integers, but also q-difference equations in which the arguments of a are related by multiplica-
tive factors.

This solves the q-difference analog of the factorial equation.

In[12]:= RSolve@a@q nD == n a@nD, a@nD, nD
1 Log@nD
K-1+ O
Out[12]= ::a@nD n 2 Log@qD C@1D>>

Here is a second-order q-difference equation.

In[13]:= RSolve@a@nD == a@q nD + a@n qD, a@nD, nD
p Log@nD p Log@nD
Out[13]= ::a@nD C@1D CosB F + C@2D SinB F>>
3 Log@qD 3 Log@qD

RSolve@8eqn1 ,eqn2 ,<,8a1 @nD,a2 @nD,<,nD

solve a coupled system of recurrence equations

Solving systems of recurrence equations.

This solves a system of two coupled recurrence equations.

In[14]:= RSolve@8a@nD == b@n - 1D + n, b@nD == a@n - 1D - n, a@1D == b@1D == 1<, 8a@nD, b@nD<, nD
1 1
Out[14]= ::a@nD I4 + 3 H-1Ln + H-1L2 n + 2 H-1L2 n nM, b@nD I4 - 3 H-1Ln - H-1L2 n - 2 H-1L2 n nM>>
4 4
208 Mathematics and Algorithms

RSolve@eqns,a@n1 ,n2 ,D,8n1 ,n2 ,<D

solve partial recurrence equations

Solving partial recurrence equations.

Just as one can set up partial differential equations that involve functions of several variables,
so one can also set up partial recurrence equations that involve multidimensional sequences.
Just as in the differential equations case, general solutions to partial recurrence equations can
involve undetermined functions.

This gives the general solution to a simple partial recurrence equation.

In[15]:= RSolve@a@i + 1, j + 1D == i j a@i, jD, a@i, jD, 8i, j<D
Gamma@iD Gamma@jD C@1D@i - jD
Out[15]= ::a@i, jD >>
Gamma@1 - i + jD

Finding Limits
In doing many kinds of calculations, you need to evaluate expressions when variables take on
particular values. In many cases, you can do this simply by applying transformation rules for
the variables using the . operator.

You can get the value of cos Ix2 M at 0 just by explicitly replacing x with 0, and then evaluating
the result.
In[1]:= Cos@x ^ 2D . x -> 0
Out[1]= 1

In some cases, however, you have to be more careful.

sin HxL
Consider, for example, finding the value of the expression x
when x = 0. If you simply replace
0 sin HxL
x by 0 in this expression, you get the indeterminate result 0
. To find the correct value of x

when x = 0, you need to take the limit.

Limit@expr,x->x0 D find the limit of expr when x approaches x0

Finding limits.
Mathematics and Algorithms 209

sin HxL
This gives the correct value for the limit of as x 0.
x

In[2]:= Limit@Sin@xD x, x -> 0D

Out[2]= 1

No finite limit exists in this case.

In[3]:= Limit@Sin@xD x ^ 2, x -> 0D
Out[3]=

Limit can find this limit, even though you cannot get an ordinary power series for x log HxL at
x = 0.
In[4]:= Limit@x Log@xD, x -> 0D
Out[4]= 0

The same is true here.

In[5]:= Limit@H1 + 2 xL ^ H1 xL, x -> 0D

Out[5]= 2

The value of Sign@xD at x = 0 is 0.

In[6]:= Sign@0D
Out[6]= 0

Its limit, however, is 1. The limit is by default taken from above.

In[7]:= Limit@Sign@xD, x -> 0D
Out[7]= 1

Not all functions have definite limits at particular points. For example, the function sinH1 xL
oscillates infinitely often near x = 0, so it has no definite limit there. Nevertheless, at least so
long as x remains real, the values of the function near x = 0 always lie between -1 and 1. Limit
represents values with bounded variation using Interval objects. In general, Interval@8xmin ,
xmax <D represents an uncertain value which lies somewhere in the interval xmin to xmax .

Limit returns an Interval object, representing the range of possible values of sin H1 xL near
its essential singularity at x = 0.
In[8]:= Limit@Sin@1 xD, x -> 0D
Out[8]= Interval@8-1, 1<D

Mathematica can do arithmetic with Interval objects.

210 Mathematics and Algorithms

Mathematica can do arithmetic with Interval objects.

In[9]:= H1 + %L ^ 3
Out[9]= Interval@80, 8<D

Mathematica represents this limit symbolically in terms of an Interval object.

In[10]:= Limit@Exp@Sin@xDD, x -> InfinityD
1
Out[10]= IntervalB: , >F

Some functions may have different limits at particular points, depending on the direction from
which you approach those points. You can use the Direction option for Limit to specify the
direction you want.

LimitAexpr,x->x0 ,Direction->1E find the limit as x approaches x0 from below

LimitAexpr,x->x0 ,Direction->-1E find the limit as x approaches x0 from above

Directional limits.

The function 1 x has a different limiting value at x = 0, depending on whether you approach
from above or below.
In[11]:= Plot@1 x, 8x, - 1, 1<D
10

Out[11]= -1.0 -0.5 0.5 1.0

-5

-10

Approaching from below gives a limiting value of -.

In[12]:= Limit@1 x, x -> 0, Direction -> 1D
Out[12]= -

Approaching from above gives a limiting value of .

In[13]:= Limit@1 x, x -> 0, Direction -> - 1D
Out[13]=

Limit makes no assumptions about functions like f@xD about which it does not have definite
knowledge. As a result, Limit remains unevaluated in most cases involving symbolic functions.

Limit has no definite knowledge about f, so it leaves this limit unevaluated.

Mathematics and Algorithms 211

Limit has no definite knowledge about f, so it leaves this limit unevaluated.

In[14]:= Limit@x f@xD, x -> 0D
Out[14]= Limit@x f@xD, x 0D

Residues
Limit@expr, x -> x0 D tells you what the value of expr is when x tends to x0 . When this value is
infinite, it is often useful instead to know the residue of expr when x equals x0 . The residue is
given by the coefficient of Hx - x0 L-1 in the power series expansion of expr about the point x0 .

Residue@expr,8x,x0 <D the residue of expr when x equals x0

Computing residues.

The residue here is equal to 1.

In[1]:= Residue@1 x, 8x, 0<D
Out[1]= 1

The residue here is zero.

In[2]:= Residue@1 x ^ 2, 8x, 0<D
Out[2]= 0

Residues can be computed at the point at infinity.

In[3]:= Residue@1 x, 8x, ComplexInfinity<D
Out[3]= -1

Pad Approximation
The Pad approximation is a rational function that can be thought of as a generalization of a
Taylor polynomial. A rational function is the ratio of polynomials. Because these functions only
use the elementary arithmetic operations, they are very easy to evaluate numerically. The
polynomial in the denominator allows you to approximate functions that have rational
singularities.
212 Mathematics and Algorithms

PadeApproximant@ f ,8x,x0 ,8n,m<<D give the Pad approximation to f centered at x0 of order

Hn, mL
PadeApproximant@ f ,8x,x0 ,n<D give the diagonal Pad approximation to f centered at x0
of order n

Pad approximations.

More precisely, a Pad approximation of order Hn, mL to an analytic function f HxL at a regular
p HxL
point or pole x0 is the rational function q HxL
where pHxL is a polynomial of degree n, qHxL is a polyno -

mial of degree m, and the formal power series of f HxL qHxL - pHxL about the point x0 begins with the
term Hx - x0 Ln+m+1 . If m is equal to n, the approximation is called a diagonal Pad approximation of
order n.

Here is the Pad approximation of order H2, 4L to cosHxL at x = 0.

In[1]:= PadeApproximant@Cos@xD, 8x, 0, 82, 4<<D
61 x2
1-
150
Out[1]=
7 x2 x4
1+ +
75 200

This gives another Pad approximation of the same order.

In[2]:= pd = PadeApproximant@x , 8x, 1, 82, 4<<D

1 1
+ H-1 + xL + H-1 + xL2
3 30
Out[2]=
2 1 1 1
1- H-1 + xL + H-1 + xL2 - H-1 + xL3 + H-1 + xL4
3 5 30 360

The initial terms of this series vanish. This is the property that characterizes the Pad approxima-
tion.
In[3]:= Series@x Denominator@pdD - Numerator@pdD, 8x, 1, 8<D
Hx - 1L7 Hx - 1L8
Out[3]= + + O@x - 1D9
75 600 120 960
Mathematics and Algorithms 213

This plots the difference between the approximation and the true function. Notice that the
approximation is very good near the center of expansion, but the error increases rapidly as you
move away.
In[4]:= Plot@pd - x , 8x, 0, 2<D
0.00001

5. 10-6

Out[4]= 0.5 1.0 1.5 2.0

-5. 10-6

-0.00001

In Mathematica PadeApproximant is generalized to allow expansion about branch points.

This gives the diagonal Pad approximation of order 1 to a generalized rational function at x = 0.
Sqrt@xD
In[5]:= PadeApproximantB , 8x, 0, 1<F
H1 + Sqrt@xDL ^ 3
x
x -
3
Out[5]=
8 x
1+ +2x
3

This gives the diagonal Pad approximation of order 5 to the logarithm of a rational function at
the branch point x = 0.
x
In[6]:= PadeApproximantBLogB F, 8x, 0, 5<F
1+x
47 x3 11 x4 137 x5
-x - 2 x2 - - -
36 36 7560
Out[6]= + Log@xD
5x 20 x2 5 x3 5 x4 x5
1+ + + + +
2 9 6 42 252

The series expansion of the function agrees with the diagonal Pad approximation up to order
10.
x
In[7]:= SeriesB% - LogB F, 8x, 0, 11<F
1+x
x11
Out[7]= + O@xD12
698 544
214 Mathematics and Algorithms

Calculus

Differentiation

D@ f ,xD partial derivative f
x

D@ f ,x,y,D multiple derivative f
x y

n
D@ f 8x,n<D nth derivative xn
f

DA f ,x,NonConstants->8v1 ,v2 ,<E

x
f with the vi taken to depend on x

Partial differentiation operations.

This gives xn .
x

In[1]:= D@x ^ n, xD

Out[1]= n x-1+n

This gives the third derivative.

In[2]:= D@x ^ n, 8x, 3<D

Out[2]= H-2 + nL H-1 + nL n x-3+n

You can differentiate with respect to any expression that does not involve explicit mathematical
operations.
In[3]:= D@x@1D ^ 2 + x@2D ^ 2, x@1DD
Out[3]= 2 x@1D

D does partial differentiation. It assumes here that y is independent of x.

In[4]:= D@x ^ 2 + y ^ 2, xD
Out[4]= 2x
Mathematics and Algorithms 215

If y does in fact depend on x, you can use the explicit functional form y@xD. "The Representa-
tion of Derivatives" describes how objects like y '@xD work.
In[5]:= D@x ^ 2 + y@xD ^ 2, xD
Out[5]= 2 x + 2 y@xD y @xD

Instead of giving an explicit function y@xD, you can tell D that y implicitly depends on x.
y
D@y, x, NonConstants -> 8y<D then represents x
, with y implicitly depending on x.

In[6]:= D@x ^ 2 + y ^ 2, x, NonConstants -> 8y<D

Out[6]= 2 x + 2 y D@y, x, NonConstants 8y<D

D@ f ,88x1 ,x2 ,<<D the gradient of a scalar function f H f x1 , f x2 , L

D@ f ,88x1 ,x2 ,<,2<D the Hessian matrix for f

D@ f ,88x1 ,x2 ,<,n<D the nth -order Taylor series coefficient

D@8 f1 , f2 ,<,88x1 ,x2 ,<<D the Jacobian for a vector function f

Vector derivatives.

This gives the gradient of the function x2 + y2 .

In[7]:= D@x ^ 2 + y ^ 2, 88x, y<<D
Out[7]= 82 x, 2 y<

This gives the Hessian.

In[8]:= D@x ^ 2 + y ^ 2, 88x, y<, 2<D
Out[8]= 882, 0<, 80, 2<<

This gives the Jacobian for a vector function.

In[9]:= D@8x ^ 2 + y ^ 2, x y<, 88x, y<<D
Out[9]= 882 x, 2 y<, 8y, x<<
216 Mathematics and Algorithms

Total Derivatives

Dt@ f D total differential d f

df
Dt@ f ,xD total derivative
dx

d d
Dt@ f ,x,y,D multiple total derivative f
dx dy

DtA f ,x,Constants->8c1 ,c2 ,<E

total derivative with ci constant (i.e., d ci = 0)
dy
y:Dt@y,xD=0 set =0
dx

SetAttributesAc,ConstantE define c to be a constant in all cases

Total differentiation operations.

When you find the derivative of some expression f with respect to x, you are effectively finding
out how fast f changes as you vary x. Often f will depend not only on x, but also on other
variables, say y and z. The results that you get then depend on how you assume that y and z
vary as you change x.

There are two common cases. Either y and z are assumed to stay fixed when x changes, or they
f
are allowed to vary with x. In a standard partial derivative x
, all variables other than x are
df
assumed fixed. On the other hand, in the total derivative dx
, all variables are allowed to change

with x.

In Mathematica, D@ f , xD gives a partial derivative, with all other variables assumed indepen-
dent of x. Dt@ f , xD gives a total derivative, in which all variables are assumed to depend on x.
In both cases, you can add an argument to give more information on dependencies.

This gives the partial derivative Ix2 + y2 M. y is assumed to be independent of x.
x

In[1]:= D@x ^ 2 + y ^ 2, xD
Out[1]= 2x

d
This gives the total derivative Ix2 + y2 M. Now y is assumed to depend on x.
dx

In[2]:= Dt@x ^ 2 + y ^ 2, xD
Out[2]= 2 x + 2 y Dt@y, xD

dy
You can make a replacement for .
dx
Mathematics and Algorithms 217

dy
You can make a replacement for .
dx

In[3]:= % . Dt@y, xD -> yp

Out[3]= 2 x + 2 y yp

dy
You can also make an explicit definition for . You need to use y : to make sure that the
dx
definition is associated with y.
In[4]:= y : Dt@y, xD = 0
Out[4]= 0

With this definition made, Dt treats y as independent of x.

In[5]:= Dt@x ^ 2 + y ^ 2 + z ^ 2, xD
Out[5]= 2 x + 2 z Dt@z, xD

This removes your definition for the derivative of y.

In[6]:= Clear@yD

This takes the total derivative, with z held fixed.

In[7]:= Dt@x ^ 2 + y ^ 2 + z ^ 2, x, Constants -> 8z<D
Out[7]= 2 x + 2 y Dt@y, x, Constants 8z<D

This specifies that c is a constant under differentiation.

In[8]:= SetAttributes@c, ConstantD

The variable c is taken as a constant.

In[9]:= Dt@a ^ 2 + c x ^ 2, xD
Out[9]= 2 c x + 2 a Dt@a, xD

The function c is also assumed to be a constant.

In[10]:= Dt@a ^ 2 + c@xD x ^ 2, xD
Out[10]= 2 x c@xD + 2 a Dt@a, xD

This gives the total differential d Ix2 + c y2 M.

In[11]:= Dt@x ^ 2 + c y ^ 2D
Out[11]= 2 x Dt@xD + 2 c y Dt@yD

You can make replacements and assignments for total differentials.

218 Mathematics and Algorithms

You can make replacements and assignments for total differentials.

In[12]:= % . Dt@yD -> dy
Out[12]= 2 c dy y + 2 x Dt@xD

Derivatives of Unknown Functions

Differentiating a known function gives an explicit result.

In[1]:= D@Log@xD ^ 2, xD
2 Log@xD
Out[1]=
x

Differentiating an unknown function f gives a result in terms of f '.

In[2]:= D@f@xD ^ 2, xD
Out[2]= 2 f@xD f @xD

Mathematica applies the chain rule for differentiation, and leaves the result in terms of f '.
In[3]:= D@x f@x ^ 2D, xD

Out[3]= fAx2 E + 2 x2 f Ax2 E

Differentiating again gives a result in terms of f, f ' and f ''.

In[4]:= D@%, xD

Out[4]= 6 x f Ax2 E + 4 x3 f Ax2 E

When a function has more than one argument, superscripts are used to indicate how many
times each argument is being differentiated.
In[5]:= D@g@x ^ 2, y ^ 2D, xD

Out[5]= 2 x gH1,0L Ax2 , y2 E

This represents g Hx, yL. Mathematica assumes that the order in which derivatives are
x x y
taken with respect to different variables is irrelevant.
In[6]:= D@g@x, yD, x, x, yD

Out[6]= gH2,1L @x, yD

Mathematics and Algorithms 219

You can find the value of the derivative when x = 0 by replacing x with 0.
In[7]:= % . x -> 0

Out[7]= gH2,1L @0, yD

f ' [x] first derivative of a function of one variable

f HnL [x] nth derivative of a function of one variable

f Hn1 ,n2 , L [x] derivative of a function of several variables, ni times with

respect to variable i

Output forms for derivatives of unknown functions.

The Representation of Derivatives

Derivatives in Mathematica work essentially the same as in standard mathematics. The usual
mathematical notation, however, often hides many details. To understand how derivatives are
represented in Mathematica, we must look at these details.

d
The standard mathematical notation f H0L is really a shorthand for dt
f HtL t=0 , where t is a
d
"dummy variable". Similarly, f Ix2 M is a shorthand for dt
f HtL t=x2 . As suggested by the notation f ,
d
the object dt
f HtL can in fact be viewed as a "pure function", to be evaluated with a particular

choice of its parameter t. You can think of the operation of differentiation as acting on a func-
tion f , to give a new function, usually called f .

With functions of more than one argument, the simple notation based on primes breaks down.
d d
You cannot tell for example whether g H0, 1L stands for dt
gHt, 1L t=0 or dt
gH0, tL t=1 , and for almost

any g, these will have totally different values. Once again, however, t is just a dummy variable,
whose sole purpose is to show with respect to which "slot" g is to be differentiated.

In Mathematica, as in some branches of mathematics, it is convenient to think about a kind of

differentiation that acts on functions, rather than expressions. We need an operation that takes
the function f , and gives us the derivative function f . Operations such as this that act on
functions, rather than variables, are known in mathematics as operators.

The object f ' in Mathematica is the result of applying the differentiation operator to the func-
tion f. The full form of f ' is in fact Derivative@1D@fD. Derivative@1D is the Mathematica
differentiation operator.
220 Mathematics and Algorithms

The arguments in the operator Derivative@n1 , n2 , D specify how many times to differentiate
with respect to each "slot" of the function on which it acts. By using operators to represent
differentiation, Mathematica avoids any need to introduce explicit "dummy variables".

This is the full form of the derivative of the function f.

In[1]:= f ' FullForm
Out[1]//FullForm= Derivative@1D@fD

Here an argument x is supplied.

In[2]:= f '@xD FullForm
Out[2]//FullForm= Derivative@1D@fD@xD

This is the second derivative.

In[3]:= f ''@xD FullForm
Out[3]//FullForm= Derivative@2D@fD@xD

This gives a derivative of the function g with respect to its second "slot".
In[4]:= D@g@x, yD, yD

Out[4]= gH0,1L @x, yD

Here is the full form.

In[5]:= % FullForm
Out[5]//FullForm= Derivative@0, 1D@gD@x, yD

Here is the second derivative with respect to the variable y, which appears in the second slot of
g.
In[6]:= D@g@x, yD, 8y, 2<D FullForm
Out[6]//FullForm= Derivative@0, 2D@gD@x, yD

This is a mixed derivative.

In[7]:= D@g@x, yD, x, y, yD FullForm
Out[7]//FullForm= Derivative@1, 2D@gD@x, yD

Since Derivative only specifies how many times to differentiate with respect to each slot, the
order of the derivatives is irrelevant.
Mathematics and Algorithms 221

Since Derivative only specifies how many times to differentiate with respect to each slot, the
order of the derivatives is irrelevant.
In[8]:= D@g@x, yD, y, y, xD FullForm
Out[8]//FullForm= Derivative@1, 2D@gD@x, yD

Here is a more complicated case, in which both arguments of g depend on the differentiation
variable.
In[9]:= D@g@x, xD, xD

Out[9]= gH0,1L @x, xD + gH1,0L @x, xD

This is the full form of the result.

In[10]:= % FullForm
Out[10]//FullForm= Plus@Derivative@0, 1D@gD@x, xD, Derivative@1, 0D@gD@x, xDD

The object f ' behaves essentially like any other function in Mathematica. You can evaluate the
function with any argument, and you can use standard Mathematica . operations to change
the argument. (This would not be possible if explicit dummy variables had been introduced in
the course of the differentiation.)

This is the Mathematica representation of the derivative of a function f, evaluated at the origin.
In[11]:= f '@0D FullForm
Out[11]//FullForm= Derivative@1D@fD@0D

The result of this derivative involves f ' evaluated with the argument x ^ 2.
In[12]:= D@f@x ^ 2D, xD

Out[12]= 2 x f Ax2 E

You can evaluate the result at the point x = 2 by using the standard Mathematica replacement
operation.
In[13]:= % . x -> 2

Out[13]= 4 f @4D
222 Mathematics and Algorithms

There is some slight subtlety when you need to deduce the value of f ' based on definitions for
objects like f@x_D.

Here is a definition for a function h.

In[14]:= h@x_D := x ^ 4

When you take the derivative of h@xD, Mathematica first evaluates h@xD, then differentiates
the result.
In[15]:= D@h@xD, xD

Out[15]= 4 x3

You can get the same result by applying the function h ' to the argument x.
In[16]:= h '@xD

Out[16]= 4 x3

Here is the function h ' on its own.

In[17]:= h'

Out[17]= 4 13 &

The function f ' is completely determined by the form of the function f. Definitions for objects
like f@x_D do not immediately apply however to expressions like f '@xD. The problem is that
f '@xD has the full form Derivative@1D@fD@xD, which nowhere contains anything that explicitly
matches the pattern f@x_D. In addition, for many purposes it is convenient to have a representa-
tion of the function f ' itself, without necessarily applying it to any arguments.

What Mathematica does is to try and find the explicit form of a pure function which represents
the object f '. When Mathematica gets an expression like Derivative@1D@fD, it effectively
converts it to the explicit form D@f@D, D & and then tries to evaluate the derivative. In the
explicit form, Mathematica can immediately use values that have been defined for objects like
f@x_D. If Mathematica succeeds in doing the derivative, it returns the explicit pure-function
result. If it does not succeed, it leaves the derivative in the original f ' form.

This gives the derivative of Tan in pure-function form.

In[18]:= Tan '

Out[18]= Sec@1D2 &

Mathematics and Algorithms 223

Here is the result of applying the pure function to the specific argument y.
In[19]:= %@yD

Out[19]= Sec@yD2

Defining Derivatives
You can define the derivative in Mathematica of a function f of one argument simply by an
assignment like f '@x_D = fp@xD.

This defines the derivative of f HxL to be f pHxL. In this case, you could have used = instead of :=.
In[1]:= f '@x_D := fp@xD

The rule for f '@x_D is used to evaluate this derivative.

In[2]:= D@f@x ^ 2D, xD

Out[2]= 2 x fpAx2 E

Differentiating again gives derivatives of f p.

In[3]:= D@%, xD

Out[3]= 2 fpAx2 E + 4 x2 fp Ax2 E

This defines a value for the derivative of g at the origin.

In[4]:= g '@0D = g0
Out[4]= g0

The value for g '@0D is used.

In[5]:= D@g@xD ^ 2, xD . x -> 0
Out[5]= 2 g0 g@0D

This defines the second derivative of g, with any argument.

In[6]:= g ''@x_D = gpp@xD
Out[6]= gpp@xD
224 Mathematics and Algorithms

The value defined for the second derivative is used.

In[7]:= D@g@xD ^ 2, 8x, 2<D

Out[7]= 2 g@xD gpp@xD + 2 g @xD2

To define derivatives of functions with several arguments, you have to use the general represen-
tation of derivatives in Mathematica.

f '@x _D:=rhs define the first derivative of f

Derivative@nD@ f D@x _D:=rhs define the nth derivative of f

Derivative@m,n,D@gD@x _,_,D:=rhs
define derivatives of g with respect to various arguments

Defining derivatives.

This defines the second derivative of g with respect to its second argument.
In[8]:= Derivative@0, 2D@gD@x_, y_D := g2p@x, yD

This uses the definition just given.

In[9]:= D@g@a ^ 2, x ^ 2D, x, xD

Out[9]= 4 x2 g2pAa2 , x2 E + 2 gH0,1L Aa2 , x2 E

Integration

Here is the integral xn d x in Mathematica.

In[1]:= Integrate@x ^ n, xD
x1+n
Out[1]=
1+n

Here is a slightly more complicated example.

In[2]:= Integrate@1 Hx ^ 4 - a ^ 4L, xD
x
ArcTanB F Log@a - xD Log@a + xD
a
Out[2]= - + -
3 3
2a 4a 4 a3

Mathematica knows how to do almost any integral that can be done in terms of standard mathe-
matical functions. But you should realize that even though an integrand may contain only fairly
simple functions, its integral may involve much more complicated functions~or may not be
expressible at all in terms of standard mathematical functions.
Mathematics and Algorithms 225

Here is a fairly straightforward integral.

In[3]:= Integrate@Log@1 - x ^ 2D, xD

Out[3]= -2 x - Log@-1 + xD + Log@1 + xD + x LogA1 - x2 E

This integral can be done only in terms of a dilogarithm function.

In[4]:= Integrate@Log@1 - x ^ 2D x, xD
1
Out[4]= - PolyLogA2, x2 E
2

This integral involves Erf.

In[5]:= Integrate@Exp@1 - x ^ 2D, xD
1
Out[5]= p Erf@xD
2

And this one involves a Fresnel function.

In[6]:= Integrate@Sin@x ^ 2D, xD

p 2
Out[6]= FresnelSB xF
2 p

Even this integral requires a hypergeometric function.

In[7]:= Integrate@H1 - x ^ 2L ^ n, xD
1 3
Out[7]= x Hypergeometric2F1B , -n, , x2 F
2 2

This integral simply cannot be done in terms of standard mathematical functions. As a result,
Mathematica just leaves it undone.
In[8]:= Integrate@x ^ x, xD
x
Out[8]= x x
226 Mathematics and Algorithms

Integrate@ f ,xD the indefinite integral f d x

Integrate@ f ,x,yD the multiple integral d x d y f

xmax
Integrate@ f ,8x,xmin ,xmax <D the definite integral f dx
xmin

Integrate@ f ,8x,xmin ,xmax <,8y,ymin ,ymax <D

xmax y
the multiple integral d x y max d y f
x min min

Integration.

b
Here is the definite integral a sin2 HxL d x.

In[9]:= Integrate@Sin@xD ^ 2, 8x, a, b<D

1
Out[9]= H-a + b + Cos@aD Sin@aD - Cos@bD Sin@bDL
2

Here is another definite integral.

In[10]:= Integrate@Exp@- x ^ 2D, 8x, 0, Infinity<D

p
Out[10]=
2

Mathematica cannot give you a formula for this definite integral.

In[11]:= Integrate@x ^ x, 8x, 0, 1<D
1
x
Out[11]= x x
0

You can still get a numerical result, though.

In[12]:= N@%D
Out[12]= 0.783431

1 x
This evaluates the multiple integral 0 d x 0 d y Ix2 + y2 M. The range of the outermost integration
variable appears first.
In[13]:= Integrate@x ^ 2 + y ^ 2, 8x, 0, 1<, 8y, 0, x<D
1
Out[13]=
3
Mathematics and Algorithms 227

This integrates x10 over a circular region.

In[14]:= Integrate@x ^ 10 Boole@x ^ 2 + y ^ 2 <= 1D, 8x, - 1, 1<, 8y, - 1, 1<D
21 p
Out[14]=
512

Indefinite Integrals
The Mathematica function Integrate@ f , xD gives you the indefinite integral f d x. You can

think of the operation of indefinite integration as being an inverse of differentiation. If you take
the result from Integrate@ f , xD, and then differentiate it, you always get a result that is mathe-
matically equal to the original expression f .

In general, however, there is a whole family of results which have the property that their deriva-
tive is f . Integrate@ f , xD gives you an expression whose derivative is f . You can get other
expressions by adding an arbitrary constant of integration, or indeed by adding any function
that is constant except at discrete points.

If you fill in explicit limits for your integral, any such constants of integration must cancel out.
But even though the indefinite integral can have arbitrary constants added, it is still often very
convenient to manipulate it without filling in the limits.

Mathematica applies standard rules to find indefinite integrals.

In[1]:= Integrate@x ^ 2, xD
x3
Out[1]=
3

You can add an arbitrary constant to the indefinite integral, and still get the same derivative.
Integrate simply gives you an expression with the required derivative.
In[2]:= D@% + c, xD

Out[2]= x2

dx
This gives the indefinite integral .
x2 -1

In[3]:= Integrate@1 Hx ^ 2 - 1L, xD

1 1
Out[3]= Log@-1 + xD - Log@1 + xD
2 2
228 Mathematics and Algorithms

Differentiating should give the original function back again.

In[4]:= D@%, xD
1 1
Out[4]= -
2 H-1 + xL 2 H1 + xL

You need to manipulate it to get it back into the original form.

In[5]:= Simplify@%D
1
Out[5]=
-1 + x2

The Integrate function assumes that any object that does not explicitly contain the integration
variable is independent of it, and can be treated as a constant. As a result, Integrate is like an
inverse of the partial differentiation function D.

The variable a is assumed to be independent of x.

In[6]:= Integrate@a x ^ 2, xD
a x3
Out[6]=
3

The integration variable can be any expression that does not involve explicit mathematical
operations.
In[7]:= Integrate@x b@xD ^ 2, b@xDD
1
Out[7]= x b@xD3
3

Another assumption that Integrate implicitly makes is that all the symbolic quantities in your
xn+1
integrand have "generic" values. Thus, for example, Mathematica will tell you that xn x is n+1

even though this is not true in the special case n = -1.

Mathematica gives the standard result for this integral, implicitly assuming that n is not equal to
- 1.
In[8]:= Integrate@x ^ n, xD
x1+n
Out[8]=
1+n

If you specifically give an exponent of - 1, Mathematica produces a different result.

In[9]:= Integrate@x ^ - 1, xD
Out[9]= Log@xD

You should realize that the result for any particular integral can often be written in many differ-
ent forms. Mathematica tries to give you the most convenient form, following principles such as
Mathematics and Algorithms 229

This integral is given in terms of ArcTan.

In[10]:= Integrate@1 H1 + a x ^ 2L, xD

ArcTanB a xF
Out[10]=
a

This integral is given in terms of ArcTanh .

In[11]:= Integrate@1 H1 - b x ^ 2L, xD

ArcTanhB b xF
Out[11]=
b

This is mathematically equal to the first integral, but is given in a somewhat different form.
In[12]:= % . b -> - a

ArcTanhB -a xF
Out[12]=
-a

The derivative is still correct.

In[13]:= D@%, xD
1
Out[13]=
1 + a x2

Even though they look quite different, both ArcTan@xD and - ArcTan@1 xD are indefinite
integrals of 1 I1 + x2 M.
In[14]:= Simplify@D@8ArcTan@xD, - ArcTan@1 xD<, xDD
1 1
Out[14]= : , >
1 + x2 1 + x2

Integrate chooses to use the simpler of the two forms.

In[15]:= Integrate@1 H1 + x ^ 2L, xD
Out[15]= ArcTan@xD
230 Mathematics and Algorithms

Integrals That Can and Cannot Be Done

Evaluating integrals is much more difficult than evaluating derivatives. For derivatives, there is
a systematic procedure based on the chain rule that effectively allows any derivative to be
worked out. But for integrals, there is no such systematic procedure.

One of the main problems is that it is difficult to know what kinds of functions will be needed to
evaluate a particular integral. When you work out a derivative, you always end up with func-
tions that are of the same kind or simpler than the ones you started with. But when you work
out integrals, you often end up needing to use functions that are much more complicated than
the ones you started with.

This integral can be evaluated using the same kind of functions that appeared in the input.
In[1]:= Integrate@Log@xD ^ 2, xD

Out[1]= 2 x - 2 x Log@xD + x Log@xD2

But for this integral the special function LogIntegral is needed.

In[2]:= Integrate@Log@Log@xDD, xD
Out[2]= x Log@Log@xDD - LogIntegral@xD

It is not difficult to find integrals that require all sorts of functions.

In[3]:= Integrate@Sin@x ^ 2D, xD

p 2
Out[3]= FresnelSB xF
2 p

This integral involves an incomplete gamma function. Note that the power is carefully set up to
allow any complex value of x.
In[4]:= Integrate@Exp@- x ^ aD, xD
1
x Hxa L-1a GammaB , xa F
a
Out[4]= -
a

Mathematica includes a very wide range of mathematical functions, and by using these func-
tions a great many integrals can be done. But it is still possible to find even fairly simple-look-
ing integrals that just cannot be done in terms of any standard mathematical functions.

Here is a fairly simple-looking integral that cannot be done in terms of any standard mathemati-
cal functions.
Mathematics and Algorithms 231

Here is a fairly simple-looking integral that cannot be done in terms of any standard mathemati-
cal functions.
In[5]:= Integrate@Sin@xD Log@xD, xD
Sin@xD
Out[5]= x
Log@xD

The main point of being able to do an integral in terms of standard mathematical functions is
that it lets one use the known properties of these functions to evaluate or manipulate the result
one gets.

In the most convenient cases, integrals can be done purely in terms of elementary functions
such as exponentials, logarithms and trigonometric functions. In fact, if you give an integrand
that involves only such elementary functions, then one of the important capabilities of
Integrate is that if the corresponding integral can be expressed in terms of elementary func-
tions, then Integrate will essentially always succeed in finding it.

Integrals of rational functions are straightforward to evaluate, and always come out in terms of
rational functions, logarithms and inverse trigonometric functions.
In[6]:= Integrate@x HHx - 1L Hx + 2LL, xD
1 2
Out[6]= Log@-1 + xD + Log@2 + xD
3 3

The integral here is still of the same form, but now involves an implicit sum over the roots of a
polynomial.
In[7]:= Integrate@1 H1 + 2 x + x ^ 3L, xD
Log@x - 1D
Out[7]= RootSumB1 + 2 1 + 13 &, &F
2 + 3 12

This finds numerical approximations to all the root objects.

In[8]:= N@%D
Out[8]= H-0.19108 - 0.088541 L [email protected] - 1.46771 L + xD -
H0.19108 - 0.088541 L [email protected] + 1.46771 L + xD + 0.38216 [email protected] + xD

Integrals of trigonometric functions usually come out in terms of other trigonometric functions.
In[9]:= Integrate@Sin@xD ^ 3 Cos@xD ^ 2, xD
Cos@xD 1 1
Out[9]= - - Cos@3 xD + Cos@5 xD
8 48 80
232 Mathematics and Algorithms

This is a fairly simple integral involving algebraic functions.

In[10]:= Integrate@Sqrt@xD Sqrt@1 + xD, xD
1
Out[10]= x 1 + x H1 + 2 xL - ArcSinhB x F
4

Here is an integral involving nested square roots.

In[11]:= Integrate@Sqrt@x + Sqrt@xDD, xD
1 1
Out[11]= x + x J-3 + 2 x + 8 xN + LogB1 + 2 x +2 x +x F
12 8

By nesting elementary functions you sometimes get integrals that can be done in terms of
elementary functions.
In[12]:= Integrate@Cos@Log@xDD, xD
1 1
Out[12]= x Cos@Log@xDD + x Sin@Log@xDD
2 2

But more often other kinds of functions are needed.

In[13]:= Integrate@Log@Cos@xDD, xD
x2 1
Out[13]= - x LogA1 + 2 x E + x Log@Cos@xDD + PolyLogA2, -2 x E
2 2

Integrals like this typically come out in terms of elliptic functions.

In[14]:= Integrate@Sqrt@Cos@xDD, xD
x
Out[14]= 2 EllipticEB , 2F
2

But occasionally one can get results in terms of elementary functions alone.
In[15]:= Integrate@Sqrt@Tan@xDD, xD
1
Out[15]= -2 ArcTanB1 - 2 Tan@xD F + 2 ArcTanB1 + 2 Tan@xD F +
2 2
LogB-1 + 2 Tan@xD - Tan@xDF - LogB1 + 2 Tan@xD + Tan@xDF

Integrals like this can systematically be done using Piecewise.

In[16]:= Integrate@2 ^ Max@x, 1 - xD, xD
21-x 1
- x
Log@2D 2
Out[16]=
2 2 2x
- + True
Log@2D Log@2D

Beyond working with elementary functions, Integrate includes a large number of algorithms
for dealing with special functions. Sometimes it uses a direct generalization of the procedure for
elementary functions. But more often its strategy is first to try to write the integrand in a form
that can be integrated in terms of certain sophisticated special functions, and then having done
this to try to find reductions of these sophisticated functions to more familiar functions.

To integrate this Bessel function requires a generalized hypergeometric function.

In[17]:= Integrate@BesselJ@0, xD, xD
1 3 x2
Out[17]= x HypergeometricPFQB: >, :1, >, - F
2 2 4

Sometimes the integrals can be reduced to more familiar forms.

In[18]:= Integrate@x ^ 3 BesselJ@0, xD, xD

Out[18]= -x2 H-2 BesselJ@2, xD + x BesselJ@3, xDL

A large book of integral tables will list perhaps a few thousand indefinite integrals. Mathematica
can do essentially all of these integrals. And because it contains general algorithms rather than
just specific cases, Mathematica can actually do a vastly wider range of integrals.

You could expect to find this integral in any large book of integral tables.
In[19]:= Integrate@Log@1 - xD x, xD
Out[19]= -PolyLog@2, xD

To do this integral, however, requires a more general algorithm, rather than just a direct table
lookup.
In[20]:= Integrate@Log@1 + 3 x + x ^ 2D x, xD
1 2x 1 2x
Out[20]= Log@xD LogB 3- 5 + xF - LogB1 + F + Log@xD LogB 3+ 5 + xF - LogB1 + F +
2 3- 5 2 3+ 5
1 1
Log@xD -LogB 3 - 5 + xF - LogB 3+ 5 + xF + LogA1 + 3 x + x2 E -
2 2
2x 2x
PolyLogB2, - F - PolyLogB2, - F
3- 5 3+ 5

Particularly if you introduce new mathematical functions of your own, you may want to teach
Mathematica new kinds of integrals. You can do this by making appropriate definitions for
Integrate.

In the case of differentiation, the chain rule allows one to reduce all derivatives to a standard
form, represented in Mathematica using Derivative. But for integration, no such similar stan-
dard form exists, and as a result you often have to make definitions for several different ver-
sions of the same integral. Changes of variables and other transformations can rarely be done
234 Mathematics and Algorithms

This integral cannot be done in terms of any of the standard mathematical functions built into
Mathematica.
In[21]:= Integrate@Sin@Sin@xDD, xD

Out[21]= Sin@Sin@xDD x

Before you add your own rules for integration, you have to remove write protection.
In[22]:= Unprotect@IntegrateD
Out[22]= 8Integrate<

You can set up your own rule to define the integral to be, say, a "Jones" function.
In[23]:= Integrate@Sin@Sin@a_. + b_. x_DD, x_D := Jones@a, xD b

Now Mathematica can do integrals that give Jones functions.

In[24]:= Integrate@Sin@Sin@3 xDD, xD
1
Out[24]= Jones@0, xD
3

As it turns out, the integral sinHsinHxLL x can in principle be represented as an infinite sum of 2 F1

hypergeometric functions, or as a suitably generalized Kamp de Friet hypergeometric function

of two variables.

Definite Integrals

Integrate@ f ,xD the indefinite integral f dx

xmax
Integrate@ f ,8x,xmin ,xmax <D the definite integral x f dx
min

Integrate@ f ,8x,xmin ,xmax <,8y,ymin ,ymax <D

xmax ymax
the multiple integral x d x y d y f
min min

Integration functions.

b
Here is the integral a x2 d x.
Mathematics and Algorithms 235

b
Here is the integral a x2 d x.

In[1]:= Integrate@x ^ 2, 8x, a, b<D

a3 b3
Out[1]= - +
3 3

a b
This gives the multiple integral 0 d x 0 dyIx2 + y2 M.

In[2]:= Integrate@x ^ 2 + y ^ 2, 8x, 0, a<, 8y, 0, b<D

1
Out[2]= a b Ia2 + b2 M
3

The y integral is done first. Its limits can depend on the value of x. This ordering is the same as
is used in functions like Sum and Table .
In[3]:= Integrate@x ^ 2 + y ^ 2, 8x, 0, a<, 8y, 0, x<D
a4
Out[3]=
3

In simple cases, definite integrals can be done by finding indefinite forms and then computing
appropriate limits. But there is a vast range of integrals for which the indefinite form cannot be
expressed in terms of standard mathematical functions, but the definite form still can be.

This indefinite integral cannot be done in terms of standard mathematical functions.

In[4]:= Integrate@Cos@Sin@xDD, xD

Out[4]= Cos@Sin@xDD x

This definite integral, however, can be done in terms of a Bessel function.

In[5]:= Integrate@Cos@Sin@xDD, 8x, 0, 2 Pi<D
Out[5]= 2 p BesselJ@0, 1D

Here is an integral where the indefinite form can be found, but it is much more efficient to work
out the definite form directly.
In[6]:= Integrate@Log@xD Exp@- x ^ 2D, 8x, 0, Infinity<D
1
Out[6]= - p HEulerGamma + Log@4DL
4
236 Mathematics and Algorithms

Just because an integrand may contain special functions, it does not mean that the definite
integral will necessarily be complicated.
In[7]:= Integrate@BesselK@0, xD ^ 2, 8x, 0, Infinity<D
p2
Out[7]=
4

Special functions nevertheless occur in this result.

In[8]:= Integrate@BesselK@0, xD BesselJ@0, xD, 8x, 0, Infinity<D
1 2
GammaB F
4
Out[8]=
4 2p

The integrand here is simple, but the definite integral is not.

In[9]:= Integrate@Sin@x ^ 2D Exp@- xD, 8x, 0, Infinity<D
1 3 5 1 1 1
Out[9]= -2 HypergeometricPFQB81<, : , >, - F+ 2p CosB F + SinB F
4 4 4 64 4 4

Even when you can find the indefinite form of an integral, you will often not get the correct
answer for the definite integral if you just subtract the values of the limits at each end point.
The problem is that within the domain of integration there may be singularities whose effects
are ignored if you follow this procedure.

Here is the indefinite integral of 1 x2 .

In[10]:= Integrate@1 x ^ 2, xD
1
Out[10]= -
x

This subtracts the limits at each end point.

In[11]:= Limit@%, x -> 2D - Limit@%, x -> - 2D
Out[11]= -1

The true definite integral is divergent because of the double pole at x = 0.

In[12]:= Integrate@1 x ^ 2, 8x, - 2, 2<D

1
Integrate::idiv : Integral of does not converge on 8-2, 2<.
x2
2 1
Out[12]= x
-2 x2

Here is a more subtle example, involving branch cuts rather than poles.
Mathematics and Algorithms 237

Here is a more subtle example, involving branch cuts rather than poles.
In[13]:= Integrate@1 H1 + a Sin@xDL, xD
x
a+TanB F
2
2 ArcTanB F
1-a2
Out[13]=
1 - a2

Taking limits in the indefinite integral gives 0.

In[14]:= Limit@%, x -> 2 PiD - Limit@%, x -> 0D
Out[14]= 0

The definite integral, however, gives the correct result which depends on a. The assumption
assures convergence.
In[15]:= Integrate@1 H1 + a Sin@xDL, 8x, 0, 2 Pi<, Assumptions -> - 1 < a < 1D
2p
Out[15]=
1 - a2

IntegrateA f ,9x,xmin ,xmax =,PrincipalValue->TrueE

the Cauchy principal value of a definite integral

Principal value integrals.

Here is the indefinite integral of 1 x.

In[16]:= Integrate@1 x, xD
Out[16]= Log@xD

Substituting in the limits -1 and +2 yields a strange result involving i p.

In[17]:= Limit@%, x -> 2D - Limit@%, x -> - 1D
Out[17]= - p + Log@2D

The ordinary Riemann definite integral is divergent.

In[18]:= Integrate@1 x, 8x, - 1, 2<D

1
Integrate::idiv : Integral of does not converge on 8-1, 2<.
x
2 1
Out[18]= x
-1 x

The Cauchy principal value, however, is finite.

238 Mathematics and Algorithms

The Cauchy principal value, however, is finite.

In[19]:= Integrate@1 x, 8x, - 1, 2<, PrincipalValue -> TrueD
Out[19]= Log@2D

When parameters appear in an indefinite integral, it is essentially always possible to get results
that are correct for almost all values of these parameters. But for definite integrals this is no
longer the case. The most common problem is that a definite integral may converge only when
the parameters that appear in it satisfy certain specific conditions.

This indefinite integral is correct for all n -1.

In[20]:= Integrate@x ^ n, xD
x1+n
Out[20]=
1+n

For the definite integral, however, n must satisfy a condition in order for the integral to be
convergent.
In[21]:= Integrate@x ^ n, 8x, 0, 1<D
1
Out[21]= IfBRe@nD > -1, , IntegrateAxn , 8x, 0, 1<, Assumptions Re@nD -1EF
1+n

If n is replaced by 2, the condition is satisfied.

In[22]:= % . n -> 2
1
Out[22]=
3

option name default value

GenerateConditions Automatic whether to generate explicit conditions
Assumptions $Assumptions what relations about parameters to assume

Options for Integrate.

With the assumption n > 2, the result is always 1 H1 + nL.

In[23]:= Integrate@x ^ n, 8x, 0, 1<, Assumptions -> Hn > 2LD
1
Out[23]=
1+n

Even when a definite integral is convergent, the presence of singularities on the integration
path can lead to discontinuous changes when the parameters vary. Sometimes a single formula
containing functions like Sign can be used to summarize the result. In other cases, however,
an explicit If is more convenient.
Mathematics and Algorithms 239

The If here gives the condition for the integral to be convergent.

In[24]:= Integrate@Sin@a xD x, 8x, 0, Infinity<D
1 Sin@a xD
Out[24]= IfBa Reals, p Sign@aD, IntegrateB , 8x, 0, <, Assumptions a Re@aDFF
2 x

Here is the result assuming that a is real.

In[25]:= Integrate@Sin@a xD x, 8x, 0, Infinity<, Assumptions -> Im@aD == 0D
1
Out[25]= p Sign@aD
2

The result is discontinuous as a function of a. The discontinuity can be traced to the essential
singularity of sinHxL at x = .
In[26]:= Plot@%, 8a, - 5, 5<D
1.5

1.0

0.5

Out[26]=
-4 -2 2 4

-0.5

-1.0

-1.5

There is no convenient way to represent this answer in terms of Sign , so Mathematica gener-
ates an explicit If .
In[27]:= Integrate@Sin@xD BesselJ@0, a xD x, 8x, 0, Infinity<, Assumptions -> Im@aD == 0D
1
a ArcSinB F p
a
Out[27]= IfBa < -1 a > 1, , F
Abs@aD 2

Here is a plot of the resulting function of a.

In[28]:= Plot@Evaluate@%D, 8a, - 5, 5<D
1.5

1.0
Out[28]=

0.5

-4 -2 2 4

Integrals over
Integrals over Regions
Regions
240 Mathematics and Algorithms

Integrals over Regions

This does an integral over the interior of the unit circle.

In[1]:= Integrate@If@x ^ 2 + y ^ 2 < 1, 1, 0D, 8x, - 1, 1<, 8y, - 1, 1<D
Out[1]= p

Here is an equivalent form.

In[2]:= Integrate@Boole@x ^ 2 + y ^ 2 < 1D, 8x, - 1, 1<, 8y, - 1, 1<D
Out[2]= p

Even though an integral may be straightforward over a simple rectangular region, it can be
significantly more complicated even over a circular region.

This gives a Bessel function.

In[3]:= Integrate@Exp@xD Boole@x ^ 2 + y ^ 2 < 1D, 8x, - 1, 1<, 8y, - 1, 1<D
Out[3]= 2 p BesselI@1, 1D

Integrate@ f Boole@condD,8x,xmin ,xmax <,8y,ymin ,ymax <D

integrate f over the region where cond is True

Integrals over regions.

Particularly if there are parameters inside the conditions that define regions, the results for
integrals over regions may break into several cases.

This gives a piecewise function of a.

In[4]:= Integrate@Boole@a x < yD, 8x, 0, 1<, 8y, 0, 1<D
1 a0
2-a
0<a1
Out[4]= 2
1
True
2a

With two parameters even this breaks into quite a few cases.
In[5]:= Integrate@Boole@a x < bD, 8x, 0, 1<D
1 Ha > 0 && a - b 0L Ha 0 && b > 0L
a-b
a 0 && a - b < 0 && b 0
Out[5]= a
b
a > 0 && b > 0 && a - b > 0
a

This involves intersecting a circle with a square.

Mathematics and Algorithms 241

This involves intersecting a circle with a square.

In[6]:= Integrate@Boole@x ^ 2 + y ^ 2 < aD, 8x, 0, 1<, 8y, 0, 1<D
1 a2
ap
0<a1
Out[6]= 4
1
J2 -1 + a + a ArcCotB -1 + a F - a ArcTanB -1 + a FN 1 < a < 2
2

The region can have an infinite number of components.

In[7]:= Integrate@Boole@Sin@xD > 1 2D Exp@- xD, 8x, 0, Infinity<D

7 p6
Out[7]=
1 + 2 p3 + 4 p3

Manipulating Integrals in Symbolic Form

When Mathematica cannot give you an explicit result for an integral, it leaves the integral in a
symbolic form. It is often useful to manipulate this symbolic form.

Mathematica cannot give an explicit result for this integral, so it leaves the integral in symbolic
form.
In[1]:= Integrate@x ^ 2 f@xD, xD
2
Out[1]= x f@xD x

Differentiating the symbolic form gives the integrand back again.

In[2]:= D@%, xD

Out[2]= x2 f@xD

Here is a definite integral which cannot be done explicitly.

In[3]:= Integrate@f@xD, 8x, a@xD, b@xD<D
b@xD
Out[3]= f@xD x
a@xD

This gives the derivative of the definite integral.

In[4]:= D@%, xD
Out[4]= -f@a@xDD a @xD + f@b@xDD b @xD
242 Mathematics and Algorithms

Here is a definite integral with end points that do not explicitly depend on x.
In[5]:= defint = Integrate@f@xD, 8x, a, b<D
b
Out[5]= f@xD x
a

The partial derivative of this with respect to u is zero.

In[6]:= D@defint, uD
Out[6]= 0

There is a non-trivial total derivative, however.

In[7]:= Dt@defint, uD
Out[7]= -Dt@a, uD f@aD + Dt@b, uD f@bD

Differential Equations
You can use the Mathematica function DSolve to find symbolic solutions to ordinary and partial
differential equations.

Solving a differential equation consists essentially in finding the form of an unknown function.
In Mathematica, unknown functions are represented by expressions like y@xD. The derivatives
of such functions are represented by y '@xD, y ''@xD and so on.

The Mathematica function DSolve returns as its result a list of rules for functions. There is a
question of how these functions are represented. If you ask DSolve to solve for y@xD, then
DSolve will indeed return a rule for y@xD. In some cases, this rule may be all you need. But this
rule, on its own, does not give values for y '@xD or even y@0D. In many cases, therefore, it is
better to ask DSolve to solve not for y@xD, but instead for y itself. In this case, what DSolve
will return is a rule which gives y as a pure function, in the sense discussed in "Pure Functions".

If you ask DSolve to solve for y@xD, it will give a rule specifically for y@xD.
In[1]:= DSolve@y '@xD + y@xD == 1, y@xD, xD
Out[1]= 99y@xD 1 + -x C@1D==

The rule applies only to y@xD itself, and not, for example, to objects like y@0D or y '@xD.
In[2]:= y@xD + 2 y '@xD + y@0D . %
Out[2]= 91 + -x C@1D + y@0D + 2 y @xD=

If you ask DSolve to solve for y, it gives a rule for the object y on its own as a pure function.
Mathematics and Algorithms 243

If you ask DSolve to solve for y, it gives a rule for the object y on its own as a pure function.
In[3]:= DSolve@y '@xD + y@xD == 1, y, xD

Out[3]= 99y FunctionA8x<, 1 + -x C@1DE==

Now the rule applies to all occurrences of y.

In[4]:= y@xD + 2 y '@xD + y@0D . %
Out[4]= 92 + C@1D - -x C@1D=

Substituting the solution into the original equation yields True .

In[5]:= y '@xD + y@xD == 1 . %%
Out[5]= 8True<

DSolve@eqn,y@xD,xD solve a differential equation for y@xD

DSolve@eqn,y,xD solve a differential equation for the function y

Getting solutions to differential equations in different forms.

In standard mathematical notation, one typically represents solutions to differential equations

by explicitly introducing "dummy variables" to represent the arguments of the functions that
appear. If all you need is a symbolic form for the solution, then introducing such dummy vari-
ables may be convenient. However, if you actually intend to use the solution in a variety of
other computations, then you will usually find it better to get the solution in pure-function form,
without dummy variables. Notice that this form, while easy to represent in Mathematica, has no
direct analog in standard mathematical notation.

DSolve@8eqn1 ,eqn2 ,<,8y1 ,y2 ,<,xD

solve a list of differential equations

Solving simultaneous differential equations.

This solves two simultaneous differential equations.

In[6]:= DSolve@8y@xD == - z '@xD, z@xD == - y '@xD<, 8y, z<, xD
1 1
Out[6]= ::z FunctionB8x<, -x I1 + 2 x M C@1D - -x I-1 + 2 x M C@2DF,
2 2
1 1
y FunctionB8x<, - -x I-1 + 2 x M C@1D + -x I1 + 2 x M C@2DF>>
2 2
244 Mathematics and Algorithms

Mathematica returns two distinct solutions for y in this case.

In[7]:= DSolve@y@xD y '@xD == 1, y, xD

Out[7]= ::y FunctionB8x<, - 2 x + C@1D F>, :y FunctionB8x<, 2 x + C@1D F>>

You can add constraints and boundary conditions for differential equations by explicitly giving
additional equations such as y@0D == 0.

This asks for a solution which satisfies the condition y@0D == 1.

In[8]:= DSolve@8y '@xD == a y@xD, y@0D == 1<, y@xD, xD
Out[8]= 99y@xD a x ==

If you ask Mathematica to solve a set of differential equations and you do not give any con-
straints or boundary conditions, then Mathematica will try to find a general solution to your
equations. This general solution will involve various undetermined constants. One new constant
is introduced for each order of derivative in each equation you give.

The default is that these constants are named C@nD, where the index n starts at 1 for each
invocation of DSolve. You can override this choice, by explicitly giving a setting for the option
GeneratedParameters. Any function you give is applied to each successive index value n to get
the constants to use for each invocation of DSolve.

The general solution to this fourth-order equation involves four undetermined constants.
In[9]:= DSolve@y ''''@xD == y@xD, y@xD, xD
Out[9]= 99y@xD x C@1D + -x C@3D + C@2D Cos@xD + C@4D Sin@xD==

Each independent initial or boundary condition you give reduces the number of undetermined
constants by one.
In[10]:= DSolve@8y ''''@xD == y@xD, y@0D == y '@0D == 0<, y@xD, xD

Out[10]= 99y@xD -x IC@3D + 2 x C@3D - 2 x C@4D - 2 x C@3D Cos@xD + x C@4D Cos@xD + x C@4D Sin@xDM==

You should realize that finding exact formulas for the solutions to differential equations is a
difficult matter. In fact, there are only fairly few kinds of equations for which such formulas can
be found, at least in terms of standard mathematical functions.

The most widely investigated differential equations are linear ones, in which the functions you
are solving for, as well as their derivatives, appear only linearly.
Mathematics and Algorithms 245

This is a homogeneous first-order linear differential equation, and its solution is quite simple.
In[11]:= DSolve@y '@xD - x y@xD == 0, y@xD, xD
x2

Out[11]= ::y@xD 2 C@1D>>

Making the equation inhomogeneous leads to a significantly more complicated solution.

In[12]:= DSolve@y '@xD - x y@xD == 1, y@xD, xD

x2 x2 p x
Out[12]= ::y@xD 2 C@1D + 2 ErfB F>>
2 2

If you have only a single linear differential equation, and it involves only a first derivative of the
function you are solving for, then it turns out that the solution can always be found just by
doing integrals.

But as soon as you have more than one differential equation, or more than a first-order deriva-
tive, this is no longer true. However, some simple second-order linear differential equations can
nevertheless be solved using various special functions from "Special Functions". Indeed, histori-
cally many of these special functions were first introduced specifically in order to represent the
solutions to such equations.

This is Airys equation, which is solved in terms of Airy functions.

In[13]:= DSolve@y ''@xD - x y@xD == 0, y@xD, xD
Out[13]= 88y@xD AiryAi@xD C@1D + AiryBi@xD C@2D<<

This equation comes out in terms of Bessel functions.

In[14]:= DSolve@y ''@xD - Exp@xD y@xD == 0, y@xD, xD

Out[14]= ::y@xD BesselIB0, 2 x F C@1D + 2 BesselKB0, 2 x F C@2D>>

This requires Mathieu functions.

In[15]:= DSolve@y ''@xD + Cos@xD y@xD == 0, y, xD
x x
Out[15]= ::y FunctionB8x<, C@1D MathieuCB0, -2, F + C@2D MathieuSB0, -2, FF>>
2 2
246 Mathematics and Algorithms

And this Legendre functions.

In[16]:= DSolve@y ''@xD - Cot@xD ^ 2 y@xD == 0, y@xD, xD

14 1 5
Out[16]= ::y@xD C@1D I-1 + Cos@xD2 M LegendrePB , , Cos@xDF +
2 2
14 1 5
C@2D I-1 + Cos@xD2 M LegendreQB , , Cos@xDF>>
2 2

Occasionally second-order linear equations can be solved using only elementary functions.
In[17]:= DSolve@x ^ 2 y ''@xD + y@xD == 0, y@xD, xD
1 1
Out[17]= ::y@xD x C@1D CosB 3 Log@xDF + x C@2D SinB 3 Log@xDF>>
2 2

Beyond second order, the kinds of functions needed to solve even fairly simple linear differen-
tial equations become extremely complicated. At third order, the generalized Meijer G function
MeijerG can sometimes be used, but at fourth order and beyond absolutely no standard mathe-
matical functions are typically adequate, except in very special cases.

Here is a third-order linear differential equation which can be solved in terms of generalized
hypergeometric functions.
In[18]:= DSolve@y '''@xD + x y@xD == 0, y@xD, xD

Out[18]= ::y@xD
1 3 x4 1 3 5 x4
C@1D HypergeometricPFQB8<, : , >, - F+ x C@2D HypergeometricPFQB8<, : , >, - F+
2 4 64 2 2 4 4 64
1 5 3 x4
x2 C@3D HypergeometricPFQB8<, : , >, - F>>
8 4 2 64

This requires more general Meijer G functions.

In[19]:= DSolve@y '''@xD + Exp@xD y@xD == 0, y@xD, xD

Out[19]= 99y@xD C@1D HypergeometricPFQA8<, 81, 1<, -x E +

C@2D MeijerGA88<, 8<<, 880, 0<, 80<<, -x E + C@3D MeijerGA88<, 8<<, 880, 0, 0<, 8<<, x E==

For nonlinear differential equations, only rather special cases can usually ever be solved in
terms of standard mathematical functions. Nevertheless, DSolve includes fairly general proce-
dures which allow it to handle almost all nonlinear differential equations whose solutions are
found in standard reference books.
Mathematics and Algorithms 247

First-order nonlinear differential equations in which x does not appear on its own are fairly easy
to solve.
In[20]:= DSolve@y '@xD - y@xD ^ 2 == 0, y@xD, xD
1
Out[20]= ::y@xD >>
-x - C@1D

This Riccati equation already gives a significantly more complicated solution.

In[21]:= DSolve@y '@xD - y@xD ^ 2 == x, y@xD, xD FullSimplify

2 2 x32 2 2 x32
Out[21]= ::y@xD x -BesselJB- , F + BesselJB , F C@1D
3 3 3 3
1 2 x32 1 2 x32
BesselJB , F + BesselJB- , F C@1D >>
3 3 3 3

This Bernoulli equation, however, has a fairly simple solution.

In[22]:= DSolve@y '@xD - x y@xD ^ 2 - y@xD == 0, y@xD, xD
x
Out[22]= ::y@xD - >>
x x
- + x - C@1D

An nth order Bernoulli equation typically has n - 1 distinct solutions.

In[23]:= DSolve@y '@xD - x y@xD ^ 3 + y@xD == 0, y@xD, xD

2 2
Out[23]= ::y@xD - >, :y@xD >>
1 + 2 x + 2 2 x C@1D 1 + 2 x + 2 2 x C@1D

This Abel equation can be solved, but only implicitly.

In[24]:= DSolve@y '@xD + x y@xD ^ 3 + y@xD ^ 2 == 0, y@xD, xD

Solve::tdep :
The equations appear to involve the variables to be solved for in an essentially non-algebraic way.
-1-2 x y@xD
2 ArcTanhB F
1 5 -1 - x y@xD H-1 - x y@xDL
Out[24]= SolveB + LogB F C@1D - Log@xD, y@xDF
2 5 x2 y@xD2

In practical applications, it is quite often convenient to set up differential equations that involve
piecewise functions. You can use DSolve to find symbolic solutions to such equations.
248 Mathematics and Algorithms

This equation involves a piecewise forcing function.

In[25]:= DSolve@y '@xD - y@xD == UnitStep@xD, y@xD, xD

Out[25]=

Here the solution is explicitly broken into three cases.

In[26]:= DSolve@y '@xD + Clip@xD y@xD == 0, y@xD, xD
x x-1
1 x2
- - -1<x1
2 2
-x True
Out[26]= ::y@xD C@1D>>

Beyond ordinary differential equations, one can consider differential-algebraic equations that
involve a mixture of differential and algebraic equations.

This solves a differential-algebraic equation.

In[27]:= DSolve@8y '@xD + 3 z '@xD == 4 y@xD + 1 x, y@xD + z@xD == 1<, 8y@xD, z@xD<, xD
3 1
Out[27]= ::y@xD +I--2 x C@1D - 9 -2 x I3 2 x + ExpIntegralEi@2 xDMM,
218
1 1
z@xD - + I-2 x C@1D + 9 -2 x I3 2 x + ExpIntegralEi@2 xDMM>>
2 18

DSolve@eqn,y@x1 ,x2 ,D,8x1 ,x2 ,<D

solve a partial differential equation for y@x1 , x2 , D
DSolve@eqn,y,8x1 ,x2 ,<D solve a partial differential equation for the function y

Solving partial differential equations.

DSolve is set up to handle not only ordinary differential equations in which just a single indepen-
dent variable appears, but also partial differential equations in which two or more independent
variables appear.

This finds the general solution to a simple partial differential equation with two independent
variables.
In[28]:= DSolve@D@y@x1, x2D, x1D + D@y@x1, x2D, x2D == 1 Hx1 x2L, y@x1, x2D, 8x1, x2<D
1
Out[28]= ::y@x1, x2D H-Log@x1D + Log@x2D + x1 C@1D@-x1 + x2D - x2 C@1D@-x1 + x2DL>>
x1 - x2
Mathematics and Algorithms 249

Here is the result represented as a pure function.

In[29]:= DSolve@D@y@x1, x2D, x1D + D@y@x1, x2D, x2D == 1 Hx1 x2L, y, 8x1, x2<D
1
Out[29]= ::y FunctionB8x1, x2<, H-Log@x1D + Log@x2D + x1 C@1D@-x1 + x2D - x2 C@1D@-x1 + x2DLF>>
x1 - x2

The basic mathematics of partial differential equations is considerably more complicated than
that of ordinary differential equations. One feature is that whereas the general solution to an
ordinary differential equation involves only arbitrary constants, the general solution to a partial
differential equation, if it can be found at all, must involve arbitrary functions. Indeed, with m
independent variables, arbitrary functions of m - 1 arguments appear. DSolve by default names
these functions C@nD.

Here is a simple PDE involving three independent variables.

In[30]:= HD@, x1D + D@, x2D + D@, x3DL &@y@x1, x2, x3DD == 0

Out[30]= yH0,0,1L @x1, x2, x3D + yH0,1,0L @x1, x2, x3D + yH1,0,0L @x1, x2, x3D 0

The solution involves an arbitrary function of two variables.

In[31]:= DSolve@%, y@x1, x2, x3D, 8x1, x2, x3<D
Out[31]= 88y@x1, x2, x3D C@1D@-x1 + x2, -x1 + x3D<<

Here is the one-dimensional wave equation.

In[32]:= Hc ^ 2 D@, x, xD - D@, t, tDL &@y@x, tDD == 0

Out[32]= -yH0,2L @x, tD + c2 yH2,0L @x, tD 0

The solution to this second-order equation involves two arbitrary functions.

In[33]:= DSolve@%, y@x, tD, 8x, t<D

c2 x c2 x
Out[33]= ::y@x, tD C@1DBt - F + C@2DBt + F>>
c2 c2

For an ordinary differential equation, it is guaranteed that a general solution must exist, with
the property that adding initial or boundary conditions simply corresponds to forcing specific
choices for arbitrary constants in the solution. But for partial differential equations this is no
longer true. Indeed, it is only for linear partial differential and a few other special types that
such general solutions exist.

Other partial differential equations can be solved only when specific initial or boundary values
are given, and in the vast majority of cases no solutions can be found as exact formulas in
terms of standard mathematical functions.
250 Mathematics and Algorithms

Since y and its derivatives appear only linearly here, a general solution exists.
In[34]:= DSolve@x1 D@y@x1, x2D, x1D + x2 D@y@x1, x2D, x2D == Exp@x1 x2D, y@x1, x2D, 8x1, x2<D
1 x2
Out[34]= ::y@x1, x2D ExpIntegralEi@x1 x2D + 2 C@1DB F >>
2 x1

This weakly nonlinear PDE turns out to have a general solution.

In[35]:= DSolve@D@y@x1, x2D, x1D + D@y@x1, x2D, x2D == Exp@y@x1, x2DD, y@x1, x2D, 8x1, x2<D
Out[35]= 88y@x1, x2D -Log@-x1 - C@1D@-x1 + x2DD<<

Here is a nonlinear PDE which has no general solution.

In[36]:= DSolve@D@y@x1, x2D, x1D D@y@x1, x2D, x2D == a, y@x1, x2D, 8x1, x2<D

DSolve::nlpde :
Solution requested to nonlinear partial differential equation. Trying to build a complete integral.
a x1
Out[36]= ::y@x1, x2D C@1D + + x2 C@2D>>
C@2D

Integral Transforms and Related Operations

Laplace Transforms

LaplaceTransform@expr,t,sD the Laplace transform of expr

InverseLaplaceTransform@expr,s,tD
the inverse Laplace transform of expr

One-dimensional Laplace transforms.

The Laplace transform of a function f HtL is given by 0 f HtL e-s t t. The inverse Laplace transform
1 g+i
of FHsL is given for suitable g by 2 p i g-i
FHsL est s.
Mathematics and Algorithms 251

Here is a simple Laplace transform.

In[1]:= LaplaceTransform@t ^ 4 Sin@tD, t, sD
24 I1 - 10 s2 + 5 s4 M
Out[1]=
5
I1 + s2 M

Here is the inverse.

In[2]:= InverseLaplaceTransform@%, s, tD

Out[2]= t4 Sin@tD

Even simple transforms often involve special functions.

In[3]:= LaplaceTransform@1 H1 + t ^ 2L, t, sD
1
Out[3]= CosIntegral@sD Sin@sD + Cos@sD Hp - 2 SinIntegral@sDL
2

Here the result involves a Meijer G function.

In[4]:= LaplaceTransform@1 H1 + t ^ 3L, t, sD
2 1 2 2 s3
MeijerGB:: >, 8<>, ::0, , , >, 8<>, F
3 3 3 3 27
Out[4]=
2 3 p

InverseLaplaceTransform returns the original function.

In[6]:= InverseLaplaceTransform@%, s, tD
1
Out[6]=
1 + t3

The Laplace transform of this Bessel function just involves elementary functions.
In[5]:= LaplaceTransform@BesselJ@n, tD, t, sD
-n
s+ 1 + s2
Out[5]=
1 + s2

Laplace transforms have the property that they turn integration and differentiation into essen-
tially algebraic operations. They are therefore commonly used in studying systems governed by
differential equations.
252 Mathematics and Algorithms

Integration becomes multiplication by 1 s when one does a Laplace transform.

In[6]:= LaplaceTransform@Integrate@f@uD, 8u, 0, t<D, t, sD
LaplaceTransform@f@tD, t, sD
Out[6]=
s

LaplaceTransform@expr,8t1 ,t2 ,<,8s1 ,s2 ,<D

the multidimensional Laplace transform of expr
InverseLaplaceTransform@expr,8s1 ,s2 ,<,8t1 ,t2 ,<D
the multidimensional inverse Laplace transform of expr

Multidimensional Laplace transforms.

Fourier Transforms

FourierTransform@expr,t,wD the Fourier transform of expr

InverseFourierTransform@expr,w,tD
the inverse Fourier transform of expr

One-dimensional Fourier transforms.

Integral transforms can produce results that involve "generalized functions" such as
HeavisideTheta.
In[1]:= FourierTransform@1 H1 + t ^ 4L, t, wD
I1+M w
1 -
2 w
Out[1]= + 2 p - + 2 w
HeavisideTheta@-wD + 1 - 2 w
HeavisideTheta@wD
4 4

This finds the inverse.

In[2]:= InverseFourierTransform@%, w, tD
1
Out[2]=
1 + t4

In Mathematica the Fourier transform of a function f HtL is by default defined to be

1
- f HtL e
iwt
t. The inverse Fourier transform of FHwL is similarly defined as
2p

1 -i w t
- FHwL e w.
2p

In different scientific and technical fields different conventions are often used for defining
Fourier transforms. The option FourierParameters in Mathematica allows you to choose any of
these conventions you want.
Mathematics and Algorithms 253

common convention setting Fourier transform inverse Fourier transform

1 iwt 1 -i w t
Mathematica default 80, 1< - f HtL e t - FHwL e w
2p 2p

-i w t 1
pure mathematics 81, - 1< - f HtL e t 2 p -
FHwL ei w t w
1
classical physics 8- 1, 1< 2 p -
f HtL ei w t t - FHwL e
-i w t
w
1 iwt 1 -i w t
modern physics 80, 1< - f HtL e t - FHwL e w
2p 2p

-i w t 1
systems engineering 81, - 1< - f HtL e t 2 p -
FHwL ei w t w
-2 p i w t 2piwt
signal processing 80, - 2 Pi< - f HtL e t - FHwL e w

b
general case 8a, b< H b L H2 pL1-a - FHwL e
-i b w t
w
H2 pL1+a
ibwt
- f HtL e t

Typical settings for FourierParameters with various conventions.

Here is a Fourier transform with the default choice of parameters.

In[3]:= FourierTransform@Exp@- t ^ 2D, t, wD
w2
-
4

Out[3]=
2

Here is the same Fourier transform with the choice of parameters typically used in signal
processing.
In[4]:= FourierTransform@Exp@- t ^ 2D, t, w, FourierParameters -> 80, - 2 Pi<D
2
w2
Out[4]= -p p

FourierSinTransform@expr,t,wD
Fourier sine transform
FourierCosTransform@expr,t,wD
Fourier cosine transform
254 Mathematics and Algorithms

InverseFourierSinTransform@expr,w,tD
inverse Fourier sine transform
InverseFourierCosTransform@expr,w,tD
inverse Fourier cosine transform

Fourier sine and cosine transforms.

In some applications of Fourier transforms, it is convenient to avoid ever introducing complex

exponentials. Fourier sine and cosine transforms correspond to integrating respectively with
sinHw tL and cosHw tL instead of expHi w tL, and using limits 0 and rather than - and .

Here are the Fourier sine and cosine transforms of e-t .

In[5]:= 8FourierSinTransform@Exp@- tD, t, wD, FourierCosTransform@Exp@- tD, t, wD<

2 2
w
p p
Out[5]= : , >
2
1+w 1 + w2

FourierTransform@expr,8t1 ,t2 ,<,8w1 ,w2 ,<D

the multidimensional Fourier transform of expr
InverseFourierTransform@expr,8w1 ,w2 ,<,8t1 ,t2 ,<D
the multidimensional inverse Fourier transform of expr
FourierSinTransform@expr,8t1 ,t2 ,<,8w1 ,w2 ,<D ,
FourierCosTransform@expr,8t1 ,t2 ,<,8w1 ,w2 ,<D
the multidimensional sine and cosine Fourier transforms of
expr
InverseFourierSinTransform@expr,8w1 ,w2 ,<,8t1 ,t2 ,<D ,
InverseFourierCosTransform@expr,8w1 ,w2 ,<,8t1 ,t2 ,<D
the multidimensional inverse Fourier sine and cosine
transforms of expr

Multidimensional Fourier transforms.

This evaluates a two-dimensional Fourier transform.

In[6]:= FourierTransform@Hu vL ^ 2 Exp@- u ^ 2 - v ^ 2D, 8u, v<, 8a, b<D
1 -
a2
-
b2

Out[6]= I-2 + a2 M I-2 + b2 M 4 4

This inverts the transform.

In[7]:= InverseFourierTransform@%, 8a, b<, 8u, v<D
2
-v2
Out[7]= -u u2 v2
Mathematics and Algorithms 255

Z Transforms

ZTransform@expr,n,zD Z transform of expr

InverseZTransform @expr,z,nD inverse Z transform of expr

Z transforms.

The Z transform of a function f HnL is given by

n=0 f HnL z . The inverse Z transform of FHzL is given
-n

1
by the contour integral 2pi
n-1
FHzL z z. Z transforms are effectively discrete analogs of Laplace

transforms. They are widely used for solving difference equations, especially in digital signal
processing and control theory. They can be thought of as producing generating functions, of the
kind commonly used in combinatorics and number theory.

This computes the Z transform of 2-n .

In[1]:= ZTransform@2 ^ - n, n, zD
2z
Out[1]=
-1 + 2 z

Here is the inverse Z transform.

In[2]:= InverseZTransform@%, z, nD
Out[2]= 2-n

The generating function for 1 n ! is an exponential function.

In[3]:= ZTransform@1 n !, n, zD
1

Out[3]= z

Generalized Functions and Related Objects

In many practical situations it is convenient to consider limits in which a fixed amount of some-
thing is concentrated into an infinitesimal region. Ordinary mathematical functions of the kind
normally encountered in calculus cannot readily represent such limits. However, it is possible to
introduce generalized functions or distributions which can represent these limits in integrals and
other types of calculations.
256 Mathematics and Algorithms

DiracDelta@xD Dirac delta function dHxL

HeavisideTheta@xD Heaviside theta function qHxL, equal to 0 for x < 0 and 1 for
x>0

Dirac delta and Heaviside theta functions.

Here is a function concentrated around x = 0.

In[1]:= Plot@Sqrt@50 PiD Exp@- 50 x ^ 2D, 8x, - 2, 2<, PlotRange -> AllD
4

Out[1]=
2

-2 -1 1 2

As n gets larger, the functions become progressively more concentrated.

In[2]:= Plot@Evaluate@Sqrt@n PiD Exp@- n x ^ 2D . n -> 81, 10, 100<D,
8x, - 2, 2<, PlotRange -> AllD

5
4
Out[2]=
3
2
1

2 1 1 2

For any n > 0, their integrals are nevertheless always equal to 1.

In[3]:= Integrate@Sqrt@n PiD Exp@- n x ^ 2D, 8x, - Infinity, Infinity<, Assumptions -> n > 0D
Out[3]= 1

The limit of the functions for infinite n is effectively a Dirac delta function, whose integral is
again 1.
In[4]:= Integrate@DiracDelta@xD, 8x, - Infinity, Infinity<D
Out[4]= 1

DiracDelta evaluates to 0 at all real points except x = 0.

In[5]:= Table@DiracDelta@xD, 8x, - 3, 3<D
Out[5]= 80, 0, 0, DiracDelta@0D, 0, 0, 0<

Inserting a delta function in an integral effectively causes the integrand to be sampled at dis-
crete points where the argument of the delta function vanishes.
Mathematics and Algorithms 257

This samples the function f with argument 2.

In[6]:= Integrate@DiracDelta@x - 2D f@xD, 8x, - 4, 4<D
Out[6]= f@2D

Here is a slightly more complicated example.

In[7]:= Integrate@DiracDelta@x ^ 2 - x - 1D, 8x, 0, 2<D
1
Out[7]=
5

This effectively counts the number of zeros of cosHxL in the region of integration.
In[8]:= Integrate@DiracDelta@Cos@xDD, 8x, - 30, 30<D
Out[8]= 20

The Heaviside function HeavisideTheta@xD is the indefinite integral of the delta function. It is
variously denoted HHxL, qHxL, mHxL, and UHxL. As a generalized function, the Heaviside function is
defined only inside an integral. This distinguishes it from the unit step function UnitStep@xD,
which is a piecewise function.

The indefinite integral of the delta function is the Heaviside theta function.
In[9]:= Integrate@DiracDelta@xD, xD
Out[9]= HeavisideTheta@xD

The value of this integral depends on whether a lies in the interval H-2, 2L.
In[10]:= Integrate@f@xD DiracDelta@x - aD, 8x, - 2, 2<, Assumptions -> a RealsD
Out[10]= f@aD HeavisideTheta@2 - aD HeavisideTheta@2 + aD

DiracDelta and HeavisideTheta often arise in doing integral transforms.

The Fourier transform of a constant function is a delta function.

In[11]:= FourierTransform@1, t, wD

Out[11]= 2 p DiracDelta@wD
258 Mathematics and Algorithms

The Fourier transform of cosHtL involves the sum of two delta functions.
In[12]:= FourierTransform@Cos@tD, t, wD

p p
Out[12]= DiracDelta@-1 + wD + DiracDelta@1 + wD
2 2

Dirac delta functions can be used in DSolve to find the impulse response or Green's function of
systems represented by linear and certain other differential equations.

This finds the behavior of a harmonic oscillator subjected to an impulse at t = 0.

In[13]:= DSolve@8x ''@tD + r x@tD == DiracDelta@tD, x@0D == 0, x '@0D == 1<, x@tD, tD

HeavisideTheta@tD SinB r tF
Out[13]= ::x@tD >>
r

DiracDelta@x1 ,x2 ,D multidimensional Dirac delta function

HeavisideTheta@x1 ,x2 ,D multidimensional Heaviside theta function

Multidimensional Dirac delta and Heaviside theta functions.

Multidimensional generalized functions are essentially products of univariate generalized

functions.

Here is a derivative of a multidimensional Heaviside function.

In[14]:= D@HeavisideTheta@x, y, zD, xD
Out[14]= DiracDelta@xD HeavisideTheta@y, zD

Related to the multidimensional Dirac delta function are two integer functions: discrete delta
and Kronecker delta. Discrete delta dHn1 , n2 , L is 1 if all the ni = 0, and is zero otherwise. Kro-
necker delta dn1 n2 is 1 if all the ni are equal, and is zero otherwise.

DiscreteDelta@n1 ,n2 ,D discrete delta dHn1 , n2 , L

KroneckerDelta@n1 ,n2 ,D Kronecker delta dn1 n2

Integer delta functions.

Mathematics and Algorithms 259

Numerical Operations on Functions

Arithmetic
You can do arithmetic with Mathematica just as you would on an electronic calculator.

This is the sum of two numbers.

In[1]:= 2.3 + 5.63
Out[1]= 7.93

Here the stands for division, and the ^ stands for power.
In[2]:= 2.4 8.9 ^ 2
Out[2]= 0.0302992

Spaces denote multiplication in Mathematica. The front end automatically replaces spaces
between numbers with light gray multiplication signs.
In[3]:= 234
Out[3]= 24

You can use a * for multiplication if you want to.

In[4]:= 2*3*4
Out[4]= 24

You can type arithmetic expressions with parentheses.

In[5]:= H3 + 4L ^ 2 - 2 H3 + 1L
Out[5]= 41

Spaces are not needed, though they often make your input easier to read.
In[6]:= H3 + 4L ^ 2 - 2 H3 + 1L
Out[6]= 41
260 Mathematics and Algorithms

x^y power
-x minus
xy divide
x y z or x*y*z multiply
x+y+z add

Arithmetic operations in Mathematica.

Arithmetic operations in Mathematica are grouped according to the standard mathematical

conventions. As usual, 2 ^ 3 + 4, for example, means H2 ^ 3L + 4, and not 2 ^ H3 + 4L. You can
always control grouping by explicitly using parentheses.

This result is given in scientific notation.

In[7]:= 2.4 ^ 45

Out[7]= 1.28678 1017

You can enter numbers in scientific notation like this.

In[8]:= 2.3 10 ^ 70

Out[8]= 2.3 1070

Or like this.
In[9]:= 2.3*^70

Out[9]= 2.3 1070

With Mathematica, you can perform calculations with a particular precision, usually higher than
an ordinary calculator. When given precise numbers, Mathematica does not convert them to an
approximate representation, but gives a precise result.

This gives the result in terms of rational numbers.

In[10]:= 13+27
13
Out[10]=
21

This gives the approximate numerical result.

In[11]:= 1 3 + 2 7 N
Out[11]= 0.619048

This gives the approximate numerical result with 40 significant digits.

Mathematics and Algorithms 261

This gives the approximate numerical result with 40 significant digits.

In[12]:= N@1 3 + 2 7, 40D
Out[12]= 0.6190476190476190476190476190476190476190

Numerical Mathematics in Mathematica

One of the important features of Mathematica is its ability to give you exact, symbolic, results
for computations. There are, however, computations where it is just mathematically impossible
to get exact "closed form" results. In such cases, you can still often get approximate numerical
results.

1
There is no "closed form" result for 0 sin Hsin HxLL x. Mathematica returns the integral in
symbolic form.
In[1]:= Integrate@Sin@Sin@xDD, 8x, 0, 1<D
1
Out[1]= Sin@Sin@xDD x
0

You can now take the symbolic form of the integral, and ask for its approximate numerical
value.
In[2]:= N@%D
Out[2]= 0.430606

When Mathematica cannot find an explicit result for something like a definite integral, it returns
a symbolic form. You can take this symbolic form, and try to get an approximate numerical
value by applying N.

By giving a second argument to N , you can specify the numerical precision to use.
In[3]:= N@Integrate@Sin@Sin@xDD, 8x, 0, 1<D, 40D
Out[3]= 0.4306061031206906049123773552484657864336

If you want to evaluate an integral numerically in Mathematica, then using Integrate and
applying N to the result is not the most efficient way to do it. It is better instead to use the
function NIntegrate, which immediately gives a numerical answer, without first trying to get
an exact, symbolic, result. You should realize that even when Integrate does not in the end
manage to give you an exact result, it may spend a lot of time trying to do so.
262 Mathematics and Algorithms

NIntegrate evaluates numerical integrals directly, without first trying to get a symbolic result.
In[4]:= NIntegrate@Sin@Sin@xDD, 8x, 0, 1<D
Out[4]= 0.430606

Integrate NIntegrate definite integrals

Sum NSum sums
Product NProduct products
Solve NSolve solutions of algebraic equations
DSolve NDSolve solutions of differential equations
Maximize NMaximize maximization

Symbolic and numerical versions of some Mathematica functions.

The Uncertainties of Numerical Mathematics

Mathematica does operations like numerical integration very differently from the way it does
their symbolic counterparts.

When you do a symbolic integral, Mathematica takes the functional form of the integrand you
have given, and applies a sequence of exact symbolic transformation rules to it, to try and
evaluate the integral.

However, when Mathematica does a numerical integral, after some initial symbolic preprocess-
ing, the only information it has about your integrand is a sequence of numerical values for it. To
get a definite result for the integral, Mathematica then effectively has to make certain assump-
tions about the smoothness and other properties of your integrand. If you give a sufficiently
pathological integrand, these assumptions may not be valid, and as a result, Mathematica may
simply give you the wrong answer for the integral.

This problem may occur, for example, if you try to integrate numerically a function which has a
very thin spike at a particular position. Mathematica samples your function at a number of
points, and then assumes that the function varies smoothly between these points. As a result, if
none of the sample points come close to the spike, then the spike will go undetected, and its
contribution to the numerical integral will not be correctly included.
Mathematics and Algorithms 263

Here is a plot of the function expI-x2 M.

In[1]:= Plot@Exp@- x ^ 2D, 8x, - 10, 10<, PlotRange -> AllD
1.0

0.8

Out[1]= 0.6

0.4

0.2

-10 -5 5 10

NIntegrate gives the correct answer for the numerical integral of this function from -10 to
+10.
In[2]:= NIntegrate@Exp@- x ^ 2D, 8x, - 10, 10<D
Out[2]= 1.77245

If, however, you ask for the integral from -10000 to 10000, with its default settings
NIntegrate will miss the peak near x = 0, and give the wrong answer.
In[3]:= NIntegrate@Exp@- x ^ 2D, 8x, - 10 000, 10 000<D

NIntegrate::ncvb : NIntegrate failed to converge to prescribed accuracy after 10 recursive bisections in x

near 8x< = 86670.<. NIntegrate obtained 0.` and 0.` for the integral and error estimates.
Out[3]= 0.

NIntegrate tries to make the best possible use of the information that it can get about the
numerical values of the integrand. Thus, for example, by default, if NIntegrate notices that the
estimated error in the integral in a particular region is large, it will take more samples in that
region. In this way, NIntegrate tries to "adapt" its operation to the particular integrand you
have given.

The kind of adaptive procedure that NIntegrate uses is similar, at least in spirit, to what Plot
does in trying to draw smooth curves for functions. In both cases, Mathematica tries to go on
taking more samples in a particular region until it has effectively found a smooth approximation
to the function in that region.

The kinds of problems that can appear in numerical integration can also arise in doing other
numerical operations on functions.

For example, if you ask for a numerical approximation to the sum of an infinite series, Mathemat -
ica samples a certain number of terms in the series, and then does an extrapolation to estimate
the contributions of other terms. If you insert large terms far out in the series, they may not be
detected when the extrapolation is done, and the result you get for the sum may be incorrect.

A similar problem arises when you try to find a numerical approximation to the minimum of a
function. Mathematica samples only a finite number of values, then effectively assumes that the
actual function interpolates smoothly between these values. If in fact the function has a sharp
dip in a particular region, then Mathematica may miss this dip, and you may get the wrong
answer for the minimum.

If you work only with numerical values of functions, there is simply no way to avoid the kinds of
problems we have been discussing. Exact symbolic computation, of course, allows you to get
around these problems.

In many calculations, it is therefore worthwhile to go as far as you can symbolically, and then
resort to numerical methods only at the very end. This gives you the best chance of avoiding
the problems that can arise in purely numerical computations.

Introduction to Numerical Sums, Products and

Integrals

NSumA f ,9i,imin ,Infinity=E numerical approximation to

imin f

NProductA f ,9i,imin ,Infinity=E numerical approximation to

imin f

xmax
NIntegrate@ f ,8x,xmin ,xmax <D numerical approximation to x f dx
min

NIntegrate@ f ,8x,xmin ,xmax <,8y,ymin ,ymax <D

xmax ymax
the multiple integral x d x y d y f
min min

Numerical sums, products, and integrals.

1
Here is a numerical approximation to
i=1 .
i3

In[1]:= NSum@1 i ^ 3, 8i, 1, Infinity<D

Out[1]= 1.20206
Mathematics and Algorithms 265

NIntegrate can handle singularities in the integration region.

In[2]:= NIntegrate@1 Sqrt@x H1 - xLD, 8x, 0, 1<D
Out[2]= 3.14159

You can do numerical integrals over infinite regions.

In[3]:= NIntegrate@Exp@- x ^ 2D, 8x, - Infinity, Infinity<D
Out[3]= 1.77245

Here is a double integral over a triangular domain. Note the order in which the variables are
given.
In[4]:= NIntegrate@Sin@x yD, 8x, 0, 1<, 8y, 0, x<D
Out[4]= 0.119906

Here is a double integral over a more complicated domain.

In[5]:= NIntegrate@ Sin@x yD, 8x, 0, 1<, 8y, 0, Sqrt@x ^ 3 + 3D<D
Out[5]= 0.727332

Numerical Integration

N@Integrate@expr,8x,xmin ,xmax <DD try to perform an integral exactly, then find numerical
approximations to the parts that remain
NIntegrate@expr,8x,xmin ,xmax <D find a numerical approximation to an integral
NIntegrate@expr,8x,xmin ,xmax <,8y,ymin ,ymax <,D
xmax ymax
multidimensional numerical integral x d x y d y expr
min min

NIntegrate@expr,8x,xmin ,x1 ,x2 ,,xmax <D

do a numerical integral along a line, starting at xmin , going
through the points xi , and ending at xmax

Numerical integration functions.

3
This finds a numerical approximation to the integral 0 e-x x.

In[1]:= NIntegrate@Exp@- x ^ 3D, 8x, 0, Infinity<D

Out[1]= 0.89298

1 1
Here is the numerical value of the double integral -1 d x -1 d y Ix2 + y2 M.
266 Mathematics and Algorithms

1 1
Here is the numerical value of the double integral -1 d x -1 d y Ix2 + y2 M.
In[2]:= NIntegrate@x ^ 2 + y ^ 2, 8x, - 1, 1<, 8y, - 1, 1<D
Out[2]= 2.66667

An important feature of NIntegrate is its ability to deal with functions that "blow up" at known
points. NIntegrate automatically checks for such problems at the endpoints of the integration
region.

The function 1 x blows up at x = 0, but NIntegrate still succeeds in getting the correct
value for the integral.
In[3]:= NIntegrate@1 Sqrt@xD, 8x, 0, 1<D
Out[3]= 2.

Mathematica can find the integral of 1 x exactly.

In[4]:= Integrate@1 Sqrt@xD, 8x, 0, 1<D
Out[4]= 2

NIntegrate detects that the singularity in 1 x at x = 0 is not integrable.

In[5]:= NIntegrate@1 x, 8x, 0, 1<D

NIntegrate::slwcon :
Numerical integration converging too slowly; suspect one of the following: singularity, value
of the integration is 0, highly oscillatory integrand, or WorkingPrecision too small.

NIntegrate::ncvb :
NIntegrate failed to converge to prescribed accuracy after 9 recursive bisections in x near
8x< = 91.22413 10-225 =. NIntegrate obtained 191612.2902185145`
and 160378.51781028978` for the integral and error estimates.
Out[5]= 191 612.

NIntegrate does not automatically look for singularities except at the endpoints of your integra-
tion region. When other singularities are present, NIntegrate may not give you the right
answer for the integral. Nevertheless, in following its adaptive procedure, NIntegrate will often
detect the presence of potentially singular behavior, and will warn you about it.
Mathematics and Algorithms 267

NIntegrate warns you of a possible problem due to the singularity in the middle of the integra-
tion region. The final result is numerically quite close to the correct answer.
In[6]:= NIntegrate@x ^ 2 Sin@1 xD, 8x, - 1, 2<D

If you know that your integrand has singularities at particular points, you can explicitly tell
NIntegrate to deal with them. NIntegrate@expr, 8x, xmin , x1 , x2 , , xmax <D integrates expr from
xmin to xmax , looking for possible singularities at each of the intermediate points xi .

This gives the same integral, but now explicitly deals with the singularity at x = 0.
In[7]:= NIntegrate@x ^ 2 Sin@1 xD, 8x, - 1, 0, 2<D
Out[7]= 1.38755

You can also use the list of intermediate points xi in NIntegrate to specify an integration con-
tour to follow in the complex plane. The contour is taken to consist of a sequence of line seg-
ments, starting at xmin , going through each of the xi , and ending at xmax .

This integrates 1 x around a closed contour in the complex plane, going from -1, through the
points -i, 1 and i, then back to -1.
In[8]:= NIntegrate@1 x, 8x, - 1, - I, 1, I, - 1<D
Out[8]= 0. + 6.28319

The integral gives 2 p i, as expected from Cauchy's theorem.

In[9]:= N@2 Pi ID
Out[9]= 0. + 6.28319

option name default value

MinRecursion 0 minimum number of recursions for the
integration method
MaxRecursion Automatic maximum number of recursions for the
integration method
MaxPoints Automatic maximum total number of times to sample
the integrand

Special options for NIntegrate .

When NIntegrate tries to evaluate a numerical integral, it samples the integrand at a sequence
of points. If it finds that the integrand changes rapidly in a particular region, then it recursively
takes more sample points in that region. The parameters MinRecursion and MaxRecursion
specify the minimum and maximum number of recursions to use. Increasing the value of
MinRecursion guarantees that NIntegrate will use a larger number of sample points.
MaxPoints and MaxRecursion limit the number of sample points which NIntegrate will ever try
to use. Increasing MinRecursion or MaxRecursion will make NIntegrate work more slowly.

With the default settings for all options, NIntegrate misses the peak in expI-Hx - 1L2 M near
x = 1, and gives the wrong answer for the integral.
In[10]:= NIntegrate@Exp@- Hx - 1L ^ 2D, 8x, - 1000, 1000<D

NIntegrate::ncvb :
NIntegrate failed to converge to prescribed accuracy after 9 recursive bisections in x near
8x< = 83.87517<. NIntegrate obtained 1.6330510571683285` and
0.004736564243403896` for the integral and error estimates.
Out[10]= 1.63305

With the option MinRecursion -> 3, NIntegrate samples enough points that it notices the
peak around x = 1. With the default setting of MaxRecursion , however, NIntegrate cannot
use enough sample points to be able to expect an accurate answer.
In[11]:= NIntegrate@Exp@- Hx - 1L ^ 2D, 8x, - 50 000, 1000<, MinRecursion 3D

NIntegrate::ncvb :
NIntegrate failed to converge to prescribed accuracy after 9 recursive bisections in x near
8x< = 8-8.44584<. NIntegrate obtained 1.8181913371063452`
and 1.165089629798181` for the integral and error estimates.
Out[11]= 1.81819

With this setting of MaxRecursion , NIntegrate can get an accurate answer for the integral.
In[12]:= NIntegrate@Exp@- Hx - 1L ^ 2D, 8x, - 50 000, 1000<, MinRecursion 3, MaxRecursion 20D

Another way to solve the problem is to make NIntegrate break the integration region into
several pieces, with a small piece that explicitly covers the neighborhood of the peak.
Mathematics and Algorithms 269

Another way to solve the problem is to make NIntegrate break the integration region into
several pieces, with a small piece that explicitly covers the neighborhood of the peak.
In[13]:= NIntegrate@Exp@- Hx - 1L ^ 2D, 8x, - 1000, - 10, 10, 1000<D
Out[13]= 1.77245

For integrals in many dimensions, it can take a long time for NIntegrate to get a precise
answer. However, by setting the option MaxPoints, you can tell NIntegrate to give you just a
rough estimate, sampling the integrand only a limited number of times.

Here is a way to get a rough estimate for an integral that takes a long time to compute.
In[14]:= NIntegrate@1 Sqrt@x + Log@y + zD ^ 2D,
8x, 0, 1<, 8y, 0, 1<, 8z, 0, 1<, MaxPoints -> 1000D
NIntegrate::maxp : The integral failed to converge after 1023 integrand evaluations. NIntegrate obtained
1.4548878649546768` and 0.03247010762528413` for the integral and error estimates.
Out[14]= 1.45489

Numerical Evaluation of Sums and Products

imax
NSum@ f ,8i,imin ,imax <D find a numerical approximation to the sum i=i f
min

NSum@ f ,8i,imin ,imaxdi <D use step di in the sum

imax
NProduct@ f ,8i,imin ,imax <D find a numerical approximation to the product i=i f
min

Numerical sums and products.

1
This gives a numerical approximation to
i=1 .
i3 +i!

In[1]:= NSum@1 Hi ^ 3 + i !L, 8i, 1, Infinity<D

Out[1]= 0.64703

There is no exact result for this sum, so Mathematica leaves it in a symbolic form.
In[2]:= Sum@1 Hi ^ 3 + i !L, 8i, 1, Infinity<D
1
Out[2]=
3
i=1 i + i!
270 Mathematics and Algorithms

You can apply N explicitly to get a numerical result.

In[3]:= N@%D
Out[3]= 0.64703

The way NSum works is to include a certain number of terms explicitly, and then to try and
estimate the contribution of the remaining ones. There are three approaches to estimating this
contribution. The first uses the Euler|Maclaurin method, and is based on approximating the sum
by an integral. The second method, known as the Wynn epsilon method, samples a number of
additional terms in the sum, and then tries to fit them to a polynomial multiplied by a decaying
exponential. The third approach, useful for alternating series, uses an alternating signs method;
it also samples a number of additional terms and approximates their sum by the ratio of two
polynomials (Pad approximation).

option name default value

Method Automatic Automatic, "EulerMaclaurin",
"WynnEpsilon", or
"AlternatingSigns"
NSumTerms 15 number of terms to include explicitly
VerifyConvergence True whether the convergence of the series
should be verified

Special options for NSum .

If you do not explicitly specify the method to use, NSum will try to choose between the
EulerMaclaurin or WynnEpsilon methods. In any case, some implicit assumptions about the
functions you are summing have to be made. If these assumptions are not correct, you may get
inaccurate answers.

The most common place to use NSum is in evaluating sums with infinite limits. You can, how-
ever, also use it for sums with finite limits. By making implicit assumptions about the objects
you are evaluating, NSum can often avoid doing as many function evaluations as an explicit Sum
computation would require.

100
This finds the numerical value of n=0 e-n by extrapolation techniques.
In[4]:= NSum@Exp@- nD, 8n, 0, 100<D
Out[4]= 1.58198
Mathematics and Algorithms 271

You can also get the result, albeit much less efficiently, by constructing the symbolic form of the
sum, then evaluating it numerically.
In[5]:= Sum@Exp@- nD, 8n, 0, 100<D N
Out[5]= 1.58198

NProduct works in essentially the same way as NSum, with analogous options.

Numerical Equation Solving

NSolve@lhs==rhs,xD solve a polynomial equation numerically

NSolve@8lhs1 ==rhs1 ,lhs2 ==rhs2 ,<,8x,y,<D
solve a system of polynomial equations numerically
FindRoot@lhs==rhs,8x,x0 <D search for a numerical solution to an equation, starting at
x = x0
FindRoot@8lhs1 ==rhs1 ,lhs2 ==rhs2 ,<,88x,x0 <,8y,y0 <,<D
search for numerical solutions to simultaneous equations

Numerical root finding.

NSolve gives you numerical approximations to all the roots of a polynomial equation.
In[1]:= NSolve@x ^ 5 + x + 1 == 0, xD
Out[1]= 88x -0.754878<, 8x -0.5 - 0.866025 <, 8x -0.5 + 0.866025 <,
8x 0.877439 - 0.744862 <, 8x 0.877439 + 0.744862 <<

You can also use NSolve to solve sets of simultaneous equations numerically.
In[2]:= NSolve@8x + y == 2, x - 3 y + z == 3, x - y + z == 0<, 8x, y, z<D
Out[2]= 88x 3.5, y -1.5, z -5.<<

If your equations involve only linear functions or polynomials, then you can use NSolve to get
numerical approximations to all the solutions. However, when your equations involve more
complicated functions, there is in general no systematic procedure for finding all solutions, even
numerically. In such cases, you can use FindRoot to search for solutions. You have to give
FindRoot a place to start its search.

This searches for a numerical solution, starting at x = 1.

In[3]:= FindRoot@3 Cos@xD == Log@xD, 8x, 1<D
Out[3]= 8x 1.44726<

The equation has several solutions. If you start at a different x, FindRoot may return a differ-
ent solution.
272 Mathematics and Algorithms

The equation has several solutions. If you start at a different x, FindRoot may return a differ-
ent solution.
In[4]:= FindRoot@3 Cos@xD == Log@xD, 8x, 10<D
Out[4]= 8x 13.1064<

You can search for solutions to sets of equations. Here the solution involves complex numbers.
In[5]:= FindRoot@8x == Log@yD, y == Log@xD<, 88x, I<, 8y, 2<<D
Out[5]= 8x 0.318132 + 1.33724 , y 0.318132 + 1.33724 <

Numerical Solution of Polynomial Equations

When Solve cannot find solutions in terms of radicals to polynomial equations, it returns a
symbolic form of the result in terms of Root objects.
In[1]:= Solve@x ^ 5 + 7 x + 1 == 0, xD

Out[1]= 99x RootA1 + 7 1 + 15 &, 1E=, 9x RootA1 + 7 1 + 15 &, 2E=,

9x RootA1 + 7 1 + 15 &, 3E=, 9x RootA1 + 7 1 + 15 &, 4E=, 9x RootA1 + 7 1 + 15 &, 5E==

You can get numerical solutions by applying N .

In[2]:= N@%D
Out[2]= 88x -0.142849<, 8x -1.11308 - 1.15173 <,
8x -1.11308 + 1.15173 <, 8x 1.1845 - 1.15139 <, 8x 1.1845 + 1.15139 <<

This gives the numerical solutions to 25-digit precision.

In[3]:= N@%%, 25D
Out[3]= 88x -0.1428486455250044341134116<,
8x -1.113077976547710735600398 - 1.151734362151674305046770 <,
8x -1.113077976547710735600398 + 1.151734362151674305046770 <,
8x 1.184502299310212952657104 - 1.151390075408837074699147 <,
8x 1.184502299310212952657104 + 1.151390075408837074699147 <<

You can use NSolve to get numerical solutions to polynomial equations directly, without first
trying to find exact results.
In[4]:= NSolve@x ^ 7 + x + 1 == 0, xD
Out[4]= 88x -0.796544<, 8x -0.705298 - 0.637624 <,
8x -0.705298 + 0.637624 <, 8x 0.123762 - 1.05665 <,
8x 0.123762 + 1.05665 <, 8x 0.979808 - 0.516677 <, 8x 0.979808 + 0.516677 <<
Mathematics and Algorithms 273

NSolve@poly==0,xD get approximate numerical solutions to a polynomial

equation
NSolve@poly==0,x,nD get solutions using n-digit precision arithmetic
NSolve@8eqn1 ,eqn2 ,<,8var1 ,var2 ,<D
get solutions to a polynomial system

Numerical solution of polynomial equations and systems.

NSolve will give you the complete set of numerical solutions to any polynomial equation or
system of polynomial equations.

NSolve can find solutions to sets of simultaneous polynomial equations.

In[5]:= NSolve@8x ^ 2 + y ^ 2 == 1, x ^ 3 + y ^ 3 == 2<, 8x, y<D
Out[5]= 88x -1.09791 - 0.839887 , y -1.09791 + 0.839887 <,
8x -1.09791 + 0.839887 , y -1.09791 - 0.839887 <,
8x 1.22333 + 0.0729987 , y -0.125423 + 0.712005 <,
8x 1.22333 - 0.0729987 , y -0.125423 - 0.712005 <,
8x -0.125423 - 0.712005 , y 1.22333 - 0.0729987 <,
8x -0.125423 + 0.712005 , y 1.22333 + 0.0729987 <<

Numerical Root Finding

NSolve gives you a general way to find numerical approximations to the solutions of polynomial
equations. Finding numerical solutions to more general equations, however, can be much more
difficult, as discussed in "Equations in One Variable". FindRoot gives you a way to search for a
numerical root of a function or a numerical solution to an arbitrary equation, or set of equations.

FindRoot@ f ,8x,x0 <D search for a numerical root of f , starting with x = x0

FindRoot@lhs==rhs,8x,x0 <D search for a numerical solution to the equation lhs == rhs,
starting with x = x0

FindRoot@ f1 , f2 ,,88x,x0 <,8y,y0 <,<D

search for a simultaneous numerical root of all the fi

FindRoot@8eqn1 ,eqn2 ,<,88x,x0 <,8y,y0 <,<D

search for a numerical solution to the simultaneous equa -
tions eqni

Numerical root finding.

274 Mathematics and Algorithms

The curves for cos HxL and x intersect at one point.

In[1]:= Plot@8Cos@xD, x<, 8x, - 1, 1<D
1.0

0.5

Out[1]=
-1.0 -0.5 0.5 1.0

-0.5

-1.0

This finds a numerical approximation to the value of x at which the intersection occurs. The 0
tells FindRoot what value of x to try first.
In[2]:= FindRoot@Cos@xD == x, 8x, 0<D
Out[2]= 8x 0.739085<

In trying to find a solution to your equation, FindRoot starts at the point you specify, and then
progressively tries to get closer and closer to a solution. Even if your equations have several
solutions, FindRoot always returns the first solution it finds. Which solution this is will depend
on what starting point you chose. If you start sufficiently close to a particular solution,
FindRoot will usually return that solution.

The function sinHxL has an infinite number of roots of the form x = n p. If you start sufficiently
close to a particular root, FindRoot will give you that root.
In[3]:= FindRoot@Sin@xD, 8x, 3<D
Out[3]= 8x 3.14159<

If you start with x = 6, you get a numerical approximation to the root x = 2 p.

In[4]:= FindRoot@Sin@xD, 8x, 6<D
Out[4]= 8x 6.28319<

If you want FindRoot to search for complex solutions, then you have to give a complex
starting value.
In[5]:= FindRoot@Sin@xD == 2, 8x, I<D
Out[5]= 8x 1.5708 + 1.31696 <

This finds a zero of the Riemann zeta function.

In[6]:= FindRoot@Zeta@1 2 + I tD, 8t, 12<D
-15
Out[6]= 9t 14.1347 - 2.54839 10 =

This
This finds
finds aa solution
solution to
to aa set
set of
of simultaneous
simultaneous equations.
equations.
Mathematics and Algorithms 275

This finds a solution to a set of simultaneous equations.

In[7]:= FindRoot@8Sin@xD == Cos@yD, x + y == 1<, 88x, 1<, 8y, 1<<D
Out[7]= 8x -1.85619, y 2.85619<

The variables used by FindRoot can have values that are lists. This allows you to find roots of
functions that take vectors as arguments.

This is a way to solve a linear equation for the variable x.

In[8]:= FindRoot@881, 2<, 83, 4<<.x == 85, 6<, 8x, 81, 1<<D
Out[8]= 8x 8-4., 4.5<<

This finds a normalized eigenvector x and eigenvalue a.

In[9]:= FindRoot@8881, 2<, 83, 4<<.x == a x, x.x == 1<, 88x, 81, 1<<, 8a, 1<<D
Out[9]= 8x 80.415974, 0.909377<, a 5.37228<

Introduction to Numerical Differential Equations

NDSolve@eqns,y,8x,xmin ,xmax <D

solve numerically for the function y, with the independent
variable x in the range xmin to xmax

NDSolve@eqns,8y1 ,y2 ,<,8x,xmin ,xmax <D

solve a system of equations for the yi

Numerical solution of differential equations.

This generates a numerical solution to the equation y HxL = yHxL with 0 < x < 2. The result is given
in terms of an InterpolatingFunction .
In[1]:= NDSolve@8y '@xD == y@xD, y@0D == 1<, y, 8x, 0, 2<D
Out[1]= 88y InterpolatingFunction@880., 2.<<, <>D<<

Here is the value of y H1.5L.

In[2]:= [email protected] . %
Out[2]= 84.48169<

With an algebraic equation such as x2 + 3 x + 1 = 0, each solution for x is simply a single number.
For a differential equation, however, the solution is a function, rather than a single number. For
example, in the equation y HxL = yHxL, you want to get an approximation to the function yHxL as the
independent variable x varies over some range.
276 Mathematics and Algorithms

Mathematica represents numerical approximations to functions as InterpolatingFunction

objects. These objects are functions which, when applied to a particular x, return the approxi-
mate value of yHxL at that point. The InterpolatingFunction effectively stores a table of values
for yHxi L, then interpolates this table to find an approximation to yHxL at the particular x you
request.

[email protected] use the list of rules for the function y to get values for y@xD
InterpolatingFunction@dataD@xD evaluate an interpolated function at the point x
Plot@Evaluate@[email protected],8x,xmin ,xmax <D
plot the solution to a differential equation

Using results from NDSolve .

This solves a system of two coupled differential equations.

In[3]:= NDSolve@8y '@xD == z@xD, z '@xD == - y@xD, y@0D == 0, z@0D == 1<, 8y, z<, 8x, 0, Pi<D
Out[3]= 88y InterpolatingFunction@880., 3.14159<<, <>D, z InterpolatingFunction@880., 3.14159<<, <>D<<

Here is the value of z@2D found from the solution.

In[4]:= z@2D . %
Out[4]= 8-0.416147<

Here is a plot of the solution for z@xD found on line 3. Plot is discussed in "Basic Plotting".
In[5]:= Plot@Evaluate@z@xD . %3D, 8x, 0, Pi<D
1.0

0.5

Out[5]=
0.5 1.0 1.5 2.0 2.5 3.0
-0.5

-1.0

NDSolve@eqn,u,8x,xmin ,xmax <,8t,tmin ,tmax <,D

solve a partial differential equation

Numerical solution of partial differential equations.

Mathematics and Algorithms 277

Numerical Solution of Differential Equations

The function NDSolve discussed in "Numerical Differential Equations" allows you to find numeri-
cal solutions to differential equations. NDSolve handles both single differential equations, and
sets of simultaneous differential equations. It can handle a wide range of ordinary differential
equations as well as some partial differential equations. In a system of ordinary differential
equations there can be any number of unknown functions yi , but all of these functions must
depend on a single "independent variable" x, which is the same for each function. Partial
differential equations involve two or more independent variables. NDSolve can also handle
differential-algebraic equations that mix differential equations with algebraic ones.

NDSolve@8eqn1 ,eqn2 ,<,y,8x,xmin ,xmax <D

find a numerical solution for the function y with x in the
range xmin to xmax

NDSolve@8eqn1 ,eqn2 ,<,8y1 ,y2 ,<,8x,xmin ,xmax <D

find numerical solutions for several functions yi

Finding numerical solutions to ordinary differential equations.

NDSolve represents solutions for the functions yi as InterpolatingFunction objects. The

InterpolatingFunction objects provide approximations to the yi over the range of values xmin
to xmax for the independent variable x.

NDSolve finds solutions iteratively. It starts at a particular value of x, then takes a sequence of
steps, trying eventually to cover the whole range xmin to xmax .

In order to get started, NDSolve has to be given appropriate initial or boundary conditions for
the yi and their derivatives. These conditions specify values for yi @xD, and perhaps derivatives
yi '@xD, at particular points x. In general, at least for ordinary differential equations, the conditions
you give can be at any x: NDSolve will automatically cover the range xmin to xmax .

This finds a solution for y with x in the range 0 to 2, using an initial condition for y@0D.
In[1]:= NDSolve@8y '@xD == y@xD, y@0D == 1<, y, 8x, 0, 2<D
Out[1]= 88y InterpolatingFunction@880., 2.<<, <>D<<
278 Mathematics and Algorithms

This still finds a solution with x in the range 0 to 2, but now the initial condition is for y@3D.
In[2]:= NDSolve@8y '@xD == y@xD, y@3D == 1<, y, 8x, 0, 2<D
Out[2]= 88y InterpolatingFunction@880., 2.<<, <>D<<

Here is a simple boundary value problem.

In[3]:= NDSolve@8y ''@xD + x y@xD == 0, y@0D == 1, y@1D == - 1<, y, 8x, 0, 1<D
Out[3]= 88y InterpolatingFunction@880., 1.<<, <>D<<

When you use NDSolve, the initial or boundary conditions you give must be sufficient to deter-
mine the solutions for the yi completely. When you use DSolve to find symbolic solutions to
differential equations, you can get away with specifying fewer initial conditions. The reason is
that DSolve automatically inserts arbitrary constants C@iD to represent degrees of freedom
associated with initial conditions that you have not specified explicitly. Since NDSolve must give
a numerical solution, it cannot represent these kinds of additional degrees of freedom. As a
result, you must explicitly give all the initial or boundary conditions that are needed to deter-
mine the solution.

In a typical case, if you have differential equations with up to nth derivatives, then you need to
give initial conditions for up to Hn - 1Lth derivatives, or give boundary conditions at n points.

With a third-order equation, you need to give initial conditions for up to second derivatives.
In[4]:= NDSolve@8y '''@xD + 8 y ''@xD + 17 y '@xD + 10 y@xD == 0,
y@0D == 6, y '@0D == - 20, y ''@0D == 84<, y, 8x, 0, 1<D
Out[4]= 88y InterpolatingFunction@880., 1.<<, <>D<<

This plots the solution obtained.

In[5]:= Plot@Evaluate@y@xD . %D, 8x, 0, 1<D
6

Out[5]=
3

0.2 0.4 0.6 0.8 1.0

Mathematics and Algorithms 279

With a third-order equation, you can also give boundary conditions at three points.
In[6]:= NDSolve@8y '''@xD + Sin@xD == 0, y@0D == 4, y@1D == 7, y@2D == 0<, y, 8x, 0, 2<D
Out[6]= 88y InterpolatingFunction@880., 2.<<, <>D<<

Mathematica allows you to use any appropriate linear combination of function values and
derivatives as boundary conditions.
In[7]:= NDSolve@8y ''@xD + y@xD == 12 x, 2 y@0D - y '@0D == - 1, 2 y@1D + y '@1D == 9<, y, 8x, 0, 1<D
Out[7]= 88y InterpolatingFunction@880., 1.<<, <>D<<

In most cases, all the initial conditions you give must involve the same value of x, say x0 . As a
result, you can avoid giving both xmin and xmax explicitly. If you specify your range of x as 8x, x1 <,
then Mathematica will automatically generate a solution over the range x0 to x1 .

This generates a solution over the range 0 to 2.

In[8]:= NDSolve@8y '@xD == y@xD, y@0D == 1<, y, 8x, 2<D
Out[8]= 88y InterpolatingFunction@880., 2.<<, <>D<<

You can give initial conditions as equations of any kind. In some cases, these equations may
have multiple solutions. In such cases, NDSolve will correspondingly generate multiple solu-
tions.

The initial conditions in this case lead to multiple solutions.

In[9]:= NDSolve@8y '@xD ^ 2 - y@xD ^ 2 == 0, y@0D ^ 2 == 4<, y@xD, 8x, 1<D
Out[9]= 88y@xD InterpolatingFunction@880., 1.<<, <>D@xD<,
8y@xD InterpolatingFunction@880., 1.<<, <>D@xD<,
8y@xD InterpolatingFunction@880., 1.<<, <>D@xD<,
8y@xD InterpolatingFunction@880., 1.<<, <>D@xD<<

Here is a plot of all the solutions.

In[10]:= Plot@Evaluate@y@xD . %D, 8x, 0, 1<D

2
Out[10]=
0.2 0.4 0.6 0.8 1.0
-2
-2
-4
-4

You can use NDSolve to solve systems of coupled differential equations.

280 Mathematics and Algorithms

This finds a numerical solution to a pair of coupled equations.

In[11]:= sol = NDSolve@
8x '@tD == - y@tD - x@tD ^ 2, y '@tD == 2 x@tD - y@tD, x@0D == y@0D == 1<, 8x, y<, 8t, 10<D
Out[11]= 88x InterpolatingFunction@880., 10.<<, <>D, y InterpolatingFunction@880., 10.<<, <>D<<

This plots the solution for y from these equations.

In[12]:= Plot@Evaluate@y@tD . solD, 8t, 0, 10<, PlotRange AllD
1.0

0.5
Out[12]=

2 4 6 8 10

-0.5

This generates a parametric plot using both x and y.

In[13]:= ParametricPlot@Evaluate@8x@tD, y@tD< . solD, 8t, 0, 10<, PlotRange -> AllD

1.0

0.5
Out[13]=

-0.4 -0.2 0.2 0.4 0.6 0.8 1.0

-0.5

Unknown functions in differential equations do not necessarily have to be represented by single

symbols. If you have a large number of unknown functions, you will often find it more conve-
nient, for example, to give the functions names like y@iD.

This constructs a set of five coupled differential equations and initial conditions.
In[14]:= eqns = Join@Table@y@iD '@xD == y@i - 1D@xD - y@iD@xD, 8i, 2, 4<D,
8y@1D '@xD == - y@1D@xD, y@5D '@xD == y@4D@xD, y@1D@0D == 1<,
Table@y@iD@0D == 0, 8i, 2, 5<DD

Out[14]= 8y@2D @xD y@1D@xD - y@2D@xD, y@3D @xD y@2D@xD - y@3D@xD,
y@4D @xD y@3D@xD - y@4D@xD, y@1D @xD -y@1D@xD, y@5D @xD y@4D@xD,
y@1D@0D 1, y@2D@0D 0, y@3D@0D 0, y@4D@0D 0, y@5D@0D 0<

This solves the equations.

In[15]:= NDSolve@eqns, Table@y@iD, 8i, 5<D, 8x, 10<D
Out[15]= 88y@1D InterpolatingFunction@880., 10.<<, <>D,
y@2D InterpolatingFunction@880., 10.<<, <>D, y@3D InterpolatingFunction@880., 10.<<, <>D,
y@4D InterpolatingFunction@880., 10.<<, <>D, y@5D InterpolatingFunction@880., 10.<<, <>D<<

Here is a plot of the solutions.

Mathematics and Algorithms 281

Here is a plot of the solutions.

In[16]:= Plot@Evaluate@Table@y@iD@xD, 8i, 5<D . %D, 8x, 0, 10<D
1.0

0.8

0.6
Out[16]=
0.4

0.2

2 4 6 8 10

NDSolve can handle functions whose values are lists or arrays. If you give initial conditions like
y@0D == 8v1 , v2 , , vn <, then NDSolve will assume that y is a function whose values are lists of
length n.

This solves a system of four coupled differential equations.

In[17]:= NDSolve@8y ''@xD == - RandomReal@80, 1<, 84, 4<D.y@xD,
y@0D == y '@0D == Table@1, 84<D<, y, 8x, 0, 8<D
Out[17]= 88y InterpolatingFunction@880., 8.<<, <>D<<

Here are the solutions.

In[18]:= With@8s = y@xD . First@%D<,
Plot@8s@@1DD, s@@2DD, s@@3DD, s@@4DD<, 8x, 0, 8<, PlotRange -> AllDD

Out[18]= 2 4 6 8
-5

-10

option name default value

MaxSteps Automatic maximum number of steps in x to take
StartingStepSize Automatic starting size of step in x to use
MaxStepSize Automatic maximum size of step in x to use
NormFunction Automatic the norm to use for error estimation

Special options for NDSolve .

NDSolve has many methods for solving equations, but essentially all of them at some level
work by taking a sequence of steps in the independent variable x, and using an adaptive proce-
dure to determine the size of these steps. In general, if the solution appears to be varying
rapidly in a particular region, then NDSolve will reduce the step size or change the method so
as to be able to track the solution better.

This solves a differential equation in which the derivative has a discontinuity.

282 Mathematics and Algorithms

This solves a differential equation in which the derivative has a discontinuity.

In[19]:= NDSolve@8y '@xD == If@x < 0, 1 Hx - 1L, 1 Hx + 1LD, y@- 5D == 5<, y, 8x, - 5, 5<D
Out[19]= 88y InterpolatingFunction@88-5., 5.<<, <>D<<

NDSolve reduced the step size around x = 0 so as to reproduce the kink accurately.
In[20]:= Plot@Evaluate@y@xD . %D, 8x, - 5, 5<D
5.0

4.5

Out[20]=
4.0

-4 -2 2 4

Through its adaptive procedure, NDSolve is able to solve "stiff" differential equations in which
there are several components which vary with x at very different rates.

In these equations, y varies much more rapidly than z.

In[21]:= sol =
NDSolve@8y '@xD == - 40 y@xD, z '@xD == - z@xD 10, y@0D == z@0D == 1<, 8y, z<, 8x, 0, 1<D
Out[21]= 88y InterpolatingFunction@880., 1.<<, <>D, z InterpolatingFunction@880., 1.<<, <>D<<

NDSolve nevertheless tracks both components successfully.

In[22]:= Plot@Evaluate@8y@xD, z@xD< . solD, 8x, 0, 1<, PlotRange -> AllD
1.0
0.8
0.6
Out[22]=
0.4
0.2

0.2 0.4 0.6 0.8 1.0

NDSolve follows the general procedure of reducing step size until it tracks solutions accurately.
There is a problem, however, when the true solution has a singularity. In this case, NDSolve
might go on reducing the step size forever, and never terminate. To avoid this problem, the
option MaxSteps specifies the maximum number of steps that NDSolve will ever take in attempt-
ing to find a solution. For ordinary differential equations the default setting is
MaxSteps -> 10 000.
Mathematics and Algorithms 283

NDSolve stops after taking 10000 steps.

In[23]:= NDSolve@8y '@xD == - 1 x ^ 2, y@- 1D == - 1<, y@xD, 8x, - 1, 0<D

NDSolve::mxst : Maximum number of 10000 steps reached at the point x == -1.00413 10-172 .
-172
Out[23]= 99y@xD InterpolatingFunctionA99-1., -1.00413 10 ==, <>E@xD==

There is in fact a singularity in the solution at x = 0.

In[24]:= Plot@Evaluate@y@xD . %D, 8x, - 1, 0<D

Out[24]=

The default setting for MaxSteps should be sufficient for most equations with smooth solutions.
When solutions have a complicated structure, however, you may occasionally have to choose
larger settings for MaxSteps. With the setting MaxSteps -> Infinity there is no upper limit on
the number of steps used.

To take the solution to the Lorenz equations this far, you need to remove the default bound on
MaxSteps .
In[25]:= NDSolve@8x '@tD == - 3 Hx@tD - y@tDL, y '@tD == - x@tD z@tD + 26.5 x@tD - y@tD,
z '@tD == x@tD y@tD - z@tD, x@0D == z@0D == 0, y@0D == 1<,
8x, y, z<, 8t, 0, 200<, MaxSteps -> InfinityD
Out[25]= 88x InterpolatingFunction@880., 200.<<, <>D,
y InterpolatingFunction@880., 200.<<, <>D, z InterpolatingFunction@880., 200.<<, <>D<<

Here is a parametric plot of the solution in three dimensions.

In[26]:= ParametricPlot3D@Evaluate@8x@tD, y@tD, z@tD< . %D, 8t, 0, 200<,
PlotPoints -> 10 000, ColorFunction HColorData@"Rainbow"D@4D &LD

Out[26]=

When NDSolve solves a particular set of differential equations, it always tries to choose a step
size appropriate for those equations. In some cases, the very first step that NDSolve makes
may be too large, and it may miss an important feature in the solution. To avoid this problem,
284 Mathematics and Algorithms

The equations you give to NDSolve do not necessarily all have to involve derivatives; they can
also just be algebraic. You can use NDSolve to solve many such differential-algebraic equations.

This solves a system of differential-algebraic equations.

In[27]:= NDSolve@8x '@tD == y@tD ^ 2 + x@tD y@tD,
2 x@tD ^ 2 + y@tD ^ 2 == 1, x@0D == 0, y@0D == 1<, 8x, y<, 8t, 0, 5<D
Out[27]= 88x InterpolatingFunction@880., 5.<<, <>D, y InterpolatingFunction@880., 5.<<, <>D<<

Here is the solution.

In[28]:= Plot@Evaluate@8x@tD, y@tD< . %D, 8t, 0, 5<D
1.0

0.5
Out[28]=

1 2 3 4 5

-0.5

NDSolve@8eqn1 ,eqn2 ,<,u,8t,tmin ,tmax <,8x,xmin ,xmax <,D

solve a system of partial differential equations for u
NDSolve@8eqn1 ,eqn2 ,<,8u1 ,u2 ,<,8t,tmin ,tmax <,8x,xmin ,xmax <,D
solve a system of partial differential equations for several
functions ui

Finding numerical solutions to partial differential equations.

This finds a numerical solution to the wave equation. The result is a two-dimensional interpolat-
ing function.
In[29]:= NDSolve@8D@u@t, xD, t, tD == D@u@t, xD, x, xD, u@0, xD == Exp@- x ^ 2D,
Derivative@1, 0D@uD@0, xD == 0, u@t, - 6D == u@t, 6D<, u, 8t, 0, 6<, 8x, - 6, 6<D
Out[29]= 88u InterpolatingFunction@880., 6.<, 8..., -6., 6., ...<<, <>D<<
Mathematics and Algorithms 285

This generates a plot of the result.

In[30]:= Plot3D@Evaluate@u@t, xD . First@%DD, 8t, 0, 6<, 8x, - 6, 6<, PlotPoints -> 50D

Out[30]=

This finds a numerical solution to a nonlinear wave equation.

In[31]:= NDSolve@8D@u@t, xD, t, tD == D@u@t, xD, x, xD + H1 - u@t, xD ^ 2L H1 + 2 u@t, xDL,
u@0, xD == Exp@- x ^ 2D, Derivative@1, 0D@uD@0, xD == 0,
u@t, - 10D == u@t, 10D<, u, 8t, 0, 10<, 8x, - 10, 10<D
Out[31]= 88u InterpolatingFunction@880., 10.<, 8..., -10., 10., ...<<, <>D<<

Here is a 3D plot of the result.

In[32]:= Plot3D@Evaluate@u@t, xD . First@%DD, 8t, 0, 10<, 8x, - 10, 10<D

Out[32]=

This is a higher-resolution density plot of the solution.

In[33]:= DensityPlot@Evaluate@u@10 - t, xD . First@%%DD,
8x, - 10, 10<, 8t, 0, 10<, PlotPoints -> 200, Mesh -> FalseD
10

6
Out[33]=
4

0
10 5 0 5 10
286 Mathematics and Algorithms

Here is a version of the equation in 2+1 dimensions.

In[34]:= eqn = D@u@t, x, yD, t, tD ==
D@u@t, x, yD, x, xD + D@u@t, x, yD, y, yD 2 + H1 - u@t, x, yD ^ 2L H1 + 2 u@t, x, yDL
1
H2,0,0L
Out[34]= u @t, x, yD H1 + 2 u@t, x, yDL I1 - u@t, x, yD2 M + uH0,0,2L @t, x, yD + uH0,2,0L @t, x, yD
2

This solves the equation.

In[35]:= NDSolve@8eqn, u@0, x, yD == Exp@- Hx ^ 2 + y ^ 2LD, u@t, - 5, yD == u@t, 5, yD,
u@t, x, - 5D == u@t, x, 5D, Derivative@1, 0, 0D@uD@0, x, yD == 0<,
u, 8t, 0, 4<, 8x, - 5, 5<, 8y, - 5, 5<D
Out[35]= 88u InterpolatingFunction@880., 4.<, 8..., -5., 5., ...<, 8..., -5., 5., ...<<, <>D<<

This generates a list of plots of the solution.

In[36]:= Table@Plot3D@Evaluate@u@t, x, yD . First@%DD, 8x, - 5, 5<, 8y, - 5, 5<,
PlotRange -> All, PlotPoints -> 100, Mesh -> FalseD, 8t, 1, 4<D

Out[36]=
Mathematics and Algorithms 287

Numerical Optimization

FindMinimum @ f ,8x,x0 <D search for a local minimum of f , starting at x = x0

FindMinimum @ f ,xD search for a local minimum of f

FindMinimum @ f ,88x,x0 <,8y,y0 <,<D
search for a local minimum in several variables
FindMinimum @8 f ,cons<,88x,x0 <,8y,y0 <,<D
search for a local minimum subject to the constraints cons
starting at x = x0 , y = y0 ,

FindMinimum @8 f ,cons<,8x,y,<D search for a local minimum subject to the constraints cons
FindMaximum @ f ,xD , etc. search for a local maximum

Searching for local minima and maxima.

This finds the value of x which minimizes GHxL, starting at x = 2.

In[1]:= FindMinimum@Gamma@xD, 8x, 2<D
Out[1]= 80.885603, 8x 1.46163<<

The last element of the list gives the value at which the minimum is achieved.
In[2]:= Gamma@xD . Last@%D
Out[2]= 0.885603

Like FindRoot, FindMinimum and FindMaximum work by starting from a point, then progres-
sively searching for a minimum or maximum. But since they return a result as soon as they find
anything, they may give only a local minimum or maximum of your function, not a global one.

This curve has two local minima.

In[3]:= Plot@x ^ 4 - 3 x ^ 2 + x, 8x, - 3, 2<D

Out[3]= 10

-3 -2 -1 1 2
288 Mathematics and Algorithms

Starting at x = 1, you get the local minimum on the right.

In[4]:= FindMinimum@x ^ 4 - 3 x ^ 2 + x, 8x, 1<D
Out[4]= 8-1.07023, 8x 1.1309<<

This gives the local minimum on the left, which in this case is also the global minimum.
In[5]:= FindMinimum@x ^ 4 - 3 x ^ 2 + x, 8x, - 1<D
Out[5]= 8-3.51391, 8x -1.30084<<

You can specify variables without initial values.

In[6]:= FindMinimum@x ^ 4 - 3 x ^ 2 + x, xD
Out[6]= 8-1.07023, 8x 1.1309<<

You can specify a constraint.

In[7]:= FindMinimum@8x ^ 4 - 3 x ^ 2 + x, x < 0<, xD
Out[7]= 8-3.51391, 8x -1.30084<<

NMinimize@ f ,xD try to find the global minimum of f

NMinimize@ f ,8x,y,<D try to find the global minimum over several variables
NMaximize@ f ,xD try to find the global maximum of f
NMaximize@ f ,8x,y,<D try to find the global maximum over several variables

Finding global minima and maxima.

This immediately finds the global minimum.

In[8]:= NMinimize@x ^ 4 - 3 x ^ 2 + x, xD
Out[8]= 8-3.51391, 8x -1.30084<<

NMinimize and NMaximize are numerical analogs of Minimize and Maximize. But unlike
Minimize and Maximize they usually cannot guarantee to find absolute global minima and
maxima. Nevertheless, they typically work well when the function f is fairly smooth, and has a
limited number of local minima and maxima.
Mathematics and Algorithms 289

NMinimize@8 f ,cons<,8x,y,<D try to find the global minimum of f subject to constraints

cons
NMaximize@8 f ,cons<,8x,y,<D try to find the global maximum of f subject to constraints
cons

Finding global minima and maxima subject to constraints.

With the constraint x > 0, NMinimize will give the local minimum on the right.
In[9]:= NMinimize@8x ^ 4 - 3 x ^ 2 + x, x > 0<, xD
Out[9]= 8-1.07023, 8x 1.1309<<

This finds the minimum of x + 2 y within the unit circle.

In[10]:= NMinimize@8x + 2 y, x ^ 2 + y ^ 2 <= 1<, 8x, y<D
Out[10]= 8-2.23607, 8x -0.447214, y -0.894427<<

In this case Minimize can give an exact result.

In[11]:= Minimize@8x + 2 y, x ^ 2 + y ^ 2 <= 1<, 8x, y<D
4 2
Out[11]= :- 5 , :x - 5 , y- >>
5 5

But in this case it cannot.

In[12]:= Minimize@8Cos@x + 2 yD, x ^ 2 + y ^ 2 <= 1<, 8x, y<D

Out[12]= MinimizeA9Cos@x + 2 yD, x2 + y2 1=, 8x, y<E

This gives a numerical approximation, effectively using NMinimize.

In[13]:= N@%D
Out[13]= 8-0.617273, 8x 0.447214, y 0.894427<<

If both the objective function f and the constraints cons are linear in all variables, then minimiza-
tion and maximization correspond to a linear programming problem. Sometimes it is convenient
to state such problems not in terms of explicit equations, but instead in terms of matrices and
vectors.

LinearProgramming@c,m,bD find the vector x which minimizes c.x subject to the con-
straints m.x b and x 0
LinearProgramming@c,m,b,lD use the constraints m.x b and x l

Linear programming in matrix form.

Here is a linear programming problem in equation form.

290 Mathematics and Algorithms

Here is a linear programming problem in equation form.

In[14]:= Minimize@82 x + 3 y, x + 5 y >= 10, x - y >= 2, x >= 1<, 8x, y<D
32 10 4
Out[14]= : , :x , y >>
3 3 3

Here is the corresponding problem in matrix form.

In[15]:= LinearProgramming@82, 3<, 881, 5<, 81, - 1<, 81, 0<<, 810, 2, 1<D
10 4
Out[15]= : , >
3 3

You can specify a mixture of equality and inequality constraints by making the list b be a
sequence of pairs 8bi , si <. If si is 1, then the ith constraint is mi .x bi . If si is 0 then it is mi .x == bi ,
and if si is -1 then it is mi .x bi .

This makes the first inequality use .

In[16]:= LinearProgramming@82, 3<, 881, 5<, 81, - 1<, 81, 0<<, 8810, - 1<, 82, 1<, 81, 1<<D
Out[16]= 82, 0<

In LinearProgramming@c, m, b, lD, you can make l be a list of pairs 88l1 , u1 <, 8l2 , u2 <, < represent-
ing lower and upper bounds on the xi .

In doing large linear programming problems, it is often convenient to give the matrix m as a
SparseArray object.

Controlling the Precision of Results

In doing numerical operations like NDSolve and NMinimize, Mathematica by default uses
machine numbers. But by setting the option WorkingPrecision -> n you can tell it to use
arbitrary-precision numbers with n-digit precision.

This does a machine-precision computation of a numerical integral.

In[1]:= NIntegrate@Sin@Sin@xDD, 8x, 0, 1<D
Out[1]= 0.430606
Mathematics and Algorithms 291

This does the computation with 30-digit arbitrary-precision numbers.

In[2]:= NIntegrate@Sin@Sin@xDD, 8x, 0, 1<, WorkingPrecision -> 30D
Out[2]= 0.430606103120690604912377355248

When you give a setting for WorkingPrecision, this typically defines an upper limit on the
precision of the results from a computation. But within this constraint you can tell Mathematica
how much precision and accuracy you want it to try to get. You should realize that for many
kinds of numerical operations, increasing precision and accuracy goals by only a few digits can
greatly increase the computation time required. Nevertheless, there are many cases where it is
important to ensure that high precision and accuracy are obtained.

WorkingPrecision the number of digits to use for computations

PrecisionGoal the number of digits of precision to try to get

AccuracyGoal the number of digits of accuracy to try to get

Options for controlling precision and accuracy.

This gives a result to 25-digit precision.

In[3]:= NIntegrate@Sin@Sin@xDD, 8x, 0, 1<, WorkingPrecision -> 30, PrecisionGoal -> 25D
Out[3]= 0.430606103120690604912377355248

50-digit precision cannot be achieved with 30-digit working precision.

In[4]:= NIntegrate@Sin@Sin@xDD, 8x, 0, 1<,
WorkingPrecision -> 30, PrecisionGoal -> 50, MaxRecursion 20D
NIntegrate::slwcon :
Numerical integration converging too slowly; suspect one of the following: singularity, value
of the integration is 0, highly oscillatory integrand, or WorkingPrecision too small.

NIntegrate::eincr :
The global error of the strategy GlobalAdaptive has increased more than 400 times. The global error is
expected to decrease monotonically after a number of integrand evaluations. Suspect one
of the following: the difference between the values of PrecisionGoal and WorkingPrecision
is too small; the integrand is highly oscillatory or it is not a HpiecewiseL smooth function;
or the true value of the integral is 0. Increasing the value of the GlobalAdaptive option
MaxErrorIncreases might lead to a convergent numerical integration. NIntegrate obtained
0.43060610312069060491237735524846578643219268469700477957788899453862440935
086147`79.99999999999999 and
5.03891680239785224285840796406833800958006097055414813173023183082827274593
35312`79.99999999999999*^-40 for the integral and error estimates.
Out[4]= 0.430606103120690604912377355248

Given a particular setting for WorkingPrecision, each of the functions for numerical operations
in Mathematica uses certain default settings for PrecisionGoal and AccuracyGoal. Typical is
the case of NDSolve, in which these default settings are equal to half the settings given for
WorkingPrecision.
292 Mathematics and Algorithms

The precision and accuracy goals normally apply both to the final results returned, and to
various norms or error estimates for them. Functions for numerical operations in Mathematica
typically try to refine their results until either the specified precision goal or accuracy goal is
reached. If the setting for either of these goals is Infinity, then only the other goal is consid-
ered.

In doing ordinary numerical evaluation with N@expr, nD, Mathematica automatically adjusts its
internal computations to achieve n-digit precision in the result. But in doing numerical opera-
tions on functions, it is in practice usually necessary to specify WorkingPrecision and
PrecisionGoal more explicitly.

Monitoring and Selecting Algorithms

Functions in Mathematica are carefully set up so that you normally do not have to know how
they work inside. But particularly for numerical functions that use iterative algorithms, it is
sometimes useful to be able to monitor the internal progress of these algorithms.

StepMonitor an expression to evaluate whenever a successful step is

taken
EvaluationMonitor an expression to evaluate whenever functions from the
input are evaluated

Options for monitoring progress of numerical functions.

This prints the value of x every time a step is taken.

In[1]:= FindRoot@Cos@xD == x, 8x, 1<, StepMonitor :> Print@xDD

0.750364
0.739113
0.739085
0.739085
Out[1]= 8x 0.739085<

Note the importance of using option :> expr rather than option -> expr. You need a delayed rule :>
to make expr be evaluated each time it is used, rather than just when the rule is given.
Mathematics and Algorithms 293

Note the importance of using option :> expr rather than option -> expr. You need a delayed rule :>
to make expr be evaluated each time it is used, rather than just when the rule is given.

Reap and Sow provide a convenient way to make a list of the steps taken.
In[2]:= Reap@FindRoot@Cos@xD == x, 8x, 1<, StepMonitor :> Sow@xDDD
Out[2]= 88x 0.739085<, 880.750364, 0.739113, 0.739085, 0.739085<<<

This counts the steps.

In[3]:= Block@8ct = 0<, 8FindRoot@Cos@xD == x, 8x, 1<, StepMonitor :> ct ++D, ct<D
Out[3]= 88x 0.739085<, 4<

To take a successful step toward an answer, iterative numerical algorithms sometimes have to
do several evaluations of the functions they have been given. Sometimes this is because each
step requires, say, estimating a derivative from differences between function values, and some-
times it is because several attempts are needed to achieve a successful step.

This shows the successful steps taken in reaching the answer.

In[4]:= Reap@FindRoot@Cos@xD == x, 8x, 5<, StepMonitor :> Sow@xDDD
Out[4]= 88x 0.739085<, 88-1., -0.0283783, 1.02962, 0.752589, 0.739125, 0.739085, 0.739085<<<

This shows every time the function was evaluated.

In[5]:= Reap@FindRoot@Cos@xD == x, 8x, 5<, EvaluationMonitor :> Sow@xDDD
Out[5]= 88x 0.739085<,
885., -55., -1., 8.71622, -0.0283783, 1.02962, 0.752589, 0.739125, 0.739085, 0.739085<<<

The pattern of evaluations done by algorithms in Mathematica can be quite complicated.

In[6]:= ListPlot@
Reap@NIntegrate@1 Sqrt@xD, 8x, - 1, 0, 1<, EvaluationMonitor :> Sow@xDDD@@2, 1DDD
1.0

0.5

Out[6]=
50 100 150 200 250

-0.5

-1.0
294 Mathematics and Algorithms

Method->Automatic pick methods automatically (default)

Method->"name" specify an explicit method to use
Method->8"name",8"par1 "->val1 ,<<
specify more details of a method

Method options.

There are often several different methods known for doing particular types of numerical compu-
tations. Typically Mathematica supports most generally successful ones that have been dis-
cussed in the literature, as well as many that have not. For any specific problem, it goes to
considerable effort to pick the best method automatically. But if you have sophisticated knowl-
edge of a problem, or are studying numerical methods for their own sake, you may find it
useful to tell Mathematica explicitly what method it should use. Function reference pages list
some of the methods built into Mathematica; others are discussed in "Numerical and Related
Functions" or in advanced documentation.

This solves a differential equation using method m, and returns the number of steps and evalua-
tions needed.
In[7]:= try@m_D := Block@8s = e = 0<, NDSolve@8y ''@xD + Sin@y@xDD == 0, y '@0D == y@0D == 1<, y,
8x, 0, 100<, StepMonitor :> s ++, EvaluationMonitor :> e ++, Method -> mD; 8s, e<D

With the method selected automatically, this is the number of steps and evaluations that are
needed.
In[8]:= try@AutomaticD
Out[8]= 81118, 2329<

This shows what happens with several other possible methods. The Adams method that is
selected automatically is the fastest.
In[9]:= try
8"Adams", "BDF", "ExplicitRungeKutta", "ImplicitRungeKutta", "Extrapolation"<
Out[9]= 881118, 2329<, 82415, 2861<, 8287, 4595<, 8882, 13 092<, 884, 4146<<

This shows what happens with the explicit Runge-Kutta method when the difference order
parameter is changed.
In[10]:= Table@try@8"ExplicitRungeKutta", "DifferenceOrder" -> n<D, 8n, 4, 9<D
Out[10]= 883519, 14 078<, 8614, 4300<, 8849, 6794<, 8472, 4722<, 8288, 3746<, 8287, 4594<<
Mathematics and Algorithms 295

Functions with Sensitive Dependence on Their Input

Functions that are specified by simple algebraic formulas tend to be such that when their input
is changed only slightly, their output also changes only slightly. But functions that are instead
based on executing procedures quite often show almost arbitrarily sensitive dependence on
their input. Typically the reason this happens is that the procedure "excavates" progressively
less and less significant digits in the input.

This shows successive steps in a simple iterative procedure with input 0.1111.
In[1]:= NestList@FractionalPart@2 D &, 0.1111, 10D
Out[1]= 80.1111, 0.2222, 0.4444, 0.8888, 0.7776, 0.5552, 0.1104, 0.2208, 0.4416, 0.8832, 0.7664<

Here is the result with input 0.1112. Progressive divergence from the result with input 0.1111 is
seen.
In[2]:= NestList@FractionalPart@2 D &, 0.1112, 10D
Out[2]= 80.1112, 0.2224, 0.4448, 0.8896, 0.7792, 0.5584, 0.1168, 0.2336, 0.4672, 0.9344, 0.8688<

The action of FractionalPart@2 xD is particularly simple in terms of the binary digits of the
number x: it just drops the first one, and shifts the remaining ones to the left. After several
steps, this means that the results one gets are inevitably sensitive to digits that are far to the
right, and have an extremely small effect on the original value of x.

This shows the shifting process achieved by FractionalPart@2 xD in the first 8 binary digits
of x.
In[3]:= RealDigits@Take@%, 5D, 2, 8, - 1D
Out[3]= 8880, 0, 0, 1, 1, 1, 0, 0<, 0<, 880, 0, 1, 1, 1, 0, 0, 0<, 0<,
880, 1, 1, 1, 0, 0, 0, 1<, 0<, 881, 1, 1, 0, 0, 0, 1, 1<, 0<, 881, 1, 0, 0, 0, 1, 1, 1<, 0<<

If you give input only to a particular precision, you are effectively specifying only a certain
number of digits. And once all these digits have been "excavated" you can no longer get accu-
rate results, since to do so would require knowing more digits of your original input. So long as
you use arbitrary-precision numbers, Mathematica automatically keeps track of this kind of
degradation in precision, indicating a number with no remaining significant digits by 0. 10e , as
discussed in "Arbitrary-Precision Numbers".
296 Mathematics and Algorithms

Successive steps yield numbers of progressively lower precision, and eventually no precision at
all.
In[4]:= NestList@FractionalPart@40 D &, N@1 9, 20D, 20D
Out[4]= 90.11111111111111111111, 0.4444444444444444444, 0.77777777777777778, 0.1111111111111111,
0.44444444444444, 0.777777777778, 0.11111111111, 0.444444444, 0.77777778, 0.111111,
0.4444, 0.778, 0.1, 0. 10-1 , 0. 101 , 0. 103 , 0. 104 , 0. 106 , 0. 107 , 0. 109 , 0. 1011 =

This asks for the precision of each number. Zero precision indicates that there are no correct
significant digits.
In[5]:= Map@Precision, %D
Out[5]= 820., 19., 17.641, 15.1938, 14.1938, 12.8348, 10.3876, 9.38764,
8.02862, 5.58146, 4.58146, 3.22244, 0.77528, 0., 0., 0., 0., 0., 0., 0., 0.<

This shows that the exact result is a periodic sequence.

In[6]:= NestList@FractionalPart@40 D &, 1 9, 10D
1 4 7 1 4 7 1 4 7 1 4
Out[6]= : , , , , , , , , , , >
9 9 9 9 9 9 9 9 9 9 9

It is important to realize that if you use approximate numbers of any kind, then in an example
like the one above you will always eventually run out of precision. But so long as you use arbi-
trary-precision numbers, Mathematica will explicitly show you any decrease in precision that is
occurring. However, if you use machine-precision numbers, then Mathematica will not keep
track of precision, and you cannot tell when your results become meaningless.

If you use machine-precision numbers, Mathematica will no longer keep track of any degrada-
tion in precision.
In[7]:= NestList@FractionalPart@40 D &, N@1 9D, 20D
Out[7]= 80.111111, 0.444444, 0.777778, 0.111111, 0.444444, 0.777778, 0.111111, 0.444445, 0.77781,
0.112405, 0.496185, 0.847383, 0.89534, 0.813599, 0.543945, 0.757813, 0.3125, 0.5, 0., 0., 0.<

By iterating the operation FractionalPart@2 xD you extract successive binary digits in what-
ever number you start with. And if these digits are apparently random~as in a number like p~
then the results will be correspondingly random. But if the digits have a simple pattern~as in
any rational number~then the results you get will be correspondingly simple.

By iterating an operation such as FractionalPart@3 2 xD it turns out however to be possible

to get seemingly random sequences even from very simple input. This is an example of a very
general phenomenon first identified by Stephen Wolfram in the mid-1980s, which has nothing
directly to do with sensitive dependence on input.
Mathematics and Algorithms 297

This generates a seemingly random sequence, even starting from simple input.
In[8]:= NestList@FractionalPart@3 2 D &, 1, 15D
1 3 1 3 9 27 81 243 217 651 1953 1763 5289 15 867 14 833
Out[8]= :1, , , , , , , , , , , , , , , >
2 4 8 16 32 64 128 256 512 1024 2048 4096 8192 16 384 32 768

After the values have been computed, one can safely find numerical approximations to them.
In[9]:= N@%D
Out[9]= 81., 0.5, 0.75, 0.125, 0.1875, 0.28125, 0.421875, 0.632813, 0.949219,
0.423828, 0.635742, 0.953613, 0.43042, 0.64563, 0.968445, 0.452667<

Here are the last 5 results after 1000 iterations, computed using exact numbers.
In[10]:= Take@N@NestList@FractionalPart@3 2 D &, 1, 1000DD, - 5D
Out[10]= 80.0218439, 0.0327659, 0.0491488, 0.0737233, 0.110585<

Using machine-precision numbers gives completely incorrect results.

In[11]:= Take@NestList@FractionalPart@3 2 D &, 1., 1000D, - 5D
Out[11]= 80.670664, 0.0059966, 0.0089949, 0.0134924, 0.0202385<

Many kinds of iterative procedures yield functions that depend sensitively on their input. Such
functions also arise when one looks at solutions to differential equations. In effect, varying the
independent parameter in the differential equation is a continuous analog of going from one
step to the next in an iterative procedure.

This finds a solution to the Duffing equation with initial condition 1.

In[12]:= NDSolve@8x ''@tD + 0.15 x '@tD - x@tD + x@tD ^ 3 == 0.3 Cos@tD,
x@0D == - 1, x '@0D == 1<, x, 8t, 0, 50<D
Out[12]= 88x InterpolatingFunction@880., 50.<<, <>D<<

Here is a plot of the solution.

In[13]:= Plot@Evaluate@x@tD . %D, 8t, 0, 50<D
1.5
1.0
0.5
Out[13]=
10 20 30 40 50
-0.5
-1.0
-1.5
298 Mathematics and Algorithms

Here is the same equation with initial condition 1.001.

In[14]:= NDSolve@8x ''@tD + 0.15 x '@tD - x@tD + x@tD ^ 3 == 0.3 Cos@tD,
x@0D == - 1, x '@0D == 1.001<, x, 8t, 0, 50<D
Out[14]= 88x InterpolatingFunction@880., 50.<<, <>D<<

The solution progressively diverges from the one shown above.

In[15]:= Plot@Evaluate@x@tD . %D, 8t, 0, 50<D
1.5
1.0
0.5
Out[15]=
10 20 30 40 50
-0.5
-1.0
Mathematics and Algorithms 299

Numerical Operations on Data

Basic Statistics

Mean@listD mean (average)

Median@listD median (central value)
Max@listD maximum value
Variance@listD variance
StandardDeviation@listD standard deviation

Quantile@list,qD qth quantile

Total@listD total

Basic descriptive statistics operations.

Given a list with n elements xi , the mean Mean@listD is defined to be mHxL = x = xi n.

The variance Variance@listD is defined to be varHxL = s2 HxL = Hxi - mHxLL2 Hn - 1L, for real data. (For

complex data var HxL = s2 HxL = Hxi - mHxLL Hxi - mHxLL Hn - 1L.)

The standard deviation StandardDeviation@listD is defined to be sHxL = varHxL .

If the elements in list are thought of as being selected at random according to some probability
distribution, then the mean gives an estimate of where the center of the distribution is located,
while the standard deviation gives an estimate of how wide the dispersion in the distribution is.

The median Median@listD effectively gives the value at the halfway point in the sorted version of
list. It is often considered a more robust measure of the center of a distribution than the mean,
since it depends less on outlying values.

The qth quantile Quantile@list, qD effectively gives the value that is q of the way through the
sorted version of list.

For a list of length n, Mathematica defines Quantile@list, qD to be s@@Ceiling@n qDDD, where s

is Sort@list, LessD.

There are, however, about ten other definitions of quantile in use, all potentially giving slightly
different results. Mathematica covers the common cases by introducing four quantile parame-
ters in the form Quantile@list, q, 88a, b<, 8c, d<<D. The parameters a and b in effect define
where in the list should be considered a fraction q of the way through. If this corresponds to an
integer position, then the element at that position is taken to be the qth quantile. If it is not an
300 Mathematics and Algorithms

The position in a sorted list s for the qth quantile is taken to be k = a + Hn + bL q. If k is an integer,
then the quantile is sk . Otherwise, it is sdkt + Hs`kp - sdkt L Hc + d Hk - dktLL, with the indices taken to be 1
or n if they are out of range.

880,0<,81,0<< inverse empirical CDF (default)

880,0<,80,1<< linear interpolation (California method)
8812,0<,80,0<< element numbered closest to qn
8812,0<,80,1<< linear interpolation (hydrologist method)
880,1<,80,1<< mean-based estimate (Weibull method)
881,-1<,80,1<< mode-based estimate
8813,13<,80,1<< median-based estimate
8838,14<,80,1<< normal distribution estimate

Common choices for quantile parameters.

Whenever d = 0, the value of the qth quantile is always equal to some actual element in list, so
that the result changes discontinuously as q varies. For d = 1, the qth quantile interpolates lin-
early between successive elements in list. Median is defined to use such an interpolation.

Note that Quantile@list, qD yields quartiles when q = m 4 and percentiles when q = m 100.

Mean@8x1 ,x2 ,<D the mean of the xi

Mean@88x1 ,y1 ,<,8x2 ,y2 ,<,<D a list of the means of the xi , yi ,

Handling multidimensional data.

Sometimes each item in your data may involve a list of values. The basic statistics functions in
Mathematica automatically apply to all corresponding elements in these lists.
Mathematics and Algorithms 301

This separately finds the mean of each "column" of data.

In[1]:= Mean@88x1, y1<, 8x2, y2<, 8x3, y3<<D
1 1
Out[1]= : Hx1 + x2 + x3L, Hy1 + y2 + y3L>
3 3

Note that you can extract the elements in the ith "column" of a multidimensional list using
list@@All, iDD.

Descriptive Statistics
Descriptive statistics refers to properties of distributions, such as location, dispersion, and
shape. The functions described here compute descriptive statistics of lists of data. You can
calculate some of the standard descriptive statistics for various known distributions by using the
functions described in "Continuous Distributions" and "Discrete Distributions".

1
The statistics are calculated assuming that each value of data xi has probability equal to n
,

where n is the number of elements in the data.

1
Mean@dataD average value i xi
n

Median@dataD median (central value)

Commonest@dataD list of the elements with highest frequency
1

GeometricMean@dataD geometric mean Ii xi M n

1
HarmonicMean@dataD harmonic mean n i
xi

1
RootMeanSquare@dataD root mean square i xi2
n

TrimmedMean @data, f D mean of remaining entries, when a fraction f is removed

from each end of the sorted list of data
TrimmedMean @data,8 f1 , f2 <D mean of remaining entries, when fractions f1 and f2 are
dropped from each end of the sorted data

Quantile@data,qD qth quantile

1 th 1 th 3 th
Quartiles@dataD list of the , , quantiles of the elements in list
4 2 4

Location statistics.

Location statistics describe where the data are located. The most common functions include
measures of central tendency like the mean, median, and mode. Quantile@data, qD gives the
location before which H100 qL percent of the data lie. In other words, Quantile gives a value z
302 Mathematics and Algorithms

Here is a dataset.
In[1]:= data = 86.5, 3.8, 6.6, 5.7, 6.0, 6.4, 5.3<
Out[1]= 86.5, 3.8, 6.6, 5.7, 6., 6.4, 5.3<

This finds the mean and median of the data.

In[2]:= 8Mean@dataD, Median@dataD<
Out[2]= 85.75714, 6.<

This is the mean when the smallest entry in the list is excluded. TrimmedMean allows you to
describe the data with removed outliers.
1
In[3]:= TrimmedMeanBdata, : , 0>F
7
Out[3]= 6.08333

1 2
Variance@dataD unbiased estimate of variance, i Hxi - xL
n-1

StandardDeviation@dataD unbiased estimate of standard deviation

1
MeanDeviation@dataD mean absolute deviation, i xi - x
n

MedianDeviation@dataD median absolute deviation, median of xi - median values

InterquartileRange@dataD difference between the first and third quartiles
QuartileDeviation@dataD half the interquartile range

Dispersion statistics.

Dispersion statistics summarize the scatter or spread of the data. Most of these functions
describe deviation from a particular location. For instance, variance is a measure of deviation
from the mean, and standard deviation is just the square root of the variance.

This gives an unbiased estimate for the variance of the data with n - 1 as the divisor.
In[4]:= Variance@dataD
Out[4]= 0.962857

This compares three types of deviation.

Mathematics and Algorithms 303

This compares three types of deviation.

In[5]:= 8StandardDeviation@dataD, MeanDeviation@dataD, MedianDeviation@dataD<
Out[5]= 80.981253, 0.706122, 0.5<

Covariance@v1 ,v2 D covariance coefficient between lists v1 and v2

Covariance@mD covariance matrix for the matrix m
Covariance@m1 ,m2 D covariance matrix for the matrices m1 and m2
Correlation @v1 ,v2 D correlation coefficient between lists v1 and v2
Correlation @mD correlation matrix for the matrix m
Correlation @m1 ,m2 D correlation matrix for the matrices m1 and m2

Covariance and correlation statistics.

Covariance is the multivariate extension of variance. For two vectors of equal length, the covari-
ance is a number. For a single matrix m, the i, jth element of the covariance matrix is the
covariance between the ith and jth columns of m. For two matrices m1 and m2 , the i, jth element
of the covariance matrix is the covariance between the ith column of m1 and the jth column of m2 .

While covariance measures dispersion, correlation measures association. The correlation

between two vectors is equivalent to the covariance between the vectors divided by the stan-
dard deviations of the vectors. Likewise, the elements of a correlation matrix are equivalent to
the elements of the corresponding covariance matrix scaled by the appropriate column standard
deviations.

This gives the covariance between data and a random vector.

In[6]:= Covariance@data, RandomReal@1, Length@dataDDD
Out[6]= 0.0258505

Here is a random matrix.

In[7]:= m = RandomReal@10, 820, 2<D
Out[7]= 888.01573, 5.3642<, 86.70564, 0.352495<, 82.17328, 7.48353<, 81.33259, 3.27026<,
89.54907, 8.35172<, 81.56138, 9.4684<, 85.76737, 4.42373<, 88.65789, 6.66041<,
86.65159, 7.40813<, 83.38061, 6.22431<, 80.269599, 9.76406<, 85.23322, 4.58995<,
83.3881, 1.66902<, 85.66131, 6.06514<, 87.50919, 8.17705<, 85.92976, 0.803385<,
89.96, 1.18177<, 82.14364, 5.8279<, 88.13317, 8.79128<, 83.51722, 3.08246<<

This is the correlation matrix for the matrix m.

In[8]:= Correlation@mD
Out[8]= 881., -0.132314<, 8-0.132314, 1.<<

This is the covariance matrix.

304 Mathematics and Algorithms

This is the covariance matrix.

In[9]:= Covariance@mD
Out[9]= 888.48155, -1.13411<, 8-1.13411, 8.66215<<

Scaling the covariance matrix terms by the appropriate standard deviations gives the correla-
tion matrix.
In[10]:= With@8sd = StandardDeviation@mD<,
Transpose@Transpose@% sdD sdDD
Out[10]= 881., -0.132314<, 8-0.132314, 1.<<

1
CentralMoment@data,rD rth central moment n
r
i Hxi - xL

Skewness@dataD coefficient of skewness

Kurtosis@dataD kurtosis coefficient
QuartileSkewness@dataD quartile skewness coefficient

Shape statistics.

You can get some information about the shape of a distribution using shape statistics. Skew-
ness describes the amount of asymmetry. Kurtosis measures the concentration of data around
the peak and in the tails versus the concentration in the flanks.

Skewness is calculated by dividing the third central moment by the cube of the population
standard deviation. Kurtosis is calculated by dividing the fourth central moment by the square
of the population variance of the data, equivalent to CentralMoment@data, 2D. (The population
variance is the second central moment, and the population standard deviation is its square
root.)

QuartileSkewness is calculated from the quartiles of data. It is equivalent to

Hq1 - 2 q2 + q3 L Hq3 - q1 L, where q1 , q2 , and q3 are the first, second, and third quartiles respectively.

Here is the second central moment of the data.

In[11]:= CentralMoment@data, 2D
Out[11]= 0.825306

A negative value for skewness indicates that the distribution underlying the data has a long left-
sided tail.
In[12]:= Skewness@dataD
Out[12]= -1.20108
Mathematics and Algorithms 305

ExpectedValue@ f ,listD expected value of the pure function f with respect to the
values in list
ExpectedValue@ f @xD,list,xD expected value of the function f of x with respect to the
values of list

Expected values.

1
The expected value of a function f is n
i=1 f Hxi L for the list of values x1 , x2 , , xn . Many descrip-
n

tive statistics are expected values. For instance, the mean is the expected value of x, and the
rth central moment is the expected value of Hx - xLr where x is the mean of the xi .

Here is the expected value of the Log of the data.

In[13]:= ExpectedValue@Log, dataD
Out[13]= 1.73573

Discrete Distributions
The functions described here are among the most commonly used discrete statistical distribu-
tions. You can compute their densities, means, variances, and other related properties. The
distributions themselves are represented in the symbolic form name@param1 , param2 , D. Func-
tions such as Mean, which give properties of statistical distributions, take the symbolic represen-
tation of the distribution as an argument. "Continuous Distributions" describes many continuous
statistical distributions.

BernoulliDistribution@pD Bernoulli distribution with mean p

BetaBinomialDistribution@ binomial distribution where the success probability is a
a,b,nD BetaDistribution@a, bD random variable
BetaNegativeBinomialDistribution@a,b,nD
negative binomial distribution where the success probabil -
ity is a BetaDistribution@a, bD random variable
BinomialDistribution@n,pD binomial distribution for the number of successes that
occur in n trials, where the probability of success in a trial
is p
DiscreteUniformDistribution@8imin ,imax <D
306 Mathematics and Algorithms

discrete uniform distribution over the integers from imin to

imax
GeometricDistribution@pD geometric distribution for the number of trials before the
first success, where the probability of success in a trial is p
HypergeometricDistribution@n,nsucc ,ntot D
hypergeometric distribution for the number of successes
out of a sample of size n, from a population of size ntot
containing nsucc successes

LogSeriesDistribution@qD logarithmic series distribution with parameter q

NegativeBinomialDistribution@n,pD
negative binomial distribution with parameters n and p
PoissonDistribution@mD Poisson distribution with mean m
ZipfDistribution@rD Zipf distribution with parameter r

Discrete statistical distributions.

Most of the common discrete statistical distributions can be understood by considering a

sequence of trials, each with two possible outcomes, for example, success and failure.

The Bernoulli distribution BernoulliDistribution@pD is the probability distribution for a single

trial in which success, corresponding to value 1, occurs with probability p, and failure, corre-
sponding to value 0, occurs with probability 1 - p.

The binomial distribution BinomialDistribution@n, pD is the distribution of the number of

successes that occur in n independent trials, where the probability of success in each trial is p.

The negative binomial distribution NegativeBinomialDistribution@n, pD for positive integer n

is the distribution of the number of failures that occur in a sequence of trials before n successes
have occurred, where the probability of success in each trial is p. The distribution is defined for
any positive n, though the interpretation of n as the number of successes and p as the success
probability no longer holds if n is not an integer.

The beta binomial distribution BetaBinomialDistribution@a, b, nD is a mixture of binomial

and beta distributions. A BetaBinomialDistribution@a, b, nD random variable follows a
BinomialDistribution@n, pD distribution, where the success probability p is itself a random
variable following the beta distribution BetaDistribution@a, bD. The beta negative binomial
distribution BetaNegativeBinomialDistribution@a, b, nD is a similar mixture of the beta and
negative binomial distributions.

The geometric distribution GeometricDistribution@pD is the distribution of the total number

of trials before the first success occurs, where the probability of success in each trial is p.

The hypergeometric distribution HypergeometricDistribution@n, nsucc , ntot D is used in place of

Mathematics and Algorithms 307

The hypergeometric distribution HypergeometricDistribution@n, nsucc , ntot D is used in place of

the binomial distribution for experiments in which the n trials correspond to sampling without
replacement from a population of size ntot with nsucc potential successes.

The discrete uniform distribution DiscreteUniformDistribution@8imin , imax <D represents an

experiment with multiple equally probable outcomes represented by integers imin through imax .

The Poisson distribution PoissonDistribution@mD describes the number of events that occur in
a given time period where m is the average number of events per period.

The terms in the series expansion of log H1 - qL about q = 0 are proportional to the probabilities of
a discrete random variable following the logarithmic series distribution
LogSeriesDistribution@qD. The distribution of the number of items of a product purchased by
a buyer in a specified interval is sometimes modeled by this distribution.

The Zipf distribution ZipfDistribution@rD, sometimes referred to as the zeta distribution, was
first used in linguistics and its use has been extended to model rare events.

PDF@dist,xD probability mass function at x

CDF@dist,xD cumulative distribution function at x
InverseCDF@dist,qD the largest integer x such that CDF@dist, xD is at most q

Quantile@dist,qD qth quantile

Mean@distD mean
Variance@distD variance
StandardDeviation@distD standard deviation
Skewness@distD coefficient of skewness
Kurtosis@distD coefficient of kurtosis
CharacteristicFunction@dist,tD characteristic function fHtL
ExpectedValue@ f ,distD expected value of the pure function f in dist
ExpectedValue@ f @xD,dist,xD expected value of f @xD for x in dist
Median@distD median
1 th 1 th 3 th
Quartiles@distD list of the , , quantiles for dist
4 2 4

InterquartileRange@distD difference between the first and third quartiles

QuartileDeviation@distD half the interquartile range
308 Mathematics and Algorithms

QuartileSkewness@distD quartile-based skewness measure

RandomInteger@distD pseudorandom number with specified distribution
RandomInteger@dist,dimsD pseudorandom array with dimensionality dims, and ele-
ments from the specified distribution

Functions of statistical distributions.

Distributions are represented in symbolic form. PDF@dist, xD evaluates the mass function at x if x
is a numerical value, and otherwise leaves the function in symbolic form whenever possible.
Similarly, CDF@dist, xD gives the cumulative distribution and Mean@distD gives the mean of the
specified distribution. For a more complete description of the various functions of a statistical
distribution, see the description of their continuous analogues in "Continuous Distributions".

Here is a symbolic representation of the binomial distribution for 34 trials, each having probabil-
ity 0.3 of success.
In[1]:= bdist = BinomialDistribution@34, 0.3D
Out[1]= BinomialDistribution@34, 0.3D

This is the mean of the distribution.

In[2]:= Mean@bdistD
Out[2]= 10.2

You can get the expression for the mean by using symbolic variables as arguments.
In[3]:= Mean@BinomialDistribution@n, pDD
Out[3]= n p

Here is the 50% quantile, which is equal to the median.

In[4]:= Quantile@bdist, 0.5D
Out[4]= 10

This gives the expected value of x3 with respect to the binomial distribution.
In[5]:= ExpectedValue[x^3, bdist, x]
Out[5]= 1282.55

The elements of this matrix are pseudorandom numbers from the binomial distribution.
In[6]:= RandomInteger@bdist, 82, 3<D
Out[6]= 8810, 7, 9<, 812, 10, 11<<

Continuous Distributions
Mathematics and Algorithms 309

Continuous Distributions
The functions described here are among the most commonly used continuous statistical distribu-
tions. You can compute their densities, means, variances, and other related properties. The
distributions themselves are represented in the symbolic form name@param1 , param2 , D. Func-
tions such as Mean, which give properties of statistical distributions, take the symbolic represen-
tation of the distribution as an argument. "Discrete Distributions" describes many discrete
statistical distributions.

NormalDistribution@m,sD normal (Gaussian) distribution with mean m and standard

deviation s
HalfNormalDistribution@qD half-normal distribution with scale inversely proportional to
parameter q
LogNormalDistribution@m,sD lognormal distribution based on a normal distribution with
mean m and standard deviation s
InverseGaussianDistribution@ inverse Gaussian distribution with mean m and scale l
m,lD

Distributions related to the normal distribution.

The lognormal distribution LogNormalDistribution@m, sD is the distribution followed by the

exponential of a normally distributed random variable. This distribution arises when many
independent random variables are combined in a multiplicative fashion. The half-normal distribu-
tion HalfNormalDistribution@qD is proportional to the distribution
NormalDistribution@0, 1 Hq Sqrt@2 pDLD limited to the domain @0, L.

The inverse Gaussian distribution InverseGaussianDistribution@m, lD, sometimes called the

Wald distribution, is the distribution of first passage times in Brownian motion with positive
drift.

ChiSquareDistribution@nD c2 distribution with n degrees of freedom

InverseChiSquareDistribution@ inverse c2 distribution with n degrees of freedom

nD
FRatioDistribution@n,mD F-ratio distribution with n numerator and m denominator
degrees of freedom
StudentTDistribution@nD Student t distribution with n degrees of freedom
310 Mathematics and Algorithms

NoncentralChiSquareDistribu@ noncentral c2 distribution with n degrees of freedom and

tion noncentrality parameter l
n,lD
NoncentralStudentTDistribut@ noncentral Student t distribution with n degrees of freedom
ion and noncentrality parameter d
n,dD
NoncentralFRatioDistribution@ noncentral F -ratio distribution with n numerator degrees of
n,m,lD freedom and m denominator degrees of freedom and
numerator noncentrality parameter l

Distributions related to normally distributed samples.

If X1 ,, Xn are independent normal random variables with unit variance and mean zero, then
n
i=1 Xi2 has a c2 distribution with n degrees of freedom. If a normal variable is standardized by

subtracting its mean and dividing by its standard deviation, then the sum of squares of such
quantities follows this distribution. The c2 distribution is most typically used when describing
the variance of normal samples.

If Y follows a c2 distribution with n degrees of freedom, 1 Y follows the inverse c 2 distribution

InverseChiSquareDistribution@nD. A scaled inverse c 2 distribution with n degrees of freedom

and scale x can be given as InverseChiSquareDistribution@n, xD. Inverse c2 distributions are

commonly used as prior distributions for the variance in Bayesian analysis of normally dis-
tributed samples.

A variable that has a Student t distribution can also be written as a function of normal random
variables. Let X and Z be independent random variables, where X is a standard normal distribu-

tion and Z is a c2 variable with n degrees of freedom. In this case, X Z n has a t distribution

with n degrees of freedom. The Student t distribution is symmetric about the vertical axis, and
characterizes the ratio of a normal variable to its standard deviation. Location and scale parame-
ters can be included as m and s in StudentTDistribution@m, s, nD. When n = 1, the t distribu-
tion is the same as the Cauchy distribution.

The F-ratio distribution is the distribution of the ratio of two independent c2 variables divided by
their respective degrees of freedom. It is commonly used when comparing the variances of two
populations in hypothesis testing.

Distributions that are derived from normal distributions with nonzero means are called noncen-
tral distributions.

The sum of the squares of n normally distributed random variables with variance s2 = 1 and
nonzero means follows a noncentral c2 distribution NoncentralChiSquareDistribution@n, lD.
The noncentrality parameter l is the sum of the squares of the means of the random variables
Mathematics and Algorithms 311

The noncentral Student t distribution NoncentralStudentTDistribution@n, dD describes the

ratio X c2n n where c2n is a central c2 random variable with n degrees of freedom, and X is an

independent normally distributed random variable with variance s2 = 1 and mean d.

The noncentral F-ratio distribution NoncentralFRatioDistribution@n, m, lD is the distribution

1 1
of the ratio of n
c2n HlL to m
c2m , where c2n HlL is a noncentral c2 random variable with noncentrality

parameter l and n1 degrees of freedom and c2m is a central c2 random variable with m degrees of

freedom.

TriangularDistribution@8a,b<D symmetric triangular distribution on the interval 8a, b<

TriangularDistribution@8a,b<,cD triangular distribution on the interval 8a, b< with maxi -
mum at c
UniformDistribution@8min,max<D uniform distribution on the interval 8min, max<

Piecewise linear distributions.

The triangular distribution TriangularDistribution@8a, b<, cD is a triangular distribution for

a+b
a<X<b with maximum probability at c and a < c < b. If c is 2
,

TriangularDistribution@8a, b<, cD is the symmetric triangular distribution

TriangularDistribution@8a, b<D.

The uniform distribution UniformDistribution@8min, max<D, commonly referred to as the

rectangular distribution, characterizes a random variable whose value is everywhere equally
likely. An example of a uniformly distributed random variable is the location of a point chosen
randomly on a line from min to max.

BetaDistribution@a, bD continuous beta distribution with shape parameters a and b

CauchyDistribution@a,bD Cauchy distribution with location parameter a and scale
parameter b
312 Mathematics and Algorithms

ChiDistribution@nD c distribution with n degrees of freedom

ExponentialDistribution@lD exponential distribution with scale inversely proportional to
parameter l
ExtremeValueDistribution@a, bD extreme maximum value (Fisher|Tippett) distribution with
location parameter a and scale parameter b
GammaDistribution@a, bD gamma distribution with shape parameter a and scale
parameter b
GumbelDistribution@a, bD Gumbel minimum extreme value distribution with location
parameter a and scale parameter b
InverseGammaDistribution@a, bD inverse gamma distribution with shape parameter a and
scale parameter b
LaplaceDistribution@m,bD Laplace (double exponential) distribution with mean m and
scale parameter b
LevyDistribution@m,sD Lvy distribution with location parameter m and dispersion
parameter s
LogisticDistribution@m, bD logistic distribution with mean m and scale parameter b
MaxwellDistribution@sD Maxwell (Maxwell|Boltzmann) distribution with scale
parameter s
ParetoDistribution@k,aD Pareto distribution with minimum value parameter k and
shape parameter a
RayleighDistribution@sD Rayleigh distribution with scale parameter s
WeibullDistribution@a, bD Weibull distribution with shape parameter a and scale
parameter b

Other continuous statistical distributions.

If X is uniformly distributed on@- p, pD, then the random variable tanHXL follows a Cauchy distribu -
tion CauchyDistribution@a, bD, with a = 0 and b = 1.

When a = n 2 and l = 2, the gamma distribution GammaDistribution@a, lD describes the distribu-

tion of a sum of squares of n-unit normal random variables. This form of the gamma distribu-
tion is called a c2 distribution with n degrees of freedom. When a = 1, the gamma distribution
takes on the form of the exponential distribution ExponentialDistribution@lD, often used in
describing the waiting time between events.

If a random variable X follows the gamma distribution GammaDistribution@a, bD, 1 X follows

the inverse gamma distribution InverseGammaDistribution@a, 1 bD. If a random variable X
follows InverseGammaDistribution@1 2, s 2D, X+m follows a Lvy distribution
LevyDistribution@m, sD.

When X1 and X2 have independent gamma distributions with equal scale parameters, the ran-
X1
dom variable X1 +X2
follows the beta distribution BetaDistribution@a, bD, where a and b are

the shape parameters of the gamma variables.

Mathematics and Algorithms 313

When X1 and X2 have independent gamma distributions with equal scale parameters, the ran-
X1
dom variable X1 +X2
follows the beta distribution BetaDistribution@a, bD, where a and b are

the shape parameters of the gamma variables.

The c distribution ChiDistribution@nD is followed by the square root of a c2 random variable.

p
For n = 1, the c distribution is identical to HalfNormalDistribution@qD with q = 2
. For n = 2,

the c distribution is identical to the Rayleigh distribution RayleighDistribution@sD with s = 1.

For n = 3, the c distribution is identical to the Maxwell|Boltzmann distribution
MaxwellDistribution@sD with s = 1.

The Laplace distribution LaplaceDistribution@m, bD is the distribution of the difference of two

independent random variables with identical exponential distributions. The logistic distribution
LogisticDistribution@m, bD is frequently used in place of the normal distribution when a
distribution with longer tails is desired.

The Pareto distribution ParetoDistribution@k, aD may be used to describe income, with k

representing the minimum income possible.

The Weibull distribution WeibullDistribution@a, bD is commonly used in engineering to

describe the lifetime of an object. The extreme value distribution
ExtremeValueDistribution@a, bD is the limiting distribution for the largest values in large
samples drawn from a variety of distributions, including the normal distribution. The limiting
distribution for the smallest values in such samples is the Gumbel distribution,
GumbelDistribution@a, bD. The names "extreme value" and "Gumbel distribution" are some-
times used interchangeably because the distributions of the largest and smallest extreme val-
ues are related by a linear change of variable. The extreme value distribution is also sometimes
referred to as the log-Weibull distribution because of logarithmic relationships between an
extreme value-distributed random variable and a properly shifted and scaled Weibull-distributed
random variable.

PDF@dist,xD probability density function at x

CDF@dist,xD cumulative distribution function at x
InverseCDF@dist,qD the value of x such that CDF@dist, xD equals q
314 Mathematics and Algorithms

Quantile@dist,qD qth quantile

InterquartileRange@distD difference between the first and third quartiles

QuartileDeviation@distD half the interquartile range
QuartileSkewness@distD quartile-based skewness measure
RandomReal@distD pseudorandom number with specified distribution
RandomReal@dist,dimsD pseudorandom array with dimensionality dims, and ele-
ments from the specified distribution

Functions of statistical distributions.

The cumulative distribution function (cdf) at x is given by the integral of the probability density
function (pdf) up to x. The pdf can therefore be obtained by differentiating the cdf (perhaps in a
generalized sense). In this package the distributions are represented in symbolic form.
PDF@dist, xD evaluates the density at x if x is a numerical value, and otherwise leaves the func-
tion in symbolic form. Similarly, CDF@dist, xD gives the cumulative distribution.

The inverse cdf InverseCDF@dist, qD gives the value of x at which CDF@dist, xD reaches q. The
median is given by InverseCDF@dist, 1 2D. Quartiles, deciles and percentiles are particular
values of the inverse cdf. Quartile skewness is equivalent to Hq1 - 2 q2 + q3 L Hq3 - q1 L, where q1 , q2
and q3 are the first, second, and third quartiles, respectively. Inverse cdfs are used in construct-
ing confidence intervals for statistical parameters. InverseCDF@dist, qD and Quantile@dist, qD
are equivalent for continuous distributions.

The mean Mean@distD is the expectation of the random variable distributed according to dist and
is usually denoted by m. The mean is given by y x f HxL x, where f HxL is the pdf of the distribu-

tion. The variance Variance@distD is given by Hx - mL2 f HxL x. The square root of the variance is

called the standard deviation, and is usually denoted by s.

The Skewness@distD and Kurtosis@distD functions give shape statistics summarizing the asymme-
1
try and the peakedness of a distribution, respectively. Skewness is given by 3
Hx - mL f HxL x
s3
Mathematics and Algorithms 315

The Skewness@distD and Kurtosis@distD functions give shape statistics summarizing the asymme -
1
try and the peakedness of a distribution, respectively. Skewness is given by 3
Hx - mL f HxL x
s3

1
and kurtosis is given by 4
Hx - mL f HxL x.
s4

The characteristic function CharacteristicFunction@dist, tD is given by f HtL = f HxL expHi t xL x.

In the discrete case, f HtL = f HxL expHi t xL. Each distribution has a unique characteristic function,
which is sometimes used instead of the pdf to define a distribution.

The expected value ExpectedValue@g, distD of a function g is given by f HxL gHxL x. In the dis-

crete case, the expected value of g is given by f HxL gHxL. ExpectedValue@g@xD, dist, xD is equiva-
lent to ExpectedValue@g, distD.

RandomReal@distD gives pseudorandom numbers from the specified distribution.

This gives a symbolic representation of the gamma distribution with a = 3 and b = 1.

In[1]:= gdist = GammaDistribution@3, 1D
Out[1]= GammaDistribution@3, 1D

Here is the cumulative distribution function evaluated at 10.

In[2]:= CDF@gdist, 10D
Out[2]= GammaRegularized@3, 0, 10D

This is the cumulative distribution function. It is given in terms of the built-in function
GammaRegularized .
In[3]:= cdfunction = CDF@gdist, xD
Out[3]= GammaRegularized@3, 0, xD

Here is a plot of the cumulative distribution function.

In[4]:= Plot@cdfunction, 8x, 0, 10<D
1.0

0.8

0.6
Out[4]=
0.4

0.2

2 4 6 8 10
316 Mathematics and Algorithms

This is a pseudorandom array with elements distributed according to the gamma distribution.
In[5]:= RandomReal@gdist, 5D
Out[5]= 81.46446, 8.56359, 2.70647, 1.97748, 2.97108<

Partitioning Data into Clusters

Cluster analysis is an unsupervised learning technique used for classification of data. Data
elements are partitioned into groups called clusters that represent proximate collections of data
elements based on a distance or dissimilarity function. Identical element pairs have zero dis-
tance or dissimilarity, and all others have positive distance or dissimilarity.

FindClusters@dataD partition data into lists of similar elements

FindClusters@data,nD partition data into exactly n lists of similar elements

General clustering function.

The data argument of FindClusters can be a list of data elements or rules indexing elements
and labels.

8e1 ,e2 ,< data specified as a list of data elements ei

8e1 v1 ,e2 v2 ,< data specified as a list of rules between data elements ei
and labels vi

8e1 ,e2 ,<8v1 ,v2 ,< data specified as a rule mapping data elements ei to labels
vi

Ways of specifying data in FindClusters .

The data elements ei can be numeric lists, matrices, tensors, lists of True and False elements,
or lists of strings. All data elements ei must have the same dimensions.

Here is a list of numbers.

In[1]:= data = 81.2, 9.1, 2.3, 15.4, 71.8<;

FindClusters clusters the numbers based on their proximity.

In[2]:= FindClusters@dataD
Out[2]= 881.2, 2.3<, 89.1, 15.4<, 871.8<<

The rule-based data syntax allows for clustering data elements and returning labels for those
elements.
Mathematics and Algorithms 317

The rule-based data syntax allows for clustering data elements and returning labels for those
elements.

Here two-dimensional points are clustered and labeled with their positions in the data list.
In[3]:= data1 = 881, 2<, 83, 7<, 80, 3<, 83, 1<< ;
FindClusters@data1 -> Range@Length@data1DDD
Out[3]= 881, 3, 4<, 82<<

The rule-based data syntax can also be used to cluster data based on parts of each data entry.
For instance, you might want to cluster data in a data table while ignoring particular columns in
the table.

Here is a list of data entries.

In[4]:= datarecords = 88"Joe", "Smith", 158, 64.4<, 8"Mary", "Davis", 137, 64.4<,
8"Bob", "Lewis", 141, 62.8<, 8"John", "Thompson", 235, 71.1<,
8"Lewis", "Black", 225, 71.4<, 8"Sally", "Jones", 168, 62.<,
8"Tom", "Smith", 243, 70.9<, 8"Jane", "Doe", 225, 71.4<<;

This clusters the data while ignoring the first two elements in each data entry.
In[5]:= FindClusters@Drop@datarecords, None, 81, 2<D datarecordsD
Out[5]= 888Joe, Smith, 158, 64.4<, 8Sally, Jones, 168, 62.<<,
88Mary, Davis, 137, 64.4<, 8Bob, Lewis, 141, 62.8<<, 88John, Thompson, 235, 71.1<,
8Lewis, Black, 225, 71.4<, 8Tom, Smith, 243, 70.9<, 8Jane, Doe, 225, 71.4<<<

In principle, it is possible to cluster points given in an arbitrary number of dimensions. How-

ever, it is difficult at best to visualize the clusters above two or three dimensions. To compare
optional methods in this documentation, an easily visualizable set of two-dimensional data will
be used.

The following commands define a set of 300 two-dimensional data points chosen to group into
four somewhat nebulous clusters.
In[6]:= GaussianRandomData@n_Integer, p_, sigma_D := TableAp + 8Re@D, Im@D< &A
RandomReal@NormalDistribution@0, sigmaDD RandomReal@80,2 p<D E, 8n<E;
datapairs = BlockRandom@
SeedRandom@1234D;
Join@GaussianRandomData@100, 82, 1<, .3D,
GaussianRandomData@100, 81, 1.5<, .2D,
GaussianRandomData@100, 81, 1.1<, .4D,
GaussianRandomData@100, 81.75, 1.75<, 0.1DDD;

This clusters the data based on the proximity of points.

In[7]:= cl = FindClusters@datapairsD;
318 Mathematics and Algorithms

Here is a plot of the clusters.

In[8]:= ListPlot@clD

2.0

1.5

Out[8]= 1.0
0.5

0.5 1.0 1.5 2.0 2.5

With the default settings, FindClusters has found the four clusters of points.

You can also direct FindClusters to find a specific number of clusters.

This shows the effect of choosing 3 clusters.

In[9]:= ListPlot@FindClusters@datapairs, 3DD

2.0

1.5

Out[9]= 1.0
0.5

0.5 1.0 1.5 2.0 2.5

This shows the effect of choosing 5 clusters.

In[10]:= ListPlot@FindClusters@datapairs, 5DD

2.0

1.5

Out[10]= 1.0
0.5

0.5 1.0 1.5 2.0 2.5

option name default value

DistanceFunction Automatic the distance or dissimilarity measure to use

Method Automatic the clustering method to use

Options for FindClusters .

Randomness is used in clustering in two different ways. Some of the methods use a random
assignment of some points to a specific number of clusters as a starting point. Randomness
may also be used to help determine what seems to be the best number of clusters to use.
Changing the random seed for generating the randomness by using
Mathematics and Algorithms 319

In principle, clustering techniques can be applied to any set of data. All that is needed is a
measure of how far apart each element in the set is from other elements, that is, a function
giving the distance between elements.

FindClusters@8e1 , e2 , <, DistanceFunction -> f D treats pairs of elements as being less

similar when their distances f @ei , e j D are larger. The function f can be any appropriate distance

or dissimilarity function. A dissimilarity function f satisfies the following:

f Hei , ei L = 0
f Iei , e j M 0
f Iei , e j M = f Ie j , ei M

If the ei are vectors of numbers, FindClusters by default uses a squared Euclidean distance. If
the ei are lists of Boolean True and False (or 0 and 1) elements, FindClusters by default uses
a dissimilarity based on the normalized fraction of elements that disagree. If the ei are strings,
FindClusters by default uses a distance function based on the number of point changes
needed to get from one string to another.

EuclideanDistance@u,vD the Euclidean norm Hu - vL2

SquaredEuclideanDistance@u,vD squared Euclidean norm Hu - vL2

ManhattanDistance@u,vD the Manhattan distance u - v
ChessboardDistance@u,vD the chessboard or Chebyshev distance maxHu - vL
CanberraDistance@u,vD the Canberra distance u - v Hu + vL
CosineDistance@u,vD the cosine distance 1 - u.v Hu vL
CorrelationDistance@u,vD the correlation distance
1 - Hu - [email protected] - Mean@vDL
Hu - Mean@uD v - Mean@vDL
BrayCurtisDistance@u,vD the Bray|Curtis distance u - v u + v

Distance functions for numerical data.

This shows the clusters in datapairs found using a Manhattan distance.

320 Mathematics and Algorithms

This shows the clusters in datapairs found using a Manhattan distance.

In[11]:= ListPlot@FindClusters@datapairs, DistanceFunction ManhattanDistanceDD
2.0

1.5

Out[11]= 1.0

0.5

0.5 1.0 1.5 2.0 2.5

Dissimilarities for Boolean vectors are typically calculated by comparing the elements of two
Boolean vectors u and v pairwise. It is convenient to summarize each dissimilarity function in
terms of ni j , where ni j is the number of corresponding pairs of elements in u and v, respectively,

equal to i and j. The number ni j counts the pairs 8i, j< in 8u1 , v1 <, 8u2 , v2 < , with i and j being

either 0 or 1. If the Boolean values are True and False, True is equivalent to 1 and False is
equivalent to 0.

MatchingDissimilarity@u,vD simple matching Hn10 + n01 L Length@uD

JaccardDissimilarity@u,vD the Jaccard dissimilarity Hn10 + n01 L Hn11 + n10 + n01 L

RussellRaoDissimilarity@u,vD the Russell|Rao dissimilarity Hn10 + n01 + n00 L Length@uD

SokalSneathDissimilarity@u,vD the Sokal|Sneath dissimilarity 2 Hn10 + n01 L Hn11 + 2 Hn10 + n01 LL

RogersTanimotoDissimilarity@ the Rogers|Tanimoto dissimilarity

u,vD 2 Hn10 + n01 L Hn11 + 2 Hn10 + n01 L + n00 L
DiceDissimilarity@u,vD the dice dissimilarity Hn10 + n01 L H2 n11 + n10 + n01 L

YuleDissimilarity@u,vD the Yule dissimilarity 2 n10 n01 Hn11 n00 + n10 n01 L

Dissimilarity functions for Boolean data.

Here is some Boolean data.

In[12]:= bdata = 88False, False, False, False, False, True, False, False, True, True<,
8True, False, False, False, False, False, False, False, False, True<,
8True, False, False, True, False, False, True, False, True, True<,
8True, True, False, False, True, False, False, False, True, True<,
8True, True, False, False, True, True, True, True, True, True<<;

These are the clusters found using the default dissimilarity for Boolean data.
In[13]:= FindClusters@bdataD
Out[13]= 888False, False, False, False, False, True, False, False, True, True<<,
88True, False, False, False, False, False, False, False, False, True<,
8True, False, False, True, False, False, True, False, True, True<,
8True, True, False, False, True, False, False, False, True, True<,
8True, True, False, False, True, True, True, True, True, True<<<
Mathematics and Algorithms 321

EditDistance@u,vD the number of edits to transform u into string v

DamerauLevenshteinDistance@u,vD Damerau|Levenshtein distance between u and v
HammingDistance@u,vD the number of elements whose values disagree in u and v

Dissimilarity functions for string data.

The edit distance is determined by counting the number of deletions, insertions, and substitu-
tions required to transform one string into another while preserving the ordering of characters.
In contrast, the Damerau|Levenshtein distance counts the number of deletions, insertions,
substitutions, and transpositions, while the Hamming distance counts only the number of substi-
tutions.

Here is some string data.

In[14]:= sdata = 8"The", "quick", "brown", "fox", "jumps", "over", "the", "lazy", "dog"<;

This clusters the string data using the edit distance.

In[15]:= FindClusters@sdataD
Out[15]= 88The, fox, over, the, lazy, dog<, 8quick, brown, jumps<<

The Method option can be used to specify different methods of clustering.

"Agglomerate" find clustering hierarchically

"Optimize" find clustering by local optimization

Explicit settings for the Method option.

The methods "Agglomerate" and "Optimize" determine how to cluster the data for a particular
number of clusters k. "Agglomerate" uses an agglomerative hierarchical method starting with
each member of the set in a cluster of its own and fusing nearest clusters until there are k
remaining. "Optimize" starts by building a set of k representative objects and clustering
around those, iterating until a (locally) optimal clustering is found. The default "Optimize"
method is based on partitioning around medoids.

Additional Method suboptions are available to allow for more control over the clustering. Avail-
able suboptions depend on the Method chosen.

"SignificanceTest" test for identifying the best number of clusters

Suboption for all methods.

For a given set of data and distance function, the choice of the best number of clusters k may
be unclear. With Method -> 8methodname, "SignificanceTest" -> "stest"<, "stest" is used to
determine statistically significant clusters to help choose an appropriate number. Possible
values of "stest" are "Silhouette" and "Gap". The "Silhouette" test uses the silhouette
statistic to test how well the data is clustered. The "Gap" test uses the gap statistic to deter-
mine how well the data is clustered.

The "Silhouette" test subdivides the data into successively more clusters looking for the first
minimum of the silhouette statistic.

The "Gap" test compares the dispersion of clusters generated from the data to that derived
from a sample of null hypothesis sets. The null hypothesis sets are uniformly randomly dis-
tributed data in the box defined by the principal components of the input data. The "Gap"
method takes two suboptions: "NullSets" and "Tolerance". The suboption "NullSets" sets
the number of null hypothesis sets to compare with the input data. The option "Tolerance"
sets the sensitivity. Typically larger values of "Tolerance" will favor fewer clusters being cho-
sen. The default settings are "NullSets" -> 5 and "Tolerance" -> 1.

This shows the result of clustering datapairs using the "Silhouette" test.
In[16]:= ListPlot@FindClusters@datapairs,
Method 8Automatic, "SignificanceTest" -> "Silhouette"<DD
2.0

1.5

Out[16]= 1.0

0.5

0.5 1.0 1.5 2.0 2.5

Here are the clusters found using the "Gap" test with the tolerance parameter set to 3. The
larger value leads to fewer clusters being selected.
In[17]:= ListPlot@FindClusters@datapairs,
Method 8Automatic, "SignificanceTest" 8"Gap", "Tolerance" 3<<DD
2.0

1.5

Out[17]= 1.0

0.5

0.5 1.0 1.5 2.0 2.5

Note that the clusters found in these two examples are identical. The only difference is how the
number of clusters is chosen.
Mathematics and Algorithms 323

Note that the clusters found in these two examples are identical. The only difference is how the
number of clusters is chosen.

"Linkage" the clustering linkage to use

Suboption for the "Agglomerate" method.

With Method -> 8"Agglomerate", "Linkage" -> f <, the specified linkage function f is used for
agglomerative clustering.

"Single" smallest intercluster dissimilarity

"Average" average intercluster dissimilarity
"Complete" largest intercluster dissimilarity
"WeightedAverage" weighted average intercluster dissimilarity
"Centroid" distance from cluster centroids
"Median" distance from cluster medians
"Ward" Ward's minimum variance dissimilarity
f a pure function

Possible values for the "Linkage" suboption.

Linkage methods determine this intercluster dissimilarity, or fusion level, given the dissimilari-
ties between member elements.

With Linkage -> f , f is a pure function that defines the linkage algorithm. Distances or dissimi-
larities between clusters are determined recursively using information about the distances or
dissimilarities between unmerged clusters to determine the distances or dissimilarities for the
newly merged cluster. The function f defines a distance from a cluster k to the new cluster
formed by fusing clusters i and j. The arguments supplied to f are dik , d jk , dij , ni , n j , and nk ,

where d is the distance between clusters and n is the number of elements in a cluster.

These are the clusters found using complete linkage hierarchical clustering.
In[18]:= ListPlot@FindClusters@datapairs, Method 8"Agglomerate", "Linkage" "Complete"<DD

2.0

1.5

Out[18]= 1.0

0.5

0.5 1.0 1.5 2.0 2.5

324 Mathematics and Algorithms

"Iterations" the maximum number of iterations to use

Suboption for the "Optimize" method.

Here are the clusters determined from a single iteration of the "Optimize" method.
In[19]:= ListPlot@FindClusters@datapairs, Method 8"Optimize", "Iterations" 1<DD

2.0

1.5

Out[19]= 1.0

0.5

0.5 1.0 1.5 2.0 2.5

Using Nearest
Nearest is used to find elements in a list that are closest to a given data point.

Nearest@8elem1 ,elem2 ,<,xD give the list of elemi to which x is nearest

Nearest@8elem1 ->v1 ,elem2 ->v2 ,<,xD

give the vi corresponding to the elemi to which x is nearest

Nearest@8elem1 ,elem2 ,<->8v1 ,v2 ,<,xD

give the same result
NearestA8elem1 ,elem2 ,<->Automatic,xE
take the vi to be the integers 1, 2, 3,

Nearest@data,x,nD give the n nearest elements to x

Nearest@data,x,8n,r<D give up to the n nearest elements to x within a radius r
Nearest@dataD generate a NearestFunction@D which can be applied
repeatedly to different x

Nearest function.

Nearest works with numeric lists, tensors, or a list of strings.

This finds the elements nearest to 4.5.

In[1]:= Nearest@81, 2, 3, 4, 5, 6, 7, 8<, 4.5D
Out[1]= 84, 5<

This finds 3 elements nearest to 4.5.

Mathematics and Algorithms 325

This finds 3 elements nearest to 4.5.

In[2]:= Nearest@81, 2, 3, 4, 5, 6, 7, 8<, 4.5, 3D
Out[2]= 84, 5, 3<

This finds all elements nearest to 4.5 within a radius of 2.

In[3]:= Nearest@81, 2, 3, 4, 5, 6, 7, 8<, 4.5, 8Infinity, 2<D
Out[3]= 84, 5, 3, 6<

This finds the points nearest to 81, 2< in 2D.

In[4]:= Nearest@881, 1<, 82, 2<, 83, 3<<, 81, 2<D
Out[4]= 881, 1<, 82, 2<<

This finds the nearest string to "cat".

In[5]:= Nearest@8"bat", "sad", "cake"<, "cat"D
Out[5]= 8bat<

The rule-based data syntax lets you use nearest elements to return their labels.

Here two-dimensional points are labeled.

In[6]:= Nearest@881, 1< a, 82, 2< b, 83, 3< c<, 81, 2<D
Out[6]= 8a, b<

In[7]:= Nearest@881, 1<, 82, 2<, 83, 3<< 8a, b, c<, 81, 2<D
Out[7]= 8a, b<

This labels the elements using successive integers.

In[8]:= Nearest@881, 1<, 82, 2<, 83, 3<, 84, 5<, 87, 7<< Automatic, 84, 4<D
Out[8]= 84<

If Nearest is to be applied repeatedly to the same numerical data, you can get significant
performance gains by first generating a NearestFunction.

This generates a set of 10,000 points in 2D and a NearestFunction .

In[9]:= pts = RandomReal@1, 810 000, 2<D;
nf = Nearest@ptsD
Out[9]= NearestFunction@810 000, 2<, <>D

This finds points in the set that are closest to the 10 target points.
326 Mathematics and Algorithms

This finds points in the set that are closest to the 10 target points.
In[10]:= target = RandomReal@1, 810, 2<D;
res = Map@nf, targetD; Timing
-16
Out[10]= 94.85723 10 , Null=

It takes much longer if NearestFunction is not used.

In[11]:= res2 = Map@Nearest@pts, D &, targetD; Timing
Out[11]= 80.504032, Null<

In[12]:= res res2

Out[12]= True

option name default value

DistanceFunction Automatic the distance metric to use

Option for Nearest .

For numerical data, by default Nearest uses the EuclideanDistance. For strings,
EditDistance is used.

Manipulating Numerical Data

When you have numerical data, it is often convenient to find a simple formula that approxi-
mates it. For example, you can try to "fit" a line or curve through the points in your data.

FitA9y1 ,y2 ,=,9 f1 , f2 ,=,xE fit the values yn to a linear combination of functions fi

FitA99x1 ,y1 =,9x2 , fit the points Hxn , yn L to a linear combination of the fi
y2 =,=,9 f1 , f2 ,=,xE

Fitting curves to linear combinations of functions.

This generates a table of the numerical values of the exponential function. Table is discussed
in "Making Tables of Values".
In[1]:= data = Table@Exp@x 5.D, 8x, 7<D
Out[1]= 81.2214, 1.49182, 1.82212, 2.22554, 2.71828, 3.32012, 4.0552<
Mathematics and Algorithms 327

This finds a least-squares fit to data of the form c1 + c2 x + c3 x2 . The elements of data are
assumed to correspond to values 1, 2, ... of x.
In[2]:= Fit@data, 81, x, x ^ 2<, xD
2
Out[2]= 1.09428 + 0.0986337 x + 0.0459482 x

This finds a fit of the form c1 + c2 x + c3 x3 + c4 x5 .

In[3]:= Fit@data, 81, x, x ^ 3, x ^ 5<, xD
3 -6 5
Out[3]= 0.96806 + 0.246829 x + 0.00428281 x - 6.57948 10 x

This gives a table of x, y pairs.

In[4]:= data = Table@8x, Exp@Sin@xDD<, 8x, 0., 1., 0.2<D
Out[4]= 880., 1.<, 80.2, 1.21978<, 80.4, 1.47612<, 80.6, 1.75882<, 80.8, 2.04901<, 81., 2.31978<<

This finds a fit to the new data, of the form c1 + c2 sin HxL + c3 sin H2 xL.
In[5]:= Fit@%, 81, Sin@xD, Sin@2 xD<, xD
Out[5]= 0.989559 + 2.04199 Sin@xD - 0.418176 Sin@2 xD

FindFit@data, form,8p1 ,p2 ,<,xD

find a fit to form with parameters pi

Fitting data to general forms.

This finds the best parameters for a linear fit.

In[6]:= FindFit@data, a + b x + c x ^ 2, 8a, b, c<, xD
Out[6]= 8a 0.991251, b 1.16421, c 0.174256<

This does a nonlinear fit.

In[7]:= FindFit@data, a + b ^ Hc + d xL, 8a, b, c, d<, xD
Out[7]= 8a -3.65199, b 1.65838, c 3.03496, d 0.50107<

One common way of picking out "signals" in numerical data is to find the Fourier transform, or
frequency spectrum, of the data.

Fourier@dataD numerical Fourier transform

InverseFourier@dataD inverse Fourier transform

Fourier transforms.

Here is a simple square pulse.

328 Mathematics and Algorithms

Here is a simple square pulse.

In[8]:= data = 81, 1, 1, 1, - 1, - 1, - 1, - 1<
Out[8]= 81, 1, 1, 1, -1, -1, -1, -1<

This takes the Fourier transform of the pulse.

In[9]:= Fourier@dataD
Out[9]= 80. + 0. , 0.707107 + 1.70711 , 0. + 0. , 0.707107 + 0.292893 ,
0. + 0. , 0.707107 - 0.292893 , 0. + 0. , 0.707107 - 1.70711 <

Note that the Fourier function in Mathematica is defined with the sign convention typically
used in the physical sciences~opposite to the one often used in electrical engineering. "Fourier
Transforms" gives more details.

Curve Fitting
There are many situations where one wants to find a formula that best fits a given set of data.
One way to do this in Mathematica is to use Fit.

Fit@8 f1 , f2 ,<,8 fun1 , fun2 ,<,xD find a linear combination of the funi that best fits the
values fi

Basic linear fitting.

Here is a table of the first 20 primes.

In[1]:= fp = Table@Prime@xD, 8x, 20<D
Out[1]= 82, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71<

Here is a plot of this "data".

In[2]:= gp = ListPlot@fpD
70

40
Out[2]=
30

5 10 15 20

This gives a linear fit to the list of primes. The result is the best linear combination of the
functions 1 and x.
Mathematics and Algorithms 329

This gives a linear fit to the list of primes. The result is the best linear combination of the
functions 1 and x.
In[3]:= Fit@fp, 81, x<, xD
Out[3]= -7.67368 + 3.77368 x

Here is a plot of the fit.

In[4]:= Plot@%, 8x, 0, 20<D

Out[4]= 30

5 10 15 20

Here is the fit superimposed on the original data.

In[5]:= Show@%, gpD

Out[5]= 30

5 10 15 20

This gives a quadratic fit to the data.

In[6]:= Fit@fp, 81, x, x ^ 2<, xD
2
Out[6]= -1.92368 + 2.2055 x + 0.0746753 x
330 Mathematics and Algorithms

Here is a plot of the quadratic fit.

In[7]:= Plot@%, 8x, 0, 20<D
70

40
Out[7]=
30

5 10 15 20

This shows the fit superimposed on the original data. The quadratic fit is better than the linear
one.
In[8]:= Show@%, gpD
70

40
Out[8]=
30

5 10 15 20

8 f1 , f2 ,< data points obtained when a single coordinate takes on

values 1, 2,
88x1 , f1 <,8x2 , f2 <,< data points obtained when a single coordinate takes on
values x1 , x2 ,
88x1 ,y1 ,, f1 <,8x2 ,y2 ,, f2 <,< data points obtained with values xi , yi , of a sequence of
coordinates

Ways of specifying data.

If you give data in the form 8 f1 , f2 , < then Fit will assume that the successive fi correspond to
values of a function at successive integer points 81, 2, <. But you can also give Fit data that
corresponds to the values of a function at arbitrary points, in one or more dimensions.

Fit@data,8 fun1 , fun2 ,<,8x,y,<D fit to a function of several variables

Multivariate fitting.

This gives a table of the values of x, y and 1 + 5 x - x y. You need to use Flatten to get it in the
right form for Fit.
Mathematics and Algorithms 331

This gives a table of the values of x, y and 1 + 5 x - x y. You need to use Flatten to get it in the
right form for Fit.
In[9]:= Flatten@Table@8x, y, 1 + 5 x - x y<, 8x, 0, 1, 0.4<, 8y, 0, 1, 0.4<D, 1D
Out[9]= 880., 0., 1.<, 80., 0.4, 1.<, 80., 0.8, 1.<, 80.4, 0., 3.<, 80.4, 0.4, 2.84<,
80.4, 0.8, 2.68<, 80.8, 0., 5.<, 80.8, 0.4, 4.68<, 80.8, 0.8, 4.36<<

This produces a fit to a function of two variables.

In[10]:= Fit@%, 81, x, y, x y<, 8x, y<D
-15
Out[10]= 1. + 5. x + 5.21108 10 y - 1. x y

Fit takes a list of functions, and uses a definite and efficient procedure to find what linear
combination of these functions gives the best least-squares fit to your data. Sometimes, how-
ever, you may want to find a nonlinear fit that does not just consist of a linear combination of
specified functions. You can do this using FindFit, which takes a function of any form, and
then searches for values of parameters that yield the best fit to your data.

FindFit@data, form, search for values of the pari that make form best fit data
8par1 ,par2 ,<,xD
FindFit@data, form,pars,8x,y,<D fit multivariate data

Searching for general fits to data.

This fits the list of primes to a simple linear combination of terms.

In[11]:= FindFit@fp, a + b x + c Exp@xD, 8a, b, c<, xD
-8
Out[11]= 9a -6.78932, b 3.64309, c 1.26883 10 =

The result is the same as from Fit.

In[12]:= Fit@fp, 81, x, Exp@xD<, xD
-8 x
Out[12]= -6.78932 + 1.26883 10 + 3.64309 x

This fits to a nonlinear form, which cannot be handled by Fit.

In[13]:= FindFit@fp, a x Log@b + c xD, 8a, b, c<, xD
Out[13]= 8a 1.42076, b 1.65558, c 0.534645<

By default, both Fit and FindFit produce least-squares fits, which are defined to minimize the
quantity c2 = i ri 2 , where the ri are residuals giving the difference between each original data

NormFunction -> u, then FindFit will attempt to find the fit that minimizes the
quantity u@rD, where r is the list of residuals. The default is NormFunction -> Norm, correspond-
ing to a least-squares fit.
By default, both Fit and FindFit produce least-squares fits, which are defined to minimize the
332 Mathematics and Algorithms
quantity c2 = r 2
, where the ri are residuals giving the difference between each original data

point and its fitted value. One can, however, also consider fits based on other norms. If you set
the option NormFunction -> u, then FindFit will attempt to find the fit that minimizes the
quantity u@rD, where r is the list of residuals. The default is NormFunction -> Norm, correspond-
ing to a least-squares fit.

This uses the -norm, which minimizes the maximum distance between the fit and the data.
The result is slightly different from least-squares.
In[14]:= FindFit@fp, a x Log@b + c xD, 8a, b, c<, x, NormFunction -> HNorm@, InfinityD &LD
Out[14]= 8a 1.15077, b 1.0023, c 1.04686<

FindFit works by searching for values of parameters that yield the best fit. Sometimes you
may have to tell it where to start in doing this search. You can do this by giving parameters in
the form 88a, a0 <, 8b, b0 <, <. FindFit also has various options that you can set to control how it
does its search.

FindFit@data, finds a best fit subject to the parameter constraints cons

8 form,cons<,pars,varsD

Searching for general fits to data.

This gives a best fit subject to constraints on the parameters.

In[15]:= FindFit@fp, 8a x Log@b + c xD, 80 <= a <= 1, 0 <= b 1, c 1<<, 8a, b, c<, xD
-9
Out[15]= 9a 1., b 1.34569 10 , c 1.69145=

option name default value

NormFunction Norm the norm to use
AccuracyGoal Automatic number of digits of accuracy to try to get
PrecisionGoal Automatic number of digits of precision to try to get
WorkingPrecision Automatic precision to use in internal computations
MaxIterations Automatic maximum number of iterations to use
StepMonitor None expression to evaluate whenever a step is
taken
EvaluationMonitor None expression to evaluate whenever form is
evaluated
Method Automatic method to use

Options for FindFit .

Mathematics and Algorithms 333

Statistical Model Analysis

When fitting models of data, it is often useful to analyze how well the model fits the data and
how well the fitting meets assumptions of the fitting. For a number of common statistical mod-
els, this is accomplished in Mathematica by way of fitting functions that construct FittedModel
objects.

FittedModel represents a symbolic fitted model

Object for fitted model information.

FittedModel objects can be evaluated at a point or queried for results and diagnostic informa-
tion. Diagnostics vary somewhat across model types. Available model fitting functions fit linear,
generalized linear, and nonlinear models.

LinearModelFit constructs a linear model

GeneralizedLinearModelFit constructs a generalized linear model

LogitModelFit constructs a binomial logistic regression model

ProbitModelFit constructs a binomial probit regression model

NonlinearModelFit constructs a nonlinear least-squares model

Functions that generate FittedModel objects.

This fits a linear model assuming x values 1, 2, .

In[1]:= lm = [email protected], 3.4, 7.1, 8.3, 10.4<, x, xD

Out[1]= FittedModelB -0.67 + 2.27 x F

Here is the functional form of the fitted model.

In[2]:= Normal@lmD
Out[2]= -0.67 + 2.27 x

This evaluates the model for x = 2.5.

In[3]:= [email protected]
Out[3]= 5.005
334 Mathematics and Algorithms

Here is a shortened list of available results for the linear fitted model.
In[4]:= lm@"Properties"D Short
Out[4]//Short= 8AdjustedRSquared, AIC, 58, StudentizedResiduals, VarianceInflationFactors<

The major difference between model fitting functions such as LinearModelFit and functions
such as Fit and FindFit is the ability to easily obtain diagnostic information from the
FittedModel objects. The results are accessible without refitting the model.

This gives the residuals for the fitting.

In[5]:= lm@"FitResiduals"D
Out[5]= 8-0.1, -0.47, 0.96, -0.11, -0.28<

Here multiple results are obtained at once.

In[6]:= lm@8"BestFitParameters", "ANOVATable"<D
DF SS MS F Statistic P-Value
x 1 51.529 51.529 124.366 0.00154521 >
Out[6]= :8-0.67, 2.27<,
Error 3 1.243 0.414333
Total 4 52.772

Fitting options relevant to property computations can be passed to FittedModel objects to

override defaults.

This gives default 95% confidence intervals.

In[7]:= lm@"ParameterConfidenceIntervals"D
Out[7]= 88-2.81849, 1.47849<, 81.62221, 2.91779<<

Here 90% intervals are obtained.

In[8]:= lm@"ParameterConfidenceIntervals", ConfidenceLevel .9D
Out[8]= 88-2.25877, 0.918767<, 81.79097, 2.74903<<

Typical data for these model fitting functions takes the same form as data in other fitting func-
tions such as Fit and FindFit.

8y1 ,y2 ,< data points with a single predictor variable taking values 1,
2,
88x11 ,x12 ,,y1 <,8x21 ,x22 ,,y2 <,< data points with explicit coordinates

Data specifications.

Linear Models
Mathematics and Algorithms 335

Linear Models
Linear models with assumed independent normally distributed errors are among the most
common models for data. Models of this type can be fitted using the LinearModelFit function.

LinearModelFit@ obtain a linear model with basis functions fi and a single

8y1 ,y2 ,<,8 f1 , f2 ,<,xD predictor variable x
LinearModelFit@ obtain a linear model of multiple predictor variables xi
88x11 ,x12 ,,y1 <,8x21 ,x22 ,,y2 <<,
8 f1 , f2 ,<,8x1 ,x2 ,<D
LinearModelFit@8m,v<D obtain a linear model based on a design matrix m and
response vector v

Linear model fitting.

` `
Linear models have the form y = b0 + b1 f1 + b2 f2 + where y is the fitted or predicted value, the
bi are parameters to be fitted, and the fi are functions of the predictor variables xi . The models
are linear in the parameters bi . The fi can be any functions of the predictor variables. Quite
often the fi are simply the predictor variables xi .

This fits a linear model to the first 20 primes.

In[9]:= lm = LinearModelFit@Array@Prime, 20D, x, xD

Out[9]= FittedModelB -7.67368 + 3.77368 x F

Options for model specification and for model analysis are available.

option name default value

ConfidenceLevel 95100 confidence level to use for parameters and
predictions
IncludeConstantBasis True whether to include a constant basis function
LinearOffsetFunction None known offset in the linear predictor
NominalVariables None variables considered as nominal or
categorical
VarianceEstimatorFunction Automatic function for estimating the error variance
Weights Automatic weights for data elements
WorkingPrecision Automatic precision used in internal computations

Options for LinearModelFit.

The Weights option specifies weight values for weighted linear regression. The
NominalVariables option specifies which predictor variables should be treated as nominal or
categorical. With NominalVariables -> All, the model is an analysis of variance (ANOVA)
model. With NominalVariables -> 8x , , x ,x , , x < the model is an analysis of covari-
336 Mathematics and Algorithms

The Weights option specifies weight values for weighted linear regression. The
NominalVariables option specifies which predictor variables should be treated as nominal or
categorical. With NominalVariables -> All, the model is an analysis of variance (ANOVA)
model. With NominalVariables -> 8x1 , , xi-1 , xi+1 , , xn < the model is an analysis of covari-

ance (ANCOVA) model with all but the ith predictor treated as nominal. Nominal variables are
represented by a collection of binary variables indicating equality and inequality to the observed
nominal categorical values for the variable.

ConfidenceLevel, VarianceEstimatorFunction , and WorkingPrecision are relevant to the

computation of results after the initial fitting. These options can be set within LinearModelFit
to specify the default settings for results obtained from the FittedModel object. These options
can also be set within an already constructed FittedModel object to override the option values
originally given to LinearModelFit.

Here are the default and mean squared error variance estimates.
In[10]:= 8lm@"EstimatedVariance"D,
lm@"EstimatedVariance", VarianceEstimatorFunction HMean@ ^ 2D &LD<
Out[10]= 86.71608, 6.04447<

IncludeConstantBasis, LinearOffsetFunction, NominalVariables, and Weights are rele-

vant only to the fitting. Setting these options within an already constructed FittedModel object
will have no further impact on the result.

A major feature of the model fitting framework is the ability to obtain results after the fitting.
The full list of available results can be obtained from the "Properties" value.

This is the number of properties available for linear models.

In[11]:= lm@"Properties"D Length
Out[11]= 62

The properties include basic information about the data, fitted model, and numerous results and
diagnostics.

"BestFit" fitted function

"BestFitParameters" parameter estimates
"Data" the input data or design matrix and response vector
Mathematics and Algorithms 337

"DesignMatrix" design matrix for the model

"Function" best-fit pure function
"Response" response values in the input data

Properties related to data and the fitted function.

The "BestFitParameters" property gives the fitted parameter values 8b0 , b1 , <. "BestFit"
is the fitted function b0 + b1 f1 + b2 f2 + and "Function" gives the fitted function as a pure
function. The "DesignMatrix" is the design or model matrix for the data. "Response" gives the
list of the response or y values from the original data.

"FitResiduals" difference between actual and predicted responses

"StandardizedResiduals" fit residuals divided by the standard error for each residual
"StudentizedResiduals" fit residuals divided by single deletion error estimates

Types of residuals.

Residuals give a measure of the point-wise difference between the fitted values and the original
responses. "FitResiduals" gives the differences between the observed and fitted values
` `
8y1 - y1 , y2 - y2 , <. "StandardizedResiduals" and "StudentizedResiduals" are scaled forms

` `2 `2
of the residuals. The ith standardized residual is Hyi - yi L s H1 - hii L wi where s is the

estimated error variance, hii is the ith diagonal element of the hat matrix, and wi is the weight
`2
for the ith data point. The ith studentized residual uses the same formula with s replaced by
` 2
sHiL , the variance estimate omitting the ith data point.

"ANOVATable" analysis of variance table

"ANOVATableDegreesOfFreedom" degrees of freedom from the ANOVA table
"ANOVATableEntries" unformatted array of values from the table
"ANOVATableFStatistics" F statistics from the table
"ANOVATableMeanSquares" mean square errors from the table
"ANOVATablePValues" p-values from the table
"ANOVATableSumsOfSquares" sums of squares from the table
"CoefficientOfVariation" response mean divided by the estimated standard deviation
338 Mathematics and Algorithms

"EstimatedVariance" estimate of the error variance

"PartialSumOfSquares" changes in model sum of squares as nonconstant basis
functions are removed
"SequentialSumOfSquares" the model sum of squares partitioned componentwise

Properties related to the sum of squared errors.

"ANOVATable" gives a formatted analysis of variance table for the model.

"ANOVATableEntries" gives the numeric entries in the table and the remaining ANOVATable
properties give the elements of columns in the table so individual parts of the table can easily
be used in further computations.

This gives a formatted ANOVA table for the fitted model.

In[12]:= lm@"ANOVATable"D
DF SS MS F Statistic P-Value

Out[12]= x 1 9470.06 9470.06 1410.06 1.49794 10-18

Error 18 120.889 6.71608
Total 19 9590.95

Here are the elements of the MS column of the table.

In[13]:= lm@"ANOVATableMeanSquares"D
Out[13]= 89470.06, 6.71608<

"CorrelationMatrix" parameter correlation matrix

"CovarianceMatrix" parameter covariance matrix
"EigenstructureTable" eigenstructure of the parameter correlation matrix
"EigenstructureTableEigenvalu eigenvalues from the table
es"
"EigenstructureTableEntries" unformatted array of values from the table
"EigenstructureTableIndexes" index values from the table
"EigenstructureTablePartitions partitioning from the table
"
"ParameterConfidenceIntervals" parameter confidence intervals
"ParameterConfidenceIntervalT table of confidence interval information for the fitted
able" parameters
"ParameterConfidenceIntervalT unformatted array of values from the table
ableEntries"
"ParameterConfidenceRegion" ellipsoidal parameter confidence region
Mathematics and Algorithms 339

"ParameterErrors" standard errors for parameter estimates

"ParameterPValues" p-values for parameter t statistics
"ParameterTable" table of fitted parameter information
"ParameterTableEntries" unformatted array of values from the table
"ParameterTStatistics" t statistics for parameter estimates
"VarianceInflationFactors" list of inflation factors for the estimated parameters

Properties and diagnostics for parameter estimates.

`2 -1
"CovarianceMatrix" gives the covariance between fitted parameters. The matrix is s HX W XN
`2
where s is the variance estimate, X is the design matrix, and W is the diagonal matrix of
weights. "CorrelationMatrix" is the associated correlation matrix for the parameter esti-
mates. "ParameterErrors" is equivalent to the square root of the diagonal elements of the
covariance matrix.

"ParameterTable" and "ParameterConfidenceIntervalTable" contain information about the

individual parameter estimates, tests of parameter significance, and confidence intervals.

Here is some data.

In[14]:= data = 888.71, 6.92, 18.89<, 86.05, 5.97, 15.08<, 86.24, 0.99, 5.92<,
88.25, 3.37, 11.39<, 86.58, 8.22, 20.77<, 84.14, 9., 21.09<, 84.35, 9.94, 24.32<,
88.99, 4.47, 13.79<, 82.82, 3.91, 10.68<, 85.14, 0.4, 3.82<<;

This fits a model using both predictor variables.

In[15]:= lm2 = LinearModelFit@data, 8x, y<, 8x, y<D

Out[15]= FittedModelB 1.40308 + 0.340391 x + 2.08429 y F

These are the formatted parameter and parameter confidence interval tables.
In[16]:= lm2@8"ParameterTable", "ParameterConfidenceIntervalTable"<D
Estimate Standard Error t Statistic P-Value Estimate Standard Error Confidence Interval
1 1.40308 0.595477 2.35622 0.0506221 1 1.40308 0.595477 8-0.00500488, 2.81116<
Out[16]= : , >
x 0.340391 0.0782093 4.35231 0.00334539 x 0.340391 0.0782093 80.155456, 0.525327<
y 2.08429 0.0496681 41.9643 1.13829 10 -9
y 2.08429 0.0496681 81.96684, 2.20174<

Here 99% confidence intervals are used in the table.

In[17]:= lm2@"ParameterConfidenceIntervalTable", ConfidenceLevel .99D
Estimate Standard Error Confidence Interval
1 1.40308 0.595477 8-0.680788, 3.48694<
Out[17]=
x 0.340391 0.0782093 80.0666993, 0.614084<
y 2.08429 0.0496681 81.91048, 2.2581<

The Estimate column of these tables is equivalent to "BestFitParameters". The t statistics are
the estimates divided by the standard errors. Each p-value is the two-sided p-value for the t
statistic and can be used to assess whether the parameter estimate is statistically significantly
different from 0. Each confidence interval gives the upper and lower bounds for the parameter
confidence interval at the level prescribed by the ConfidenceLevel option. The various
ParameterTable and ParameterConfidenceIntervalTable properties can be used to get the
columns or the unformatted array of values from the table.

"VarianceInflationFactors" is used to measure the multicollinearity between basis func-

tions. The ith inflation factor is equal to 1 I1 - R2i M where R2i is the coefficient of variation from

fitting the ith basis function to a linear function of the other basis functions. With
IncludeConstantBasis -> True, the first inflation factor is for the constant term.

"EigenstructureTable" gives the eigenvalues, condition indices, and variance partitions for
the nonconstant basis functions. The Index column gives the square root of the ratios of the
eigenvalues to the largest eigenvalue. The column for each basis function gives the proportion
of variation in that basis function explained by the associated eigenvector.
"EigenstructureTablePartitions" gives the values in the variance partitioning for all basis
functions in the table.

"BetaDifferences" DFBETAS measures of influence on parameter values

"CatcherMatrix" catcher matrix
"CookDistances" list of Cook distances
"CovarianceRatios" COVRATIO measures of observation influence
"DurbinWatsonD" Durbin-Watson d statistic for autocorrelation
"FitDifferences" DFFITS measures of influence on predicted values
"FVarianceRatios" FVARATIO measures of observation influence
"HatDiagonal" diagonal elements of the hat matrix

"SingleDeletionVariances" list of variance estimates with the ith data point omitted

Properties related to influence measures.

Point-wise measures of influence are often employed to assess whether individual data points
have a large impact on the fitting. The hat matrix and catcher matrix play important roles in
`
such diagnostics. The hat matrix is the matrix H such that y = H y where y is the observed
`
y is the predicted response vector. "HatDiagonal" gives the diagonal
elements of the hat matrix. "CatcherMatrix" is the matrix C such that b = C y where b is the
fitted parameter vector.
Point-wise measures of influence are often employed to assess whether individual data points
have a large impact on the fitting. The hat matrix and catcher matrix play important rolls in
Mathematics and Algorithms 341
`
`
response vector and y is the predicted response vector. "HatDiagonal" gives the diagonal
elements of the hat matrix. "CatcherMatrix" is the matrix C such that b = C y where b is the
fitted parameter vector.

"FitDifferences" gives the DFFITS values that provide a measure of influence of each data

point on the fitted or predicted values. The ith DFFITS value is given by hii H1 - hii L rti where hii

is the ith hat diagonal and rti is the ith studentized residual.

"BetaDifferences" gives the DFBETAS values that provide measures of influence of each data
point on the parameters in the model. For a model with p parameters, the ith element of
"BetaDifferences" is a list of length p with the jth value giving the measure the of the influ-
ence of data point i on the jth parameter in the model. The ith "BetaDifferences" vector can

rti H1 - hii L Inj=1 k=1 c2jk M where c jk is the j,kth element of the catcher
p
be written as 9ci1 , , cip =

matrix.

"CookDistances" gives the Cook distance measures of leverage given. The ith Cook distance is
given by Hhii H1 - hii L rsi p where rsi is the ith standardized residual.

The ith element of "CovarianceRatios" is given by Hn - pL p IH1 - hii L Irti2 + n - p - 1M M and the ith
p

` 2 `2 ` 2
"FVarianceRatios" value is equal to sHiL Js H1 - hii L N where sHiL is the ith single deletion

variance.

The Durbin|Watson d statistic "DurbinWatsonD" is used for testing the existence of a first-order
autoregressive process. The d statistic is equivalent to n-1
i=1 Hri+1 - ri L i=1 ri where ri is the
2 n 2

ith residual.

This plots the Cook distances for the bivariate model.

In[18]:= ListPlot@lm2@"CookDistances"D, Filling 0D
0.7

0.6

0.5

0.4
Out[18]=
0.3

0.2

0.1

2 4 6 8 10
342 Mathematics and Algorithms

"MeanPredictionBands" confidence bands for mean predictions

"MeanPredictionConfidenceInte confidence intervals for the mean predictions
rvals"
"MeanPredictionConfidenceInte table of confidence intervals for the mean predictions
rvalTable"
"MeanPredictionConfidenceInte unformatted array of values from the table
rvalTableEntries"
"MeanPredictionErrors" standard errors for mean predictions
"PredictedResponse" fitted values for the data
"SinglePredictionBands" confidence bands based on single observations
"SinglePredictionConfidenceIn confidence intervals for the predicted response of single
tervals" observations
"SinglePredictionConfidenceIn table of confidence intervals for the predicted response of
tervalTable" single observations
"SinglePredictionConfidenceIn unformatted array of values from the table
tervalTableEntries"
"SinglePredictionErrors" standard errors for the predicted response of single
observations

Properties of predicted values.

Tabular results for confidence intervals are given by "MeanPredictionConfidenceInterval

Table " and "SinglePredictionConfidenceIntervalTable". These include the observed and
predicted responses, standard error estimates, and confidence intervals for each point. Mean
prediction confidence intervals are often referred to simply as confidence intervals and single
prediction confidence intervals are often referred to as prediction intervals.

"MeanPredictionBands" and "SinglePredictionBands" give functions of the predictor

variables.
Mathematics and Algorithms 343

Here is the mean prediction table.

In[19]:= lm2@"MeanPredictionConfidenceIntervalTable"D
Observed Predicted Standard Error Confidence Interval
18.89 18.7912 0.272818 818.1461, 19.4363<
15.08 15.9057 0.155811 815.5372, 16.2741<
5.92 5.59057 0.262819 84.9691, 6.21204<
11.39 11.2354 0.236917 810.6751, 11.7956<
Out[19]= 20.77 20.7757 0.215751 820.2656, 21.2859<
21.09 21.5709 0.271392 820.9292, 22.2127<
24.32 23.6016 0.295281 822.9034, 24.2999<
13.79 13.78 0.269774 813.1421, 14.4179<
10.68 10.5126 0.315597 89.76629, 11.2588<
3.82 3.9864 0.306026 83.26277, 4.71004<

This gives the 90% mean prediction intervals.

In[20]:= lm2@"MeanPredictionConfidenceIntervals", ConfidenceLevel .9D
Out[20]= 8818.2743, 19.3081<, 815.6105, 16.2009<, 85.09263, 6.0885<,
810.7865, 11.6842<, 820.367, 21.1845<, 821.0567, 22.0851<, 823.0422, 24.1611<,
813.2689, 14.2911<, 89.91464, 11.1105<, 83.40661, 4.56619<<

"AdjustedRSquared" R2 adjusted for the number of model parameters

"AIC" Akaike Information Criterion
"BIC" Bayesian Information Criterion

"RSquared" coefficient of determination R2

Goodness of fit measures.

Goodness of fit measures are used to assess how well a model fits or to compare models. The
coefficient of determination "RSquared" is the ratio of the model sum of squares to the total
sum of squares. "AdjustedRSquared" penalizes for the number of parameters in the model and
n -1
is given by 1 - H n -p L H1 - R2 L.

"AIC" and "BIC" are likelihood-based goodness of fit measures. Both are equal to -2 times the
log-likelihood for the model plus k p where p is the number of parameters to be estimated
including the estimated variance. For "AIC" k is 2, and for "BIC" k is logHnL.
344 Mathematics and Algorithms

Generalized Linear Models

The linear model can be seen as a model where each response value y is an observation from a
`
normal distribution with mean value y = b0 + b1 f1 + b2 f2 + . The generalized linear model
`
extends to models of the form y = g-1 H b0 + b1 f1 + b2 f2 + L with each y assumed to be an obser-
`
vation from a distribution of known exponential family form with mean y and g being an invert-
ible function over the support of the exponential family. Models of this sort can be obtained via
GeneralizedLinearModelFit .

GeneralizedLinearModelFit @ obtain a generalized linear model with basis functions fi

8y1 ,y2 ,<,8 f1 , f2 ,<,xD and a single predictor variable x
GeneralizedLinearModelFit @ obtain a generalized linear model of multiple predictor
88x11 ,x12 ,,y1 <,8x21 ,x22 ,,y2 <<, variables xi
8 f1 , f2 ,<,8x1 ,x2 ,<D
GeneralizedLinearModelFit @ obtain a generalized linear model based on a design matrix
8m,v<D m and response vector v

Generalized linear model fitting.

The invertible function g is called the link function and the linear combination b0 + b1 f1 + b2 f2 +
is referred to as the linear predictor. Common special cases include the linear regression model
with the identity link function and Gaussian or normal exponential family distribution, logit and
probit models for probabilities, Poisson models for count data, and gamma and inverse Gaus-
sian models.
`
The error variance is a function of the prediction y and is defined by the distribution up to a
constant f, which is referred to as the dispersion parameter. The error variance for a fitted
` ` ` `
value y can be written as f vHyL, where f is an estimate of the dispersion parameter obtained
`
from the observed and predicted response values, and vHyL is the variance function associated
`
with the exponential family evaluated at the value y.

This fits a linear regression model.

In[21]:= glm1 = GeneralizedLinearModelFit@Sqrt@Range@10DD, x, xD

Out[21]= FittedModelB 0.973709 + 0.231476 x F

Mathematics and Algorithms 345

This fits a canonical gamma regression model to the same data.

In[22]:= glm2 = GeneralizedLinearModelFit@
Sqrt@Range@10DD, x, x, ExponentialFamily "Gamma"D

1
Out[22]= FittedModelB F
0.742193 - 20 x

Here are the functional forms of the models.

In[23]:= Map@Normal, 8glm1, glm2<D
1
Out[23]= :0.973709 + 0.231476 x, >
0.742193 - 0.0467911 x

Logit and probit models are common binomial models for probabilities. The link function for the
y
logit model is logJ 1-y N and the link for the probit model is the inverse CDF for a standard normal

distribution 2 erf-1 H2 y - 1L. Models of this type can be fitted via GeneralizedLinearModelFit
with ExponentialFamily -> "Binomial" and the appropriate LinkFunction or via
LogitModelFit and ProbitModelFit.

LogitModelFit@data, funs,varsD obtain a logit model with basis functions funs and predictor
variables vars
LogitModelFit@8m,v<D obtain a logit model based on a design matrix m and
response vector v
ProbitModelFit@data, funs,varsD obtain a probit model fit to data
ProbitModelFit@8m,v<D obtain a probit model fit to a design matrix m and response
vector v

Logit and probit model fitting.

Parameter estimates are obtained via iteratively reweighted least squares with weights
obtained from the variance function of the assumed distribution. Options for
GeneralizedLinearModelFit include options for iteration fitting such as PrecisionGoal,
options for model specification such as LinkFunction, and options for further analysis such as
ConfidenceLevel.
346 Mathematics and Algorithms

option name default value

AccuracyGoal Automatic the accuracy sought
ConfidenceLevel 95100 confidence level to use for parameters and
predictions
CovarianceEstimatorFunct "ExpectedInfor estimation method for the parameter
ion mation" covariance matrix
DispersionEstimatorFunct function for estimating the dispersion
Automatic
ion
parameter
ExponentialFamily Automatic exponential family distribution for y
IncludeConstantBasis True whether to include a constant basis function
LinearOffsetFunction None known offset in the linear predictor
LinkFunction Automatic link function for the model
MaxIterations Automatic maximum number of iterations to use
NominalVariables None variables considered as nominal or
categorical
PrecisionGoal Automatic the precision sought
Weights Automatic weights for data elements
WorkingPrecision Automatic precision used in internal computations

Options for GeneralizedLinearModelFit .

The options for LogitModelFit and ProbitModelFit are the same as for
GeneralizedLinearModelFit except that ExponentialFamily and LinkFunction are defined
by the logit or probit model and so are not options to LogitModelFit and ProbitModelFit.

ExponentialFamily can be "Binomial", "Gamma", "Gaussian", "InverseGaussian",

"Poisson", or "QuasiLikelihood". Binomial models are valid for responses from 0 to 1. Pois-
son models are valid for non-negative integer responses. Gaussian or normal models are valid
for real responses. Gamma and inverse Gaussian models are valid for positive responses. Quasi-
likelihood models define the distributional structure in terms of a variance function vHmL such
`
y yi -m
that the log of the quasi-likelihood function for the ith data point is given by y i f vHmL m. The
i

variance function for a "QuasiLikelihood" model can be optionally set via

ExponentialFamily -> 8"QuasiLikelihood", "VarianceFunction" fun< where fun is a pure
function to be applied to fitted values.

DispersionEstimatorFunction defines a function for estimating the dispersion parameter f.

` `2
The estimate f is analogous to s in linear and nonlinear regression models.

ExponentialFamily, IncludeConstantBasis, LinearOffsetFunction, LinkFunction,

NominalVariables, and Weights all define some aspect of the model structure and optimiza-
tion criterion and can only be set within GeneralizedLinearModelFit . All other options can be
set either within GeneralizedLinearModelFit or passed to the FittedModel object when
obtaining results and diagnostics. Options set in evaluations of FittedModel objects take
Mathematics and Algorithms 347

ExponentialFamily, IncludeConstantBasis, LinearOffsetFunction, LinkFunction,

This gives 95% and 99% confidence intervals for the parameters in the gamma model.
In[24]:= 8glm2@"ParameterConfidenceIntervals"D,
glm2@"ParameterConfidenceIntervals", ConfidenceLevel .99D<
Out[24]= 8880.62891, 0.855475<, 8-0.0616093, -0.0319729<<,
880.593314, 0.891071<, 8-0.0662656, -0.0273166<<<

"BestFit" fitted function

"BestFitParameters" parameter estimates
"Data" the input data or design matrix and response vector
"DesignMatrix" design matrix for the model
"Function" best fit pure function
"LinearPredictor" fitted linear combination
"Response" response values in the input data

Properties related to data and the fitted function.

"BestFitParameters" gives the parameter estimates for the basis functions. "BestFit" gives
` ` `
the fitted function g-1 H b0 + b1 f1 + b2 f2 + L, and "LinearPredictor" gives the linear combina-
` ` `
tion b0 + b1 f1 + b2 f2 + . "DesignMatrix" is the design or model matrix for the basis functions.

"Deviances" deviances
"DevianceTable" deviance table
"DevianceTableDegreesOfFreedom degrees of freedom differences from the table
"
"DevianceTableDeviances" deviance differences from the table
"DevianceTableEntries" unformatted array of values from the table
"DevianceTableResidualDegrees residual degrees of freedom from the table
OfFreedom"
"DevianceTableResidualDevianc residual deviances from the table
es"
348 Mathematics and Algorithms

"EstimatedDispersion" estimated dispersion parameter

"NullDeviance" deviance for the null model
"NullDegreesOfFreedom" degrees of freedom for the null model
"ResidualDeviance" difference between the model deviance and null deviance
"ResidualDegreesOfFreedom" difference between the model degrees of freedom and null
degrees of freedom

Properties related to dispersion and model deviances.

Deviances and deviance tables generalize the model decomposition given by analysis of vari-
` `
ance in linear models. The deviance for a single data point is 2 fIm HyL - m HyLM where m is the log-

likelihood function for the fitted model. "Deviances" gives a list of the deviance values for all
data points. The sum of all deviances gives the model deviance. The model deviance can be
decomposed as sums of squares are in an ANOVA table for linear models.

Here is some data with two predictor variables.

In[31]:= glmdata = 883.58, 1.83, 0.21<, 83.58, 4.47, 0.17<, 83.58, 3.11, 0.19<,
83.58, 4.2, 0.18<, 83.58, 2.83, 0.2<, 84.01, 2.53, 0.19<, 84.01, 3.63, 0.18<,
84.01, 3.93, 0.17<, 84.01, 4.05, 0.17<, 84.01, 1.33, 0.22<,
82.01, 3.46, 0.22<, 82.01, 1.77, 0.25<, 82.01, 2.44, 0.23<, 82.01, 2.39, 0.25<,
82.01, 3.78, 0.2<, 83.59, 3.49, 0.19<, 83.59, 3.82, 0.19<, 83.59, 3.05, 0.2<,
83.59, 4.51, 0.17<, 83.59, 3.37, 0.18<, 82.62, 1.42, 0.25<, 82.62, 3.9, 0.19<,
82.62, 2.51, 0.21<, 82.62, 4.59, 0.18<, 82.62, 4.28, 0.19<<;

This fits the data to an inverse Gaussian model.

In[32]:= glm3 = GeneralizedLinearModelFit@glmdata,
8x, y<, 8x, y<, ExponentialFamily "InverseGaussian"D

1
Out[32]= FittedModelB F
-0.852313 + 18 x + 18 y

Here is the deviance table for the model.

In[33]:= glm3@"DevianceTable"D
DF Deviance Residual DF Residual Deviance
24 1.79112
Out[33]=
x 1 0.782767 23 1.00835
y 1 0.913425 22 0.0949267

As with sums of squares, deviances are additive. The Deviance column of the table gives the
increase in the model deviance when the given basis function is added. The Residual Deviance
column gives the difference between the model deviance and the deviance for the submodel
containing all previous terms in the table. For large samples, the increase in deviance is approxi-
mately c2 distributed with degrees of freedom equal to that for the basis function in the table.

"NullDeviance" is the deviance for the null model, the constant model equal to the mean of all
observed responses for models including a constant or g-1 H0L if a constant term is not included.
Mathematics and Algorithms 349

"NullDeviance" is the deviance for the null model, the constant model equal to the mean of all
observed responses for models including a constant or g-1 H0L if a constant term is not included.

As with "ANOVATable", a number of properties are included to extract the columns or unformat-
ted array of entries from "DevianceTable".

"AnscombeResiduals" Anscombe residuals

"DevianceResiduals" deviance residuals
"FitResiduals" difference between actual and predicted responses
"LikelihoodResiduals" likelihood residuals
"PearsonResiduals" Pearson residuals
"StandardizedDevianceResiduals standardized deviance residuals
"
"StandardizedPearsonResiduals" standardized Pearson residuals
"WorkingResiduals" working residuals

Types of residuals.

"FitResiduals" is the list of residuals, differences between the observed and predicted
responses. Given the distributional assumptions, the magnitude of the residuals is expected to
change as a function of the predicted response value. Various types of scaled residuals are
employed in the analysis of generalized linear models.
`
If di and ri = yi - yi are the deviance and residual for the ith data point, the ith deviance residual is
`
given by rdi = di sgnHri L. The ith Pearson residual is defined as r pi = ri vHyi L where v is the

variance function for the exponential family distribution. Standardized deviance residuals and

`
standardized Pearson residuals include division by fH1 - hii L where hii is the ith diagonal of the

hat matrix. "LikelihoodResiduals" values combine deviance and Pearson residuals. The ith

`
likelihood residual is given by sgnHri L Irdi 2 + hii r pi 2 H1 - hii LM f .

"AnscombeResiduals" provide a transformation of the residuals toward normality, so a plot of

these residuals should be expected to look roughly like white noise. The ith Anscombe residual
` ` y 3 -1
can be written as
3
vHyi L vHyi L y` i J vHmL N m.
i

"WorkingResiduals" gives the residuals from the last step of the iterative fitting. The ith work-
gHmL `
ing residual can be obtained as ri m evaluated at m = yi .
350 Mathematics and Algorithms

"WorkingResiduals" gives the residuals from the last step of the iterative fitting. The ith work-
gHmL `
ing residual can be obtained as ri m evaluated at m = yi .

This plots the residuals and Anscombe residuals for the inverse Gaussian model.
In[41]:= Map@ListPlot@, Filling 0D &, glm3@8"FitResiduals", "AnscombeResiduals"<DD

0.10
0.010

0.05
0.005

Out[41]= : , >
5 10 15 20 25
5 10 15 20 25

-0.005 -0.05

"CorrelationMatrix" asymptotic parameter correlation matrix

"CovarianceMatrix" asymptotic parameter covariance matrix
"ParameterConfidenceIntervals" parameter confidence intervals
"ParameterConfidenceIntervalT table of confidence interval information for the fitted
able" parameters
"ParameterConfidenceIntervalT unformatted array of values from the table
ableEntries"
"ParameterConfidenceRegion" ellipsoidal parameter confidence region
"ParameterTableEntries" unformatted array of values from the table
"ParameterErrors" standard errors for parameter estimates
"ParameterPValues" p-values for parameter z-statistics
"ParameterTable" table of fitted parameter information
"ParameterZStatistics" z-statistics for parameter estimates

Properties and diagnostics for parameter estimates.

"CovarianceMatrix" gives the covariance between fitted parameters and is very similar to the
definition for linear models. With CovarianceEstimatorFunction -> "ExpectedInformation"
the expected information matrix obtained from the iterative fitting is used. The matrix is
` -1
f HX W XN where X is the design matrix, and W is the diagonal matrix of weights from the final

stage of the fitting. The weights include both weights specified via the Weights option
and the weights associated with the distribution's variance function. With
CovarianceEstimatorFunction -> "ObservedInformation" the matrix is given by -f I -1
where
I is the observed Fisher information matrix, which is the Hessian of the log-likelihood function
with respect to parameters of the model.
f HX W XN where X is the design matrix, and W is the diagonal matrix of weights from the final

stage of the fitting. The weights include both weights specified via the Weights option
Mathematics and Algorithms 351

CovarianceEstimatorFunction -> "ObservedInformation" the matrix is given by -f I -1 where

I is the observed Fisher information matrix, which is the Hessian of the log-likelihood function
with respect to parameters of the model.

"CorrelationMatrix" is the associated correlation matrix for the parameter estimates.

"ParameterErrors" is equivalent to the square root of the diagonal elements of the covariance
matrix. "ParameterTable" and "ParameterConfidenceIntervalTable" contain information
about the individual parameter estimates, tests of parameter significance, and confidence
intervals. The test statistics for generalized linear models asymptotically follow normal
distributions.

"CookDistances" list of Cook distances

"HatDiagonal" diagonal elements of the hat matrix

Properties related to influence measures.

"CookDistances" and "HatDiagonal" extend the leverage measures from linear regression to
generalized linear models. The hat matrix from which the diagonal elements are extracted is
defined using the final weights of the iterative fitting.

The Cook distance measures of leverage are defined as in linear regression with standardized
residuals replaced by standardized Pearson residuals. The ith Cook distance is given by
Hhii H1 - hii L rspi p where rspi is the ith standardized Pearson residual.

"PredictedResponse" fitted values for the data

Properties of predicted values.

"AdjustedLikelihoodRatioIndex" Ben-Akiva and Lerman's adjusted likelihood ratio index

"AIC" Akaike Information Criterion
"BIC" Bayesian Information Criterion

"CoxSnellPseudoRSquared" Cox and Snell's pseudo R2

"CraggUhlerPseudoRSquared" Cragg and Uhler's pseudo R2

"EfronPseudoRSquared" Efron's pseudo R2

"LikelihoodRatioIndex" McFadden's likelihood ratio index
352 Mathematics and Algorithms

"LikelihoodRatioStatistic" likelihood ratio

"LogLikelihood" log-likelihood for the fitted model

"PearsonChiSquare" Pearson's c2 statistic

Goodness of fit measures.

"LogLikelihood" is the log-likelihood for the fitted model. "AIC" and "BIC" are penalized log-
likelihood measures 2 + k p where is the log-likelihood for the fitted model, p is the number of
parameters estimated including the dispersion parameter, and k is 2 for "AIC" and logHnL for
"BIC" for a model of n data points. "LikelihoodRatioStatistic" is given by 2 H - 0 L where 0
is the log-likelihood for the null model.

A number of the goodness of fit measures generalize R2 from linear regression as either a
measure of explained variation or as a likelihood-based measure. "CoxSnellPseudoRSquared"

is given by 1 - I0 - M . "CraggUhlerPseudoRSquared" is a scaled version of Cox and Snell's

measure J1 - I0 - M N J1 - I0 M N. "LikelihoodRatioIndex" involves the ratio of log-likelihoods

2n 2n

1 - 0 , and "AdjustedLikelihoodRatioIndex" adjusts by penalizing for the number of parame-

ters 1 - H - pL 0 . "EfronPseudoRSquared" uses the sum of squares interpretation of R2 and is

given as 1 - ni=1 ri2 ni=1 Hyi - yL2 where ri is the ith residual and y is the mean of the responses yi .

"PearsonChiSquare" is equal to ni=1 r2pi where the r pi are Pearson residuals.

Nonlinear Models
A nonlinear least-squares model is an extension of the linear model where the model need not
be a linear combination of basis function. The errors are still assumed to be independent and
normally distributed. Models of this type can be fitted using the NonlinearModelFit function.

NonlinearModelFit@ obtain a nonlinear model of the function form with parame -

8y1 ,y2 ,<, form,8 b1 ,<,xD ters bi and a single parameter predictor variable x

NonlinearModelFit@ obtain a nonlinear model as a function of multiple predictor

88x11 ,,y1 <,8x21 ,,y2 <<, variables xi
form,8b1 ,<,8x1 ,<D
NonlinearModelFit@data, obtain a nonlinear model subject to the constraints cons
8 form,cons<,8 b1 ,<,8x1 ,<D

Nonlinear model fitting.

` `
Nonlinear models have the form y = f Ix1 , , xi , b1 , , b j M where y is the fitted or predicted value,

the bi are parameters to be fitted, and the xi are predictor variables. As with any nonlinear
Mathematics and Algorithms 353

` `
Nonlinear models have the form y = f Ix1 , , xi , b1 , , b j M where y is the fitted or predicted value,

the bi are parameters to be fitted, and the xi are predictor variables. As with any nonlinear
optimization problem, a good choice of starting values for the parameters may be necessary.
Starting values can be given using the same parameter specifications as for FindFit.

This fits a nonlinear model to a sequence of square roots.

In[25]:= nlm = NonlinearModelFit@Array@Sqrt, 20D, Log@a + b xD, 8a, b<, xD

Out[25]= FittedModelB [email protected] + 2.76912 xD F

Options for model fitting and for model analysis are available.

option name default value

AccuracyGoal Automatic the accuracy sought
ConfidenceLevel 95100 confidence level to use for parameters and
predictions
EvaluationMonitor None expression to evaluate whenever expr is
evaluated
MaxIterations Automatic maximum number of iterations to use
Method Automatic method to use
PrecisionGoal Automatic the precision sought
StepMonitor None the expression to evaluate whenever a step
is taken
VarianceEstimatorFunction Automatic function for estimating the error variance
Weights Automatic weights for data elements
WorkingPrecision Automatic precision used in internal computations

Options for NonlinearModelFit.

General numeric options such as AccuracyGoal, Method, and WorkingPrecision are the same
as for FindFit.

The Weights option specifies weight values for weighted nonlinear regression. The optimal fit is
for a weighted sum of squared errors.

All other options can be relevant to computation of results after the initial fitting. They can be
set within NonlinearModelFit for use in the fitting and to specify the default settings for
results obtained from the FittedModel object. These options can also be set within an already
constructed FittedModel object to override the option values originally given to
NonlinearModelFit.
354 Mathematics and Algorithms

"BestFit" fitted function

"BestFitParameters" parameter estimates
"Data" the input data
"Function" best fit pure function
"Response" response values in the input data

Properties related to data and the fitted function.

Basic properties of the data and fitted function for nonlinear models behave like the same
properties for linear and generalized linear models with the exception that
"BestFitParameters" returns a rule as is done for the result of FindFit.

This gives the fitted function and rules for the parameter estimates.
In[26]:= nlm@8"BestFit", "BestFitParameters"<D
Out[26]= [email protected] + 2.76912 xD, 8a -0.748315, b 2.76912<<

Many diagnostics for nonlinear models extend or generalize concepts from linear regression.
These extensions often rely on linear approximations or large sample approximations.

"FitResiduals" difference between actual and predicted responses

"StandardizedResiduals" fit residuals divided by the standard error for each residual
"StudentizedResiduals" fit residuals divided by single deletion error estimates

Types of residuals.

As in linear regression, "FitResiduals" gives the differences between the observed and fitted
` `
values 8y1 - y1 , y2 - y2 , <, and "StandardizedResiduals" and "StudentizedResiduals" are
scaled forms of these differences.

` `2 `2
The ith standardized residual is Hyi - yi L s H1 - hii L wi where s is the estimated error

variance, hii is the ith diagonal element of the hat matrix, and wi is the weight for the ith data
`2
point, and the ith studentized residual is obtained by s replacing with the ith single deletion
` 2
variance sHiL . For nonlinear models a first-order approximation is used for the design matrix,
which is needed to compute the hat matrix.
Mathematics and Algorithms 355

"ANOVATable" analysis of variance table

"ANOVATableDegreesOfFreedom" degrees of freedom from the ANOVA table
"ANOVATableEntries" unformatted array of values from the table
"ANOVATableMeanSquares" mean square errors from the table
"ANOVATableSumsOfSquares" sums of squares from the table
"EstimatedVariance" estimate of the error variance

Properties related to the sum of squared errors.

"ANOVATable" provides a decomposition of the variation in the data attributable to the fitted
function and to the errors or residuals.

This gives the ANOVA table for the nonlinear model.

In[27]:= nlm@"ANOVATable"D
DF SS MS
Model 2 208.604 104.302
Out[27]= Error 18 1.39635 0.0775748
Uncorrected Total 20 210.
Corrected Total 19 19.8654

The uncorrected total sums of squares gives the sum of squared responses, while the corrected
total gives the sum of squared differences between the responses and their mean value.

"CorrelationMatrix" asymptotic parameter correlation matrix

"CovarianceMatrix" asymptotic parameter covariance matrix
"ParameterBias" estimated bias in the parameter estimates
"ParameterConfidenceIntervals" parameter confidence intervals
"ParameterConfidenceIntervalT table of confidence interval information for the fitted
able" parameters
"ParameterConfidenceIntervalT unformatted array of values from the table
ableEntries"
"ParameterConfidenceRegion" ellipsoidal parameter confidence region
"ParameterErrors" standard errors for parameter estimates
"ParameterPValues" p-values for parameter t statistics
"ParameterTable" table of fitted parameter information
"ParameterTableEntries" unformatted array of values from the table
"ParameterTStatistics" t statistics for parameter estimates

Properties and diagnostics for parameter estimates.

"CovarianceMatrix" gives the approximate covariance between fitted parameters. The matrix
`2 -1 `2
is s HX W XN where s is the variance estimate, X is the design matrix for the linear approxima-

tion to the model, and W is the diagonal matrix of weights. "CorrelationMatrix" is the associ-
356 Mathematics and Algorithms

"CovarianceMatrix" gives the approximate covariance between fitted parameters. The matrix
`2 -1 `2
is s HX W XN where s is the variance estimate, X is the design matrix for the linear approxima-

tion to the model, and W is the diagonal matrix of weights. "CorrelationMatrix" is the associ-
ated correlation matrix for the parameter estimates. "ParameterErrors" is equivalent to the
square root of the diagonal elements of the covariance matrix.

"ParameterTable" and "ParameterConfidenceIntervalTable" contain information about the

individual parameter estimates, tests of parameter significance, and confidence intervals
obtained using the error estimates.

"CurvatureConfidenceRegion" confidence region for curvature diagnostics

"FitCurvatureTable" table of curvature diagnostics
"FitCurvatureTableEntries" unformatted array of values from the table
"MaxIntrinsicCurvature" measure of maximum intrinsic curvature
"MaxParameterEffectsCurvature" measure of maximum parameter effects curvature

Curvature diagnostics.

The first-order approximation used for many diagnostics is equivalent to the model being linear
in the parameters. If the parameter space near the parameter estimates is sufficiently flat, the
linear approximations and any results that rely on first-order approximations can be deemed
reasonable. Curvature diagnostics are used to assess whether the approximate linearity is
reasonable. "FitCurvatureTable" is a table of curvature diagnostics.

"MaxIntrinsicCurvature" and "MaxParameterEffectsCurvature" are scaled measures of the

normal and tangential curvatures of the parameter spaces at the best-fit parameter values.
"CurvatureConfidenceRegion" is a scaled measure of the radius of curvature of the parameter
space at the best-fit parameter values. If the normal and tangential curvatures are small rela-
tive to the value of the "CurvatureConfidenceRegion", the linear approximation is considered
reasonable. Some rules of thumb suggest comparing the values directly, while others suggest
comparing with half the "CurvatureConfidenceRegion".

Here is the curvature table for the nonlinear model.

In[28]:= nlm@"FitCurvatureTable"D
Curvature
Max Intrinsic 0.109997
Out[28]=
Max Parameter Effects 0.311792
95. % Confidence Region 0.530405
Mathematics and Algorithms 357

"HatDiagonal" diagonal elements of the hat matrix

"SingleDeletionVariances" list of variance estimates with the ith data point omitted

Properties related to influence measures.

` `
The hat matrix is the matrix H such that y = H y where y is the observed response vector and y is
the predicted response vector. "HatDiagonal" gives the diagonal elements of the hat matrix.
As with other properties, H uses the design matrix for the linear approximation to the model.
`
The ith element of "SingleDeletionVariances" is equivalent to IHn - pL s - ri2 H1 - hii LM Hn - p - 1L

where n is the number of data points, p is the number of parameters, hii is the ith hat diagonal,
`
s is the variance estimate for the full dataset, and ri is the ith residual.

"MeanPredictionBands" confidence bands for mean predictions

Properties of predicted values.

Tabular results for confidence intervals are given by

"MeanPredictionConfidenceIntervalTable" and "SinglePredictionConfidenceInterval
Table ". These results are analogous to those for linear models obtained via LinearModelFit,
again with first-order approximations used for the design matrix.

"MeanPredictionBands" and "SinglePredictionBands" give functions of the predictor

variables.
358 Mathematics and Algorithms

"MeanPredictionBands" and "SinglePredictionBands" give functions of the predictor

variables.

Here the fitted function and mean prediction bands are obtained.
In[29]:= 8fit@x_D, mp@x_D< = nlm@8"BestFit", "MeanPredictionBands"<D

Out[29]= :[email protected] + 2.76912 xD,

0.382399 - 0.165343 x + 0.0443921 x2

:-2.10092 + [email protected] + 2.76912 xD,
H0.748315 - 2.76912 xL2

0.382399 - 0.165343 x + 0.0443921 x2

2.10092 + [email protected] + 2.76912 xD>>
H0.748315 - 2.76912 xL2

This plots the fitted curve and confidence bands.

In[30]:= Plot@8fit@xD, mp@xD<, 8x, 1, 20<D
4

Out[30]= 2

10 15 20

"AdjustedRSquared" R2 adjusted for the number of model parameters

"AIC" Akaike Information Criterion
"BIC" Bayesian Information Criterion

"RSquared" coefficient of determination R2

Goodness of fit measures.

"AdjustedRSquared", "AIC", "BIC", and "RSquared" are all direct extensions of the measures
as defined for linear models. The coefficient of determination "RSquared" is the ratio of the
model sum of squares to the total sum of squares. "AdjustedRSquared" penalizes for the
n-1
number of parameters in the model and is given by 1 - H n-p L H1 - R2 L.

"AIC" and "BIC" are equal to -2 times the log-likelihood for the model plus k p where p is the
number of parameters to be estimated including the estimated variance. For "AIC" k is 2, and
for "BIC" k is logHnL.

Approximate Functions and Interpolation

Mathematics and Algorithms 359

Approximate Functions and Interpolation

In many kinds of numerical computations, it is convenient to introduce approximate functions.
Approximate functions can be thought of as generalizations of ordinary approximate real num -
bers. While an approximate real number gives the value to a certain precision of a single numeri -
cal quantity, an approximate function gives the value to a certain precision of a quantity which
depends on one or more parameters. Mathematica uses approximate functions, for example, to
represent numerical solutions to differential equations obtained with NDSolve, as discussed in
"Numerical Differential Equations".

Approximate functions in Mathematica are represented by InterpolatingFunction objects.

These objects work like the pure functions discussed in "Pure Functions". The basic idea is that
when given a particular argument, an InterpolatingFunction object finds the approximate
function value that corresponds to that argument.

The InterpolatingFunction object contains a representation of the approximate function

based on interpolation. Typically it contains values and possibly derivatives at a sequence of
points. It effectively assumes that the function varies smoothly between these points. As a
result, when you ask for the value of the function with a particular argument, the
InterpolatingFunction object can interpolate to find an approximation to the value you want.

Interpolation@8 f1 , f2 ,<D construct an approximate function with values fi at succes -

sive integers
Interpolation@88x1 , f1 <,8x2 , f2 <,<D
construct an approximate function with values fi at points xi

Constructing approximate functions.

Here is a table of the values of the sine function.

In[1]:= Table@8x, Sin@xD<, 8x, 0, 2, 0.25<D
Out[1]= 880., 0.<, 80.25, 0.247404<, 80.5, 0.479426<, 80.75, 0.681639<,
81., 0.841471<, 81.25, 0.948985<, 81.5, 0.997495<, 81.75, 0.983986<, 82., 0.909297<<

This constructs an approximate function which represents these values.

In[2]:= sin = Interpolation@%D
Out[2]= InterpolatingFunction@880., 2.<<, <>D

The approximate function reproduces each of the values in the original table.
360 Mathematics and Algorithms

The approximate function reproduces each of the values in the original table.
In[3]:= [email protected]
Out[3]= 0.247404

It also allows you to get approximate values at other points.

In[4]:= [email protected]
Out[4]= 0.2955

In this case the interpolation is a fairly good approximation to the true sine function.
In[5]:= [email protected]
Out[5]= 0.29552

You can work with approximate functions much as you would with any other Mathematica
functions. You can plot approximate functions, or perform numerical operations such as integra-
tion or root finding.

If you give a non-numerical argument, the approximate function is left in symbolic form.
In[6]:= sin@xD
Out[6]= InterpolatingFunction@880., 2.<<, <>D@xD

Here is a numerical integral of the approximate function.

In[7]:= NIntegrate@sin@xD ^ 2, 8x, 0, Pi 2<D
Out[7]= 0.78531

Here is the same numerical integral for the true sine function.
In[8]:= NIntegrate@Sin@xD ^ 2, 8x, 0, Pi 2<D
Out[8]= 0.785398

A plot of the approximate function is essentially indistinguishable from the true sine function.
In[9]:= Plot@sin@xD, 8x, 0, 2<D
1.0

0.8

0.6
Out[9]=
0.4

0.2

0.5 1.0 1.5 2.0

If you differentiate an approximate function, Mathematica will return another approximate

function that represents the derivative.
Mathematics and Algorithms 361

If you differentiate an approximate function, Mathematica will return another approximate

function that represents the derivative.

This finds the derivative of the approximate sine function, and evaluates it at p 6.
In[10]:= sin '@Pi 6D
Out[10]= 0.865372

The result is close to the exact one.

In[11]:= N@Cos@Pi 6DD
Out[11]= 0.866025

InterpolatingFunction objects contain all the information Mathematica needs about approxi-
mate functions. In standard Mathematica output format, however, only the part that gives the
domain of the InterpolatingFunction object is printed explicitly. The lists of actual parame-
ters used in the InterpolatingFunction object are shown only in iconic form.

In standard output format, the only part of an InterpolatingFunction object printed

explicitly is its domain.
In[12]:= sin
Out[12]= InterpolatingFunction@880., 2.<<, <>D

If you ask for a value outside of the domain, Mathematica prints a warning, then uses extrapola-
tion to find a result.
In[13]:= sin@3D

InterpolatingFunction::dmval :
Input value 83< lies outside the range of data in the interpolating function. Extrapolation will be used.
Out[13]= 0.0155471

The more information you give about the function you are trying to approximate, the better the
approximation Mathematica constructs can be. You can, for example, specify not only values of
the function at a sequence of points, but also derivatives.

Interpolation@888x1 <, f1 ,df1 ,ddf1 ,<,<D

construct an approximate function with specified deriva-
tives at points xi

Constructing approximate functions with specified derivatives.

362 Mathematics and Algorithms

This interpolates through the values of the sine function and its first derivative.
In[14]:= sind = Interpolation@Table@88x<, Sin@xD, Cos@xD<, 8x, 0, 2, 0.25<DD
Out[14]= InterpolatingFunction@880., 2.<<, <>D

This finds a better approximation to the derivative than the previous interpolation.
In[15]:= sind '@Pi 6D
Out[15]= 0.865974

Interpolation works by fitting polynomial curves between the points you specify. You can use
the option InterpolationOrder to specify the degree of these polynomial curves. The default
setting is InterpolationOrder -> 3, yielding cubic curves.

This makes a table of values of the cosine function.

In[16]:= tab = Table@8x, Cos@xD<, 8x, 0, 6<D;

This creates an approximate function using linear interpolation between the values in the table.
In[17]:= Interpolation@tab, InterpolationOrder -> 1D
Out[17]= InterpolatingFunction@880, 6<<, <>D

The approximate function consists of a collection of straight-line segments.

In[18]:= Plot@%@xD, 8x, 0, 6<D
1.0

0.5

Out[18]=
1 2 3 4 5 6
-0.5

-1.0

With the default setting InterpolationOrder -> 3, cubic curves are used, and the function
looks smooth.
In[19]:= Plot@Evaluate@Interpolation@tabDD@xD, 8x, 0, 6<D
1.0

0.5

Out[19]=
1 2 3 4 5 6
-0.5

-1.0

Increasing the setting for InterpolationOrder typically leads to smoother approximate func-
tions. However, if you increase the setting too much, spurious wiggles may develop.
Mathematics and Algorithms 363

ListInterpolation@88 f11 , f12 ,<,8 f21 ,<,<D

construct an approximate function from a two-dimensional
grid of values at integer points
ListInterpolation@list,88xmin ,xmax <,8ymin ,ymax <<D
assume the values are from an evenly spaced grid with the
specified domain
ListInterpolation@list,88x1 ,x2 ,<,8y1 ,y2 ,<<D
assume the values are from a grid with the specified grid
lines

Interpolating multidimensional arrays of data.

This interpolates an array of values from integer grid points.

In[20]:= ListInterpolation@[email protected] Hx ^ 2 + y ^ 3L, 8x, 10<, 8y, 15<DD
Out[20]= InterpolatingFunction@881., 10.<, 81., 15.<<, <>D

Here is the value at a particular position.

In[21]:= %@6.5, 7.2D
Out[21]= 0.00360759

Here is another array of values.

In[22]:= tab = [email protected] Hx ^ 2 + y ^ 3L, 8x, 5.5, 7.2, .2<, 8y, 2.3, 8.9, .1<D;

To interpolate this array you explicitly have to tell Mathematica the domain it covers.
In[23]:= ListInterpolation@tab, 885.5, 7.2<, 82.3, 8.9<<D
Out[23]= [email protected], 7.2<, 82.3, 8.9<<, <>D

ListInterpolation works for arrays of any dimension, and in each case it produces an
InterpolatingFunction object which takes the appropriate number of arguments.

This interpolates a three-dimensional array.

In[24]:= ListInterpolation@Array@1 ^ 2 + 2 ^ 2 - 3 ^ 2 &, 810, 10, 10<DD;

The resulting InterpolatingFunction object takes three arguments.

In[25]:= %@3.4, 7.8, 2.6D
Out[25]= 65.64

Mathematica can handle not only purely numerical approximate functions, but also ones which
involve symbolic parameters.
364 Mathematics and Algorithms

Mathematica can handle not only purely numerical approximate functions, but also ones which
involve symbolic parameters.

This generates an InterpolatingFunction that depends on the parameters a and b.

In[26]:= sfun = ListInterpolation@81 + a, 2, 3, 4 + b, 5<D
Out[26]= InterpolatingFunction@881, 5<<, <>D

This shows how the interpolated value at 2.2 depends on the parameters.
In[27]:= [email protected] Simplify
Out[27]= 2.2 - 0.048 a - 0.032 b

With the default setting for InterpolationOrder used, the value at this point no longer
depends on a.
In[28]:= [email protected] Simplify
Out[28]= 3.8 + 0.864 b

In working with approximate functions, you can quite often end up with complicated combina-
tions of InterpolatingFunction objects. You can always tell Mathematica to produce a single
InterpolatingFunction object valid over a particular domain by using
FunctionInterpolation.

This generates a new InterpolatingFunction object valid in the domain 0 to 1.

In[29]:= FunctionInterpolation@x + sin@x ^ 2D, 8x, 0, 1<D
Out[29]= InterpolatingFunction@880., 1.<<, <>D

This generates a nested InterpolatingFunction object.

In[30]:= ListInterpolation@83, 4, 5, sin@aD, 6<D
Out[30]= InterpolatingFunction@881, 5<<, <>D

This produces a pure two-dimensional InterpolatingFunction object.

In[31]:= FunctionInterpolation@a ^ 2 + %@xD, 8x, 1, 3<, 8a, 0, 1.5<D
Out[31]= InterpolatingFunction@881., 3.<, 80., 1.5<<, <>D
Mathematics and Algorithms 365

FunctionInterpolation@expr,8x,xmin ,xmax <D

construct an approximate function by evaluating expr with
x ranging from xmin to xmax
FunctionInterpolation@expr,8x,xmin ,xmax <,8y,ymin ,ymax <,D
construct a higher-dimensional approximate function

Constructing approximate functions by evaluating expressions.

Discrete Fourier Transforms

A common operation in analyzing various kinds of data is to find the discrete Fourier transform
(or spectrum) of a list of values. The idea is typically to pick out components of the data with
particular frequencies or ranges of frequencies.

Fourier@8u1 ,u2 ,,un <D discrete Fourier transform

InverseFourier@8v1 ,v2 ,,vn <D inverse discrete Fourier transform

Discrete Fourier transforms.

Here is some data, corresponding to a square pulse.

In[1]:= 8- 1, - 1, - 1, - 1, 1, 1, 1, 1<
Out[1]= 8-1, -1, -1, -1, 1, 1, 1, 1<

Here is the discrete Fourier transform of the data. It involves complex numbers.
In[2]:= Fourier@%D
Out[2]= 80. + 0. , -0.707107 - 1.70711 , 0. + 0. , -0.707107 - 0.292893 ,
0. + 0. , -0.707107 + 0.292893 , 0. + 0. , -0.707107 + 1.70711 <

Here is the inverse discrete Fourier transform.

In[3]:= InverseFourier@%D
Out[3]= 8-1., -1., -1., -1., 1., 1., 1., 1.<

Fourier works whether or not your list of data has a length which is a power of two.
In[4]:= Fourier@81, - 1, 1<D
Out[4]= 80.57735 + 0. , 0.57735 - 1. , 0.57735 + 1. <
366 Mathematics and Algorithms

This generates a list of 200 elements containing a periodic signal with random noise added.
In[5]:= data = Table@N@Sin@30 2 Pi n 200D + HRandomReal@D - 1 2LD, 8n, 200<D;

The data looks fairly random if you plot it directly.

In[6]:= ListLinePlot@dataD
1.5

1.0

0.5

Out[6]=
50 100 150 200

-0.5

-1.0

-1.5

The discrete Fourier transform, however, shows a strong peak at 30 + 1, and a symmetric peak
at 201 - 30, reflecting the frequency component of the original signal near 30 200.
In[7]:= ListLinePlot@Abs@Fourier@dataDD, PlotRange -> AllD
7

4
Out[7]=
3

50 100 150 200

In Mathematica, the discrete Fourier transform vs of a list ur of length n is by default defined to

1
be r=1 ur e2 p i Hr-1L Hs-1Ln . Notice that the zero frequency term appears at position 1 in the result-
n
n

ing list.

The inverse discrete Fourier transform ur of a list vs of length n is by default defined to be

1
s=1 vs e-2 p i Hr-1L Hs-1Ln .
n
n

In different scientific and technical fields different conventions are often used for defining dis-
crete Fourier transforms. The option FourierParameters allows you to choose any of these
conventions you want.
Mathematics and Algorithms 367

common setting discrete Fourier transform inverse discrete Fourier transform

convention
1 n 1 n
Mathematica 80,1< r=1 ur e2 p i Hr-1L Hs-1Ln s=1 vs e-2 p i Hr-1L Hs-1Ln
n12 n12
default
1 n n
data analysis 8-1,1< n
r=1 ur e2 p i Hr-1L Hs-1Ln s=1 vs e-2 p iHr-1L Hs-1Ln
n 1 n
signal processing 81,-1< r=1 ur e-2 p iHr-1L Hs-1Ln n
s=1 vs e2 p i Hr-1L Hs-1Ln
1 n 1 n
general case 8a,b< r=1 ur e2 p i bHr-1L Hs-1Ln s=1 vs e-2 p i b Hr-1L Hs-1Ln
nH1-aL2 nH1+aL2

Typical settings for FourierParameters with various conventions.

Fourier@88u11 ,u12 ,<,8u21 ,u22 ,<,<D

two-dimensional discrete Fourier transform

Two-dimensional discrete Fourier transform.

Mathematica can find discrete Fourier transforms for data in any number of dimensions. In n
dimensions, the data is specified by a list nested n levels deep. Two-dimensional discrete
Fourier transforms are often used in image processing.

One issue with the usual discrete Fourier transform for real data is that the result is complex-
valued. There are variants of real discrete Fourier transforms that have real results. Mathemat-
ica has commands for computing the discrete cosine transform and the discrete sine transform.

FourierDCT@listD Fourier discrete cosine transform of a list of real numbers

FourierDST@listD Fourier discrete sine transform of a list of real numbers

Discrete real Fourier transforms.

Here is some data, corresponding to a square pulse.

In[8]:= pulse = 8- 1, - 1, - 1, - 1, - 1, 1, 1, 1, 1, 1<
Out[8]= 8-1, -1, -1, -1, -1, 1, 1, 1, 1, 1<

Here is the Fourier discrete cosine transform of the data.

In[9]:= FourierDCT@pulseD
-17 -17
Out[9]= 90., -2.02147, 3.29753 10 , 0.696552, -5.68065 10 ,
-0.447214, -4.12723 10-17 , 0.354911, 4.76294 10-17 , -0.32017=
368 Mathematics and Algorithms

Here is the Fourier discrete sine transform of the data.

In[10]:= FourierDST@pulseD
-17 -17 -18
Out[10]= 98.03365 10 , -2.04667, -3.85062 10 , -9.37835 10 ,
0., -0.781758, 9.84259 10-17 , 4.4886 10-17 , 5.83679 10-17 , -0.632456=

There are four types each of Fourier discrete sine and cosine transforms typically in use,
denoted by number or sometimes roman numeral as in "DCTII" for the discrete cosine trans-
form of type 2.

FourierDCT@list,mD Fourier discrete cosine transform of type m

FourierDST@list,mD Fourier discrete sine transform of type m

Discrete real Fourier transforms of different types.

The default is type 2 for both FourierDCT and FourierDST.

Mathematica does not need InverseFourierDCT or InverseFourierDST functions because

FourierDCT and FourierDST are their own inverses when used with the appropriate type. The
inverse transforms for types 1, 2, 3, 4 are types 1, 3, 2, 4, respectively.

Check that the type 3 transform is the inverse of the type 2 transform.
In[11]:= FourierDCT@FourierDCT@pulse, 2D, 3D
Out[11]= 8-1., -1., -1., -1., -1., 1., 1., 1., 1., 1.<

The discrete real transforms are convenient to use for data or image compression.

Here is data that might be like a front or an edge.

In[12]:= data = Table@ArcTan@x - 100D, 8x, 1., 200.<D;
ListLinePlot@dataD
1.5
1.0
0.5
Out[12]=
50 100 150 200
-0.5
-1.0
-1.5
Mathematics and Algorithms 369

The discrete cosine transform has most of the information in the first few modes.
In[13]:= dct = FourierDCT@dataD;
ListLinePlot@dct, PlotRange AllD
5

50 100 150 200

Out[13]=
-5

-10

Reconstruct the front from only the first 20 modes (1/10 of the original data size). The oscilla-
tions are a consequence of the truncation and are known to show up in image processing
applications as well.
In[14]:= tdata = FourierDCT@PadRight@Take@dct, 20D, 200, 0D, 3D;
ListLinePlot@8data, tdata<D
1.5
1.0
Out[14]= 0.5

50 100 150 200

-0.5
-1.0
-1.5

Convolutions and Correlations

Convolution and correlation are central to many kinds of operations on lists of data. They are
used in such areas as signal and image processing, statistical data analysis, approximations to
partial differential equations, as well as operations on digit sequences and power series.

In both convolution and correlation the basic idea is to combine a kernel list with successive
sublists of a list of data. The convolution of a kernel Kr with a list us has the general form
r Kr us-r , while the correlation has the general form r Kr us+r .

ListConvolve@kernel,listD form the convolution of kernel with list

ListCorrelate@kernel,listD form the correlation of kernel with list

Convolution and correlation of lists.

This forms the convolution of the kernel 8x, y< with a list of data.
In[1]:= ListConvolve@8x, y<, 8a, b, c, d, e<D
Out[1]= 8b x + a y, c x + b y, d x + c y, e x + d y<
370 Mathematics and Algorithms

This forms the correlation.

In[2]:= ListCorrelate@8x, y<, 8a, b, c, d, e<D
Out[2]= 8a x + b y, b x + c y, c x + d y, d x + e y<

In this case reversing the kernel gives exactly the same result as ListConvolve .
In[3]:= ListCorrelate@8y, x<, 8a, b, c, d, e<D
Out[3]= 8b x + a y, c x + b y, d x + c y, e x + d y<

This forms successive differences of the data.

In[4]:= ListCorrelate@8- 1, 1<, 8a, b, c, d, e<D
Out[4]= 8-a + b, -b + c, -c + d, -d + e<

In forming sublists to combine with a kernel, there is always an issue of what to do at the ends
of the list of data. By default, ListConvolve and ListCorrelate never form sublists which
would "overhang" the ends of the list of data. This means that the output you get is normally
shorter than the original list of data.

With an input list of length 6, the output is in this case of length 4.

In[5]:= ListCorrelate@81, 1, 1<, Range@6DD
Out[5]= 86, 9, 12, 15<

In practice one often wants to get output that is as long as the original list of data. To do this
requires including sublists that overhang one or both ends of the list of data. The additional
elements needed to form these sublists must be filled in with some kind of "padding". By
default, Mathematica takes copies of the original list to provide the padding, thus effectively
treating the list as being cyclic.

ListCorrelate@kernel,listD do not allow overhangs on either side (result shorter

than list)
ListCorrelate@kernel,list,1D allow an overhang on the right (result same length as list)
ListCorrelate@kernel,list,-1D allow an overhang on the left (result same length as list)
ListCorrelate@kernel,list,8-1,1<D allow overhangs on both sides (result longer than list)
ListCorrelate@kernel,list,8kL ,kR <D allow particular overhangs on left and right

Controlling how the ends of the list of data are treated.

Mathematics and Algorithms 371

The default involves no overhangs.

In[6]:= ListCorrelate@8x, y<, 8a, b, c, d<D
Out[6]= 8a x + b y, b x + c y, c x + d y<

The last term in the last element now comes from the beginning of the list.
In[7]:= ListCorrelate@8x, y<, 8a, b, c, d<, 1D
Out[7]= 8a x + b y, b x + c y, c x + d y, d x + a y<

Now the first term of the first element and the last term of the last element both involve
wraparound.
In[8]:= ListCorrelate@8x, y<, 8a, b, c, d<, 8- 1, 1<D
Out[8]= 8d x + a y, a x + b y, b x + c y, c x + d y, d x + a y<

In the general case ListCorrelate@kernel, list, 8kL , kR <D is set up so that in the first element of
the result, the first element of list appears multiplied by the element at position kL in kernel, and
in the last element of the result, the last element of list appears multiplied by the element at
position kR in kernel. The default case in which no overhang is allowed on either side thus corre-
sponds to ListCorrelate@kernel, list, 81, - 1<D.

With a kernel of length 3, alignments 8- 1, 2< always make the first and last elements of the
result the same.
In[9]:= ListCorrelate@8x, y, z<, 8a, b, c, d<, 8- 1, 2<D
Out[9]= 8c x + d y + a z, d x + a y + b z, a x + b y + c z, b x + c y + d z, c x + d y + a z<

For many kinds of data, it is convenient to assume not that the data is cyclic, but rather that it
is padded at either end by some fixed element, often 0, or by some sequence of elements.

ListCorrelate@kernel,list,klist,pD pad with element p

ListCorrelate@kernel,list,klist,8p1 ,p2 ,<D
pad with cyclic repetitions of the pi

ListCorrelate@kernel,list,klist,listD pad with cyclic repetitions of the original data

Controlling the padding for a list of data.

This pads with element p.

In[10]:= ListCorrelate@8x, y<, 8a, b, c, d<, 8- 1, 1<, pD
Out[10]= 8p x + a y, a x + b y, b x + c y, c x + d y, d x + p y<

A common case is to pad with zero.

372 Mathematics and Algorithms

A common case is to pad with zero.

In[11]:= ListCorrelate@8x, y<, 8a, b, c, d<, 8- 1, 1<, 0D
Out[11]= 8a y, a x + b y, b x + c y, c x + d y, d x<

When the padding is indicated by 8p, q<, the list 8a, b, c< overlays 8, p, q, p, q, < with
a p aligned under the a.
In[12]:= ListCorrelate@8x, y, z<, 8a, b, c<, 8- 1, 1<, 8p, q<D
Out[12]= 8p x + q y + a z, q x + a y + b z, a x + b y + c z, b x + c y + q z, c x + q y + p z<

Different choices of kernel allow ListConvolve and ListCorrelate to be used for different
kinds of computations.

This finds a moving average of data.

In[13]:= ListCorrelate@81, 1, 1< 3, 8a, b, c, d, e<D
a b c b c d c d e
Out[13]= : + + , + + , + + >
3 3 3 3 3 3 3 3 3

Here is a Gaussian kernel.

In[14]:= kern = Table@Exp@- n ^ 2 100D Sqrt@2. PiD, 8n, - 10, 10<D;

This generates some "data".

In[15]:= data = Table@BesselJ@1, xD + 0.2 RandomReal@D, 8x, 0, 10, .1<D;

Here is a plot of the data.

In[16]:= ListPlot@dataD

0.6
0.4
Out[16]=
0.2

20 40 60 80 100
-0.2

This convolves the kernel with the data.

In[17]:= ListConvolve@kern, dataD;
Mathematics and Algorithms 373

The result is a smoothed version of the data.

In[18]:= ListPlot@%D

2
Out[18]= 1

20 40 60 80
-1

You can use ListConvolve and ListCorrelate to handle symbolic as well as numerical data.

This forms the convolution of two symbolic lists.

In[19]:= ListConvolve@8a, b, c<, 8u, v, w<, 81, - 1<, 0D
Out[19]= 8a u, b u + a v, c u + b v + a w, c v + b w, c w<

The result corresponds exactly with the coefficients in the expanded form of this product of
polynomials.
In[20]:= Expand@Ha + b x + c x ^ 2L Hu + v x + w x ^ 2LD

Out[20]= a u + b u x + a v x + c u x 2 + b v x 2 + a w x2 + c v x 3 + b w x3 + c w x4

ListConvolve and ListCorrelate work on data in any number of dimensions.

This imports image data from a file.

In[21]:= g = ReadList@"ExampleDatafish.data", Number, RecordLists -> TrueD;

Here is the image.

In[22]:= Graphics@Raster@gDD

Out[22]=
374 Mathematics and Algorithms

This convolves the data with a two-dimensional kernel.

In[23]:= ListConvolve@881, 1, 1<, 81, - 8, 1<, 81, 1, 1<<, gD;

This shows the image corresponding to the data.

In[24]:= Graphics@Raster@%DD

Out[24]=

Cellular Automata
Cellular automata provide a convenient way to represent many kinds of systems in which the
values of cells in an array are updated in discrete steps according to a local rule.

CellularAutomaton@rnum,init,tD evolve rule rnum from init for t steps

Generating a cellular automaton evolution.

This starts with the list given, then evolves rule 30 for 4 steps.
In[1]:= CellularAutomaton@30, 80, 0, 0, 1, 0, 0, 0<, 4D
Out[1]= 880, 0, 0, 1, 0, 0, 0<, 80, 0, 1, 1, 1, 0, 0<,
80, 1, 1, 0, 0, 1, 0<, 81, 1, 0, 1, 1, 1, 1<, 80, 0, 0, 1, 0, 0, 0<<

This shows 100 steps of rule 30 evolution from random initial conditions.
In[2]:= ArrayPlot@CellularAutomaton@30, RandomInteger@1, 250D, 100DD

Out[2]=
Mathematics and Algorithms 375

8a1 ,a2 ,< explicit list of values ai

88a1 ,a2 ,<,b< values ai superimposed on a b background

88a1 ,a2 ,<,blist< values ai superimposed on a background of repetitions of

blist
8888a11 ,a12 ,<,8d1 <<,<,blist< values aij at offsets di

Ways of specifying initial conditions for one-dimensional cellular automata.

If you give an explicit list of initial values, CellularAutomaton will take the elements in this list
to correspond to all the cells in the system, arranged cyclically.

The right neighbor of the cell at the end is the cell at the beginning.
In[4]:= CellularAutomaton@30, 81, 0, 0, 0, 0<, 1D
Out[4]= 881, 0, 0, 0, 0<, 81, 1, 0, 0, 1<<

It is often convenient to set up initial conditions in which there is a small "seed" region, superim-
posed on a constant "background". By default, CellularAutomaton automatically fills in enough
background to cover the size of the pattern that can be produced in the number of steps of
evolution you specify.

This shows rule 30 evolving from an initial condition containing a single black cell.
In[5]:= ArrayPlot@CellularAutomaton@30, 881<, 0<, 100DD

Out[5]=
376 Mathematics and Algorithms

This shows rule 30 evolving from an initial condition consisting of a 81, 1< seed on a back-
ground of repeated 81, 0, 1, 1< blocks.
In[6]:= ArrayPlot@CellularAutomaton@30, 881, 1<, 81, 0, 1, 1<<, 100DD

Out[6]=

Particularly in studying interactions between structures, you may sometimes want to specify
initial conditions for cellular automata in which certain blocks are placed at particular offsets.

This sets up an initial condition with black cells at offsets 40.

In[7]:= ArrayPlot@CellularAutomaton@30, 88881<, 8- 40<<, 881<, 840<<<, 0<, 100DD

Out[7]=

n k = 2, r = 1, elementary rule
8n,k< general nearest-neighbor rule with k colors
8n,k,r< general rule with k colors and range r
8n,8k,1<< k-color nearest-neighbor totalistic rule
8n,8k,1<,r< k-color range r totalistic rule
8n,8k,8wt1 ,wt2 ,<<,r< rule in which neighbor i is assigned weight wti

9n,kspec,98off1 <,8off2 <,,9offs === rule with neighbors at specified offsets

8lhs1 ->rhs1 ,lhs2 ->rhs2 ,< explicit replacements for lists of neighbors
8 fun,8<,rspec< rule obtained by applying function fun to each neighbor list

Specifying rules for one-dimensional cellular automata.

In the simplest cases, a cellular automaton allows k possible values or "colors" for each cell, and
has rules that involve up to r neighbors on each side. The digits of the "rule number" n then
specify what the color of a new cell should be for each possible configuration of the
neighborhood.
Mathematics and Algorithms 377

This evolves a single neighborhood for 1 step.

In[8]:= CellularAutomaton@30, 81, 1, 0<, 1D
Out[8]= 881, 1, 0<, 81, 0, 0<<

Here are the 8 possible neighborhoods for a k = 2, r = 1 cellular automaton.

In[9]:= Tuples@81, 0<, 3D
Out[9]= 881, 1, 1<, 81, 1, 0<, 81, 0, 1<, 81, 0, 0<, 80, 1, 1<, 80, 1, 0<, 80, 0, 1<, 80, 0, 0<<

This shows the new color of the center cell for each of the 8 neighborhoods.
In[10]:= Map@CellularAutomaton@30, , 1D@@2, 2DD &, %D
Out[10]= 80, 0, 0, 1, 1, 1, 1, 0<

For rule 30, this sequence corresponds to the base-2 digits of the number 30.
In[11]:= FromDigits@%, 2D
Out[11]= 30

This runs the general k = 3, r = 1 rule with rule number 921408.

In[12]:= ArrayPlot@CellularAutomaton@8921 408, 3, 1<, 881<, 0<, 100DD

Out[12]=

For a general cellular automaton rule, each digit of the rule number specifies what color a
different possible neighborhood of 2 r + 1 cells should yield. To find out which digit corresponds

r = 1 cellular automaton, the number is obtained from the list of elements neig in the
neighborhood by neig.8k ^ 2, k, 1<.
378 Mathematics and Algorithms

to which neighborhood, one effectively treats the cells in a neighborhood as digits in a number.
For an r = 1 cellular automaton, the number is obtained from the list of elements neig in the
neighborhood by neig.8k ^ 2, k, 1<.

It is sometimes convenient to consider totalistic cellular automata, in which the new value of a
cell depends only on the total of the values in its neighborhood. One can specify totalistic cellu-
lar automata by rule numbers or "codes" in which each digit refers to neighborhoods with a
given total value, obtained for example from neig.81, 1, 1<.

In general, CellularAutomaton allows one to specify rules using any sequence of weights.
Another choice sometimes convenient is 8k, 1, k<, which yields outer totalistic rules.

This runs the k = 3, r = 1 totalistic rule with code number 867.

In[13]:= ArrayPlot@CellularAutomaton@8867, 83, 1<, 1<, 881<, 0<, 100DD

Out[13]=

Rules with range r involve all cells with offsets -r through +r. Sometimes it is convenient to
think about rules that involve only cells with specific offsets. You can do this by replacing a
single r with a list of offsets.

Any k = 2 cellular automaton rule can be thought of as corresponding to a Boolean function. In

the simplest case, basic Boolean functions like And or Nor take two arguments. These are
conveniently specified in a cellular automaton rule as being at offsets 880<, 81<<. Note that for
compatibility with handling higher-dimensional cellular automata, offsets must always be given
in lists, even for one-dimensional cellular automata.

This generates the truth table for 2-cell-neighborhood rule number 7, which turns out to be the
Boolean function Nand.
In[14]:= Map@CellularAutomaton@87, 2, 880<, 81<<<, , 1D@@2, 2DD &,
881, 1<, 81, 0<, 80, 1<, 80, 0<<D
Out[14]= 80, 1, 1, 1<

Rule numbers provide a highly compact way to specify cellular automaton rules. But sometimes
it is more convenient to specify rules by giving an explicit function that should be applied to
each possible neighborhood.
Mathematics and Algorithms 379

This runs an additive cellular automaton whose rule adds all values in each neighborhood
modulo 4.
In[15]:= ArrayPlot@CellularAutomaton@8Mod@Apply@Plus, D, 4D &, 8<, 1<, 881<, 0<, 100DD

Out[15]=

The function is given the step number as a second argument.

In[16]:= ArrayPlot@CellularAutomaton@8Mod@Total@D + 2, 4D &, 8<, 1<, 881<, 0<, 100DD

Out[16]=

When you specify rules by functions, the values of cells need not be integers.
In[17]:= ArrayPlot@CellularAutomaton@8Mod@1 2 Apply@Plus, D, 1D &, 8<, 1<, 881<, 0<, 100DD

Out[17]=

They can even be symbolic.

In[18]:= Simplify@
CellularAutomaton@8Mod@Apply@Plus, D, 2D &, 8<, 1<, 88a<, 0<, 2D, a IntegersD
Out[18]= 880, 0, a, 0, 0<, 80, Mod@a, 2D, Mod@a, 2D, Mod@a, 2D, 0<, 8Mod@a, 2D, 0, Mod@a, 2D, 0, Mod@a, 2D<<
380 Mathematics and Algorithms

CellularAutomaton@rnum,init,tD evolve for t steps, keeping all steps

CellularAutomaton@ evolve for t steps, keeping only the last step
rnum,init,88t<<D
CellularAutomatonA keep only steps specified by spect
rnum,init,9spect =E
CellularAutomaton@rnum,initD evolve rule for one step, giving only the last step

Selecting which steps to keep.

This runs rule 30 for 5 steps, keeping only the last step.
In[19]:= CellularAutomaton@30, 881<, 0<, 885<<D
Out[19]= 881, 1, 0, 1, 1, 1, 1, 0, 1, 1, 1<<

This keeps the last 2 steps.

In[20]:= CellularAutomaton@30, 881<, 0<, 884, 5<<D
Out[20]= 880, 1, 1, 0, 0, 1, 0, 0, 0, 1, 0<, 81, 1, 0, 1, 1, 1, 1, 0, 1, 1, 1<<

This gives 1 step.

In[21]:= CellularAutomaton@30, 80, 0, 1, 0, 0<D
Out[21]= 80, 1, 1, 1, 0<

The step specification spect works very much like taking elements from a list with Take. One

difference, though, is that the initial condition for the cellular automaton is considered to be
step 0. Note that any step specification of the form 8< must be enclosed in an additional list.

u steps 0 through u
8u< step u
8u1 ,u2 < steps u1 through u2
8u1 ,u2 ,du< steps u1 , u1 + du,

Cellular automaton step specifications.

This evolves for 100 steps, but keeps only every other step.
In[22]:= ArrayPlot@CellularAutomaton@30, 881<, 0<, 880, 100, 2<<DD

Out[22]=
Mathematics and Algorithms 381

CellularAutomaton@rnum,init,tD keep all steps, and all relevant cells

CellularAutomatonArnum,init,9spect ,specx =E
keep only specified steps and cells

Selecting steps and cells to keep.

Much as you can specify which steps to keep in a cellular automaton evolution, so also you can
specify which cells to keep. If you give an initial condition such as 8a1 , a2 , <, blist, then ai is
taken to have offset 0 for the purpose of specifying which cells to keep.

All all cells that can be affected by the specified initial condition

Automatic all cells in the region that differs from the background
(default)
0 cell aligned with beginning of aspec
x cells at offsets up to x on the right
-x cells at offsets up to x on the left
8x< cell at offset x to the right
8-x< cell at offset x to the left
8x1 ,x2 < cells at offsets x1 through x2
8x1 ,x2 ,dx< cells x1 , x1 + dx,

Cellular automaton cell specifications.

This keeps all steps, but drops cells at offsets more than 20 on the left.
In[23]:= ArrayPlot@CellularAutomaton@30, 881<, 0<, 8100, 8- 20, 100<<DD

Out[23]=
382 Mathematics and Algorithms

This keeps just the center column of cells.

In[24]:= CellularAutomaton@30, 881<, 0<, 820, 80<<D
Out[24]= 881<, 81<, 80<, 81<, 81<, 81<, 80<, 80<, 81<,
81<, 80<, 80<, 80<, 81<, 80<, 81<, 81<, 80<, 80<, 81<, 80<<

If you give an initial condition such as 88a1 , a2 , <, blist<, then CellularAutomaton will always
effectively do the cellular automaton as if there were an infinite number of cells. By using a

specx such as 8x1 , x2 < you can tell CellularAutomaton to include only cells at specific offsets x1

through x2 in its output. CellularAutomaton by default includes cells out just far enough that
their values never simply stay the same as in the background blist.

In general, given a cellular automaton rule with range r, cells out to distance rt on each side
could in principle be affected in the evolution of the system. With specx being All, all these cells

are included; with the default setting of Automatic, cells whose values effectively stay the
same as in blist are trimmed off.

By default, only the parts that are not constant black are kept.
In[25]:= ArrayPlot@CellularAutomaton@225, 881<, 0<, 100DD

Out[25]=
Mathematics and Algorithms 383

Using All for specx includes all cells that could be affected by a cellular automaton with this
range.
In[26]:= ArrayPlot@CellularAutomaton@225, 881<, 0<, 8100, All<DD

Out[26]=

CellularAutomaton generalizes quite directly to any number of dimensions. Above two dimen-
sions, however, totalistic and other special types of rules tend to be more useful, since the
number of entries in the rule table for a general rule rapidly becomes astronomical.

8n,k,8r1 ,r2 ,,rd << d -dimensional rule with H2 r1 + 1L H2 r2 + 1L H2 rd + 1L

neighborhood
8n,8k,1<,81,1<< two-dimensional 9-neighbor totalistic rule
8n,8k,880,1,0<,81,1,1<,80,1,0<<<,81,1<<
two-dimensional 5-neighbor totalistic rule
8n,8k,880,k,0<,8k,1,k<,80,k,0<<<,81,1<<
two-dimensional 5-neighbor outer totalistic rule

Higher-dimensional rule specifications.

This is the rule specification for the two-dimensional 9-neighbor totalistic cellular automaton
with code 797.
In[27]:= code797 = 8797, 82, 1<, 81, 1<<;

This gives steps 0 and 1 in its evolution.

In[28]:= CellularAutomaton@code797, 8881<<, 0<, 1D
Out[28]= 8880, 0, 0<, 80, 1, 0<, 80, 0, 0<<, 880, 0, 0<, 80, 0, 0<, 80, 0, 0<<<
384 Mathematics and Algorithms

This shows step 70 in the evolution.

In[29]:= ArrayPlot@First@CellularAutomaton@code797, 8881<<, 0<,
8870<<DDD

Out[29]=

This shows all steps in a slice along the x axis.

In[30]:= ArrayPlot@Map@First,
CellularAutomaton@code797, 8881<<, 0<,
870, 80<, All<DDD

Out[30]=
Mathematics and Algorithms 385

Mathematical Functions

Naming Conventions
Mathematical functions in Mathematica are given names according to definite rules. As with
most Mathematica functions, the names are usually complete English words, fully spelled out.
For a few very common functions, Mathematica uses the traditional abbreviations. Thus the
modulo function, for example, is Mod, not Modulo.

Mathematical functions that are usually referred to by a person's name have names in Mathemat-
ica of the form PersonSymbol. Thus, for example, the Legendre polynomials Pn HxL are denoted
LegendreP@n, xD. Although this convention does lead to longer function names, it avoids any
ambiguity or confusion.

When the standard notation for a mathematical function involves both subscripts and super-
scripts, the subscripts are given before the superscripts in the Mathematica form. Thus, for
example, the associated Legendre polynomials Pm
n HxL are denoted LegendreP@n, m, xD.

Generic and Non-Generic Cases

This gives a result for the integral of xn that is valid for almost all values of n.
In[1]:= Integrate@x ^ n, xD
x1+n
Out[1]=
1+n

For the special case of x-1 , however, the correct result is different.
In[2]:= Integrate@x ^ - 1, xD
Out[2]= Log@xD

The overall goal of symbolic computation is typically to get formulas that are valid for many
possible values of the variables that appear in them. It is however often not practical to try to
get formulas that are valid for absolutely every possible value of each variable.
386 Mathematics and Algorithms

Mathematica always replaces 0 x by 0.

In[3]:= 0x
Out[3]= 0

If x is equal to 0, however, then the true result is not 0.

In[4]:= 00

1
Power::infy : Infinite expression encountered.
0

::indet : Indeterminate expression 0 ComplexInfinity encountered.

Out[4]= Indeterminate

This construct treats both cases, but would be quite unwieldy to use.
In[5]:= If@x != 0, 0, IndeterminateD
Out[5]= If@x 0, 0, IndeterminateD

If Mathematica did not automatically replace 0 x by 0, then few symbolic computations would
get very far. But you should realize that the practical necessity of making such replacements
can cause misleading results to be obtained when exceptional values of parameters are used.

The basic operations of Mathematica are nevertheless carefully set up so that whenever possi-
ble the results obtained will be valid for almost all values of each variable.

x2 is not automatically replaced by x.

In[6]:= Sqrt@x ^ 2D

Out[6]= x2

If it were, then the result here would be -2, which is incorrect.

In[7]:= % . x -> - 2
Out[7]= 2

This makes the assumption that x is a positive real variable, and does the replacement.
In[8]:= Simplify@Sqrt@x ^ 2D, x > 0D
Out[8]= x

Numerical Functions
Mathematics and Algorithms 387

Numerical Functions

IntegerPart @xD integer part of x

FractionalPart@xD fractional part of x
Round@xD integer Xx\ closest to x
Floor@xD greatest integer dxt not larger than x
Ceiling@xD least integer `xp not smaller than x
Rationalize @xD rational number approximation to x
Rationalize @x,dxD rational approximation within tolerance dx

Functions relating real numbers and integers.

x IntegerPart @xD FractionalPart@xD Round@xD Floor@xD Ceiling@xD

2.4 2 0.4 2 2 3
2.5 2 0.5 2 2 3
2.6 2 0.6 3 2 3
-2.4 -2 -0.4 -2 -3 -2
-2.5 -2 -0.5 -2 -3 -2
-2.6 -2 -0.6 -3 -3 -2

Extracting integer and fractional parts.

IntegerPart @xD and FractionalPart@xD can be thought of as extracting digits to the left and
right of the decimal point. Round@xD is often used for forcing numbers that are close to integers
to be exactly integers. Floor@xD and Ceiling@xD often arise in working out how many ele-
ments there will be in sequences of numbers with non-integer spacings.

Sign@xD 1 for x > 0, -1 for x < 0

UnitStep@xD 1 for x 0, 0 for x < 0
Abs@xD absolute value x of x
Clip@xD x clipped to be between -1 and +1
Rescale@x,8xmin ,xmax <D x rescaled to run from 0 to 1
Max@x1 ,x2 ,D or Max@8x1 ,x2 ,<,D
the maximum of x1 , x2 ,
Min@x1 ,x2 ,D or Min@8x1 ,x2 ,<,D
the minimum of x1 , x2 ,

Numerical functions of real variables.

388 Mathematics and Algorithms

x+I y the complex number x + i y

Re@zD the real part Re z
Im@zD the imaginary part Im z
Conjugate@zD the complex conjugate z* or z
Abs@zD the absolute value z

Arg@zD the argument f such that z = z eif

Numerical functions of complex variables.

Piecewise Functions

Boole@exprD give 1 if expr is True , and 0 if it is False

Turning conditions into numbers.

Boole@exprD is a basic function that turns True and False into 1 and 0. It is sometimes known
as the characteristic function or indicator function.

This gives the area of a unit disk.

In[1]:= Integrate@Boole@x ^ 2 + y ^ 2 <= 1D, 8x, - 1, 1<, 8y, - 1, 1<D
Out[1]= p

Piecewise@88val1 ,cond1 <,8val2 ,cond2 <,<D

give the first vali for which condi is True

Piecewise@88val1 ,cond1 <,<,valD give val if all condi are False

Piecewise functions.

It is often convenient to have functions with different forms in different regions. You can do this
using Piecewise.
Mathematics and Algorithms 389

This plots a piecewise function.

In[2]:= Plot@Piecewise@88x ^ 2, x < 0<, 81 - x, x > 0<<D, 8x, - 1, 1<D
1.0

0.8

0.6
Out[2]=
0.4

0.2

-1.0 -0.5 0.5 1.0

Piecewise functions appear in systems where there is discrete switching between different
domains. They are also at the core of many computational methods, including splines and finite
elements. Special cases include such functions as Abs, UnitStep, Clip, Sign, Floor and Max.
Mathematica handles piecewise functions in both symbolic and numerical situations.

This generates a square wave.

In[3]:= Plot@UnitStep@Sin@xDD, 8x, 0, 30<D
1.0

0.8

0.6
Out[3]=
0.4

0.2

5 10 15 20 25 30

Here is the integral of the square wave.

In[4]:= Integrate@UnitStep@Sin@xDD, 8x, 0, 30<D
Out[4]= 5 p

Pseudorandom Numbers
Mathematica has three functions for generating pseudorandom numbers that are distributed
uniformly over a range of values.

1
RandomInteger@D 0 or 1 with probability
2

RandomInteger@8imin ,imax <D an integer between imin and imax , inclusive

RandomInteger@imax D an integer between 0 and imax , inclusive

390 Mathematics and Algorithms

RandomReal@ D a real number between 0 and 1

RandomReal@8xmin , xmax <D a real number between xmin and xmax

RandomReal@xmax D a real number between 0 and xmax

RandomComplex@D a complex number in the unit square

RandomComplex@8zmin ,zmax <D a complex number in the rectangle defined by zmin and zmax

RandomComplex@zmax D a complex number in the rectangle defined by 0 and zmax

Pseudorandom number generation.

RandomReal@range,nD , RandomComplex@range,nD , RandomInteger@range,nD

a list of n pseudorandom numbers from the given range
RandomReal@range,8n1 ,n2 ,<D ,
RandomComplex@range,8n1 ,n2 ,<D , RandomInteger@range,8n1 ,n2 ,<D
an n1 n2 array of pseudorandom numbers

Generating tables of pseudorandom numbers.

This will give 0 or 1 with equal probability.

In[1]:= RandomInteger@D
Out[1]= 1

This gives a pseudorandom complex number.

In[2]:= RandomComplex@D
Out[2]= 0.2597 + 0.789124

This gives a list of 10 pseudorandom integers between 0 and 9 (inclusive).

In[3]:= RandomInteger@80, 9<, 10D
Out[3]= 87, 8, 0, 7, 5, 5, 1, 4, 2, 3<

This gives a matrix of pseudorandom reals between 0 and 1.

In[4]:= RandomReal@1, 83, 3<D
Out[4]= 880.798457, 0.90866, 0.0501389<, 80.254316, 0.428327, 0.734689<, 80.293283, 0.628561, 0.463396<<

RandomReal and RandomComplex allow you to obtain pseudorandom numbers with any precision.
Mathematics and Algorithms 391

option name default value

MachinePrecis precision to use for real or complex num -
WorkingPrecision
ion
bers

Changing the precision for pseudorandom numbers.

Here is a 30-digit pseudorandom real number in the range 0 to 1.

In[5]:= RandomReal@WorkingPrecision 30D
Out[5]= 0.592425435593582531459708475075

Here is a list of four 20-digit pseudorandom complex numbers.

In[6]:= RandomComplex@8- 1 - I, 1 + I<, 4, WorkingPrecision 20D
Out[6]= 80.90904993322285593741 + 0.12394429399708576389 ,
-0.32862367115658248044 - 0.99682159655818692192 ,
0.28641328852741506413 + 0.19292256786947586951 ,
0.59587180971009053802 + 0.83133753157739276902 <

If you get arrays of pseudorandom numbers repeatedly, you should get a "typical" sequence of
numbers, with no particular pattern. There are many ways to use such numbers.

One common way to use pseudorandom numbers is in making numerical tests of hypotheses.
For example, if you believe that two symbolic expressions are mathematically equal, you can
test this by plugging in "typical" numerical values for symbolic parameters, and then comparing
the numerical results. (If you do this, you should be careful about numerical accuracy problems
and about functions of complex variables that may not have unique values.)

Here is a symbolic equation.

In[7]:= Sqrt@x ^ 2D Abs@xD

Out[7]= x2 Abs@xD

Substituting in a random numerical value shows that the equation is not always True .
In[8]:= % . x -> RandomComplex@D
Out[8]= False

Other common uses of pseudorandom numbers include simulating probabilistic processes, and
sampling large spaces of possibilities. The pseudorandom numbers that Mathematica generates
for a range of numbers are always uniformly distributed over the range you specify.

RandomInteger, RandomReal and RandomComplex are unlike almost any other Mathematica
functions in that every time you call them, you potentially get a different result. If you use
them in a calculation, therefore, you may get different answers on different occasions.
392 Mathematics and Algorithms

The sequences that you get from RandomInteger, RandomReal and RandomComplex are not in
most senses "truly random", although they should be "random enough" for practical purposes.
The sequences are in fact produced by applying a definite mathematical algorithm, starting
from a particular "seed". If you give the same seed, then you get the same sequence.

When Mathematica starts up, it takes the time of day (measured in small fractions of a second)
as the seed for the pseudorandom number generator. Two different Mathematica sessions will
therefore almost always give different sequences of pseudorandom numbers.

If you want to make sure that you always get the same sequence of pseudorandom numbers,
you can explicitly give a seed for the pseudorandom generator, using SeedRandom.

SeedRandom@D reseed the pseudorandom generator, with the time of day

SeedRandom@sD reseed with the integer s

Pseudorandom number generator seed.

This reseeds the pseudorandom generator.

In[9]:= SeedRandom@143D

Here are three pseudorandom numbers.

In[10]:= RandomReal@1, 83<D
Out[10]= 80.110762, 0.364563, 0.163681<

If you reseed the pseudorandom generator with the same seed, you get the same sequence of
pseudorandom numbers.
In[11]:= SeedRandom@143D; RandomReal@1, 83<D
Out[11]= 80.110762, 0.364563, 0.163681<

Every single time RandomInteger, RandomReal or RandomComplex is called, the internal state of
the pseudorandom generator that it uses is changed. This means that subsequent calls to these
functions made in subsidiary calculations will have an effect on the numbers returned in your
main calculation. To avoid any problems associated with this, you can localize this effect of their
use by doing the calculation inside of BlockRandom .
Mathematics and Algorithms 393

BlockRandom @exprD evaluates expr with the current state of the pseudorandom
generators localized

Localizing the effects of using RandomInteger , RandomReal or RandomComplex .

By localizing the calculation inside BlockRandom , the internal state of the pseudorandom
generator is restored after generating the first list.
In[12]:= 8BlockRandom@8RandomReal@D, RandomReal@D<D, 8RandomReal@D, RandomReal@D<<
Out[12]= 880.952312, 0.93591<, 80.952312, 0.93591<<

Many applications require random numbers from non-uniform distributions. Mathematica has
many distributions built into the system. You can give a distribution with appropriate parame-
ters instead of a range to RandomInteger or RandomReal.

RandomInteger@distD , RandomReal@distD
a pseudorandom number distributed by the random distribu-
tion dist
RandomInteger@dist,nD , RandomReal@dist,nD
a list of n pseudorandom numbers distributed by the
random distribution dist
RandomInteger@dist,8n1 ,n2 ,<D , RandomReal@dist,8n1 ,n2 ,<D
an n1 n2 array of pseudorandom numbers distributed
by the random distribution dist

Generating pseudorandom numbers with non-uniform distributions.

This generates 12 integers distributed by the Poisson distribution with mean 3.

In[13]:= RandomInteger@PoissonDistribution@3D, 12D
Out[13]= 84, 3, 1, 3, 2, 2, 5, 3, 5, 2, 1, 3<

This generates a 44 matrix of real numbers using the standard normal distribution.
In[14]:= RandomReal@NormalDistribution@D, 84, 4<D
Out[14]= 881.17681, -0.774733, -1.74139, 1.3577<, 8-1.261, 0.0408214, 0.989022, 2.80942<,
8-1.27146, 1.63037, 1.98221, 0.403135<, 81.00722, -0.927379, 0.747369, -2.28065<<

This generates five high-precision real numbers distributed normally with mean 2 and standard
deviation 4.
In[15]:= RandomReal@NormalDistribution@2, 4D, 5, WorkingPrecision 32D
Out[15]= 8-3.7899344407106290701062146195097,
-1.1607986070402381009885236231751, 12.042079595098604792470496688453,
2.3508651879131153670572237267418, 5.0287452449413463045300577818173<

An additional use of pseudorandom numbers is for selecting from a list. RandomChoice selects
with replacement and RandomSample samples without replacement.
394 Mathematics and Algorithms

An additional use of pseudorandom numbers is for selecting from a list. RandomChoice selects
with replacement and RandomSample samples without replacement.

RandomChoice@list, nD choose n items at random from list

RandomChoice@list,8n1 ,n2 ,<D an n1 n2 array of values chosen randomly from list
RandomSample@list, nD a sample of size n from list

Selecting at random.

Choose 10 items at random from the digits 0 through 9.

In[16]:= RandomChoice@Range@0, 9D, 10D
Out[16]= 88, 8, 3, 5, 3, 5, 0, 8, 4, 1<

Chances are very high that at least one of the choices was repeated in the output. That is
because when an element is chosen, it is immediately replaced. On the other hand, if you want
to select from an actual set of elements, there should be no replacement.

Sample 10 items at random from the digits 0 through 9 without replacement. The result is a
random permutation of the digits.
In[17]:= RandomSample@Range@0, 9D, 10D
Out[17]= 87, 9, 2, 5, 3, 4, 1, 8, 0, 6<

Sample 10 items from a set having different frequencies for each digit.
In[18]:= RandomSample@80, 1, 2, 2, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9, 9<, 10D
Out[18]= 86, 6, 7, 8, 9, 7, 2, 2, 9, 9<

Integer and Number Theoretic Functions

Mod@k,nD k modulo n (remainder from dividing k by n)

Quotient@m,nD the quotient of m and n (truncation of m n)
QuotientRemainder@m,nD a list of the quotient and the remainder
Divisible@m,nD test whether m is divisible by n
CoprimeQ@n1 ,n2 ,D test whether the ni are pairwise relatively prime
GCD@n1 ,n2 ,D the greatest common divisor of n1 , n2 ,
Mathematics and Algorithms 395

LCM@n1 ,n2 ,D the least common multiple of n1 , n2 ,

KroneckerDelta@n1 ,n2 ,D the Kronecker delta dn1 n2 equal to 1 if all the ni are equal,
and 0 otherwise
IntegerDigits@n,bD the digits of n in base b
IntegerExponent@n,bD the highest power of b that divides n

Some integer functions.

The remainder on dividing 17 by 3.

In[1]:= Mod@17, 3D
Out[1]= 2

The integer part of 17/3.

In[2]:= Quotient@17, 3D
Out[2]= 5

Mod also works with real numbers.

In[3]:= [email protected], 1.2D
Out[3]= 0.8

The result from Mod always has the same sign as the second argument.
In[4]:= Mod@- 5.6, 1.2D
Out[4]= 0.4

For any integers a and b, it is always true that b * Quotient@a, bD + Mod@a, bD is equal to a.

Mod@k,nD result in the range 0 to n - 1

Mod@k,n,1D result in the range 1 to n
Mod@k,n,-n2D result in the range `- n 2p to d+ n 2t
Mod@k,n,dD result in the range d to d + n - 1

Integer remainders with offsets.

Particularly when you are using Mod to get indices for parts of objects, you will often find it
convenient to specify an offset.
396 Mathematics and Algorithms

This effectively extracts the 18th part of the list, with the list treated cyclically.
In[5]:= Part@8a, b, c<, Mod@18, 3, 1DD
Out[5]= c

The greatest common divisor function GCD@n1 , n2 , D gives the largest integer that divides all
the ni exactly. When you enter a ratio of two integers, Mathematica effectively uses GCD to
cancel out common factors and give a rational number in lowest terms.

The least common multiple function LCM@n1 , n2 , D gives the smallest integer that contains all
the factors of each of the ni .

The largest integer that divides both 24 and 15 is 3.

In[6]:= GCD@24, 15D
Out[6]= 3

The Kronecker delta function KroneckerDelta@n1 , n2 , D is equal to 1 if all the ni are equal,
and is 0 otherwise. dn1 n2 can be thought of as a totally symmetric tensor.

This gives a totally symmetric tensor of rank 3.

In[7]:= Array@KroneckerDelta, 83, 3, 3<D
Out[7]= 8881, 0, 0<, 80, 0, 0<, 80, 0, 0<<,
880, 0, 0<, 80, 1, 0<, 80, 0, 0<<, 880, 0, 0<, 80, 0, 0<, 80, 0, 1<<<

FactorInteger@nD a list of the prime factors of n, and their exponents

Divisors@nD a list of the integers that divide n

Prime@kD the kth prime number

PrimePi@xD the number of primes less than or equal to x
PrimeQ@nD give True if n is a prime, and False otherwise
PrimeNu@nD the number of distinct primes nHnL in n
PrimeOmega@nD the number of prime factors counting multiplicities WHnL in
n.
LiouvilleLambda@nD the Liouville function lHnL
MangoldtLambda@nD the von Mandgoldt function LHnL
FactorIntegerAn,GaussianIntegers->TrueE
a list of the Gaussian prime factors of the Gaussian integer
n, and their exponents
Mathematics and Algorithms 397

PrimeQAn,GaussianIntegers->TrueE
give True if n is a Gaussian prime, and False otherwise

Integer factoring and related functions.

This gives the factors of 24 as 23 , 31 . The first element in each list is the factor; the second is its
exponent.
In[8]:= FactorInteger@24D
Out[8]= 882, 3<, 83, 1<<

Here are the factors of a larger integer.

In[9]:= FactorInteger@111 111 111 111 111 111D
Out[9]= 883, 2<, 87, 1<, 811, 1<, 813, 1<, 819, 1<, 837, 1<, 852 579, 1<, 8333 667, 1<<

You should realize that according to current mathematical thinking, integer factoring is a funda-
mentally difficult computational problem. As a result, you can easily type in an integer that
Mathematica will not be able to factor in anything short of an astronomical length of time. But
as long as the integers you give are less than about 50 digits long, FactorInteger should have
no trouble. And in special cases it will be able to deal with much longer integers.

Here is a rather special long integer.

In[10]:= 30 !
Out[10]= 265 252 859 812 191 058 636 308 480 000 000

Mathematica can easily factor this special integer.

In[11]:= FactorInteger@%D
Out[11]= 882, 26<, 83, 14<, 85, 7<, 87, 4<, 811, 2<, 813, 2<, 817, 1<, 819, 1<, 823, 1<, 829, 1<<

Although Mathematica may not be able to factor a large integer, it can often still test whether
or not the integer is a prime. In addition, Mathematica has a fast way of finding the kth prime
number.

It is often much faster to test whether a number is prime than to factor it.
In[12]:= PrimeQ@234 242 423D
Out[12]= False

Here is a plot of the first 100 primes.

398 Mathematics and Algorithms

Here is a plot of the first 100 primes.

In[13]:= ListPlot@Table@Prime@nD, 8n, 100<DD

500

400

300
Out[13]=
200

100

20 40 60 80 100

This is the millionth prime.

In[14]:= Prime@1 000 000D
Out[14]= 15 485 863

Particularly in number theory, it is often more important to know the distribution of primes than
their actual values. The function PrimePi@xD gives the number of primes pHxL that are less than
or equal to x.

This gives the number of primes less than a billion.

In[15]:= PrimePi@10 ^ 9D
Out[15]= 50 847 534

PrimeNu gives the number of distinct primes.

In[16]:= PrimeNu@10 ^ 9D
Out[16]= 2

PrimeOmega gives the number of prime factors counting multiplicities WHnL in n.

In[17]:= PrimeOmega@10 ^ 9D
Out[17]= 18

Liouville's function gives H-1Lk where k is the number of prime factors counting multiplicity.
In[18]:= 8LiouvilleLambda@3 ^ 5D, LiouvilleLambda@2 3 ^ 5D<
Out[18]= 8-1, 1<

The Mangoldt function returns the log of prime power base or zero when composite.
In[19]:= 8MangoldtLambda@3 ^ 5D, MangoldtLambda@2 3 ^ 5D<
Out[19]= 8Log@3D, 0<

By default, FactorInteger allows only real integers. But with the option setting
Mathematics and Algorithms 399

By default, FactorInteger allows only real integers. But with the option setting
GaussianIntegers -> True, it also handles Gaussian integers, which are complex numbers with
integer real and imaginary parts. Just as it is possible to factor uniquely in terms of real primes,
it is also possible to factor uniquely in terms of Gaussian primes. There is nevertheless some
potential ambiguity in the choice of Gaussian primes. In Mathematica, they are always chosen
to have positive real parts, and non-negative imaginary parts, except for a possible initial factor
of -1 or i.

Over the Gaussian integers, 2 can be factored as H-iL H1 + iL2 .

In[20]:= FactorInteger@2, GaussianIntegers -> TrueD
Out[20]= 88-, 1<, 81 + , 2<<

Here are the factors of a Gaussian integer.

In[21]:= FactorInteger@111 + 78 I, GaussianIntegers -> TrueD
Out[21]= 882 + , 1<, 83, 1<, 820 + 3 , 1<<

PowerMod@a,b,nD the power ab modulo n

DirichletCharacter@k, j,nD the Dirichlet character c8k, j< HnL

EulerPhi@nD the Euler totient function fHnL

MoebiusMu@nD the Mbius function mHnL
DivisorSigma@k,nD the divisor function sk HnL

DivisorSum@n, formD the sum of form@iD for all i that divide n

DivisorSum@n, form,condD the sum for only those divisors for which cond@iD gives
True
n
JacobiSymbol@n,mD the Jacobi symbol J N
m

ExtendedGCD @n1 ,n2 ,D the extended GCD of n1 , n2 ,

MultiplicativeOrder@k,nD the multiplicative order of k modulo n
MultiplicativeOrder@k,n,8r1 ,r2 ,<D
the generalized multiplicative order with residues ri

CarmichaelLambda@nD the Carmichael function lHnL

PrimitiveRoot@nD a primitive root of n

Some functions from number theory.

The modular power function PowerMod@a, b, nD gives exactly the same results as Mod@a ^ b, nD
for b > 0. PowerMod is much more efficient, however, because it avoids generating the full form
of a ^ b.
400 Mathematics and Algorithms

The modular power function PowerMod@a, b, nD gives exactly the same results as Mod@a ^ b, nD
for b > 0. PowerMod is much more efficient, however, because it avoids generating the full form
of a ^ b.

You can use PowerMod not only to find positive modular powers, but also to find modular
inverses. For negative b, PowerMod@a, b, nD gives, if possible, an integer k such that
k a-b 1 mod n. (Whenever such an integer exists, it is guaranteed to be unique modulo n.) If no
such integer k exists, Mathematica leaves PowerMod unevaluated.

PowerMod is equivalent to using Power , then Mod, but is much more efficient.
In[22]:= PowerMod@2, 13 451, 3D
Out[22]= 2

This gives the modular inverse of 3 modulo 7.

In[23]:= PowerMod@3, - 1, 7D
Out[23]= 5

Multiplying the inverse by 3 modulo 7 gives 1, as expected.

In[24]:= Mod@3 %, 7D
Out[24]= 1

This finds the smallest non-negative integer x so that x2 is equal to 3 mod 11.
In[25]:= PowerMod@3, 1 2, 11D
Out[25]= 5

This verifies the result.

In[26]:= ModA52 , 11E

Out[26]= 3

This returns all integers less than 11 which satisfy the relation.
In[27]:= PowerModList@3, 1 2, 11D
Out[27]= 85, 6<
Mathematics and Algorithms 401

If d does not have a square root modulo n, PowerMod@d, nD will remain unevaluated and
PowerModList will return an empty list.
In[28]:= PowerMod@3, 1 2, 5D

PowerMod::root : The equation x^2 = 3 Hmod 5L has no integer solutions.

1
Out[28]= PowerModB3, , 5F
2

In[29]:= PowerModList@3, 1 2, 5D
Out[29]= 8<

This checks that 3 is not a square modulo 5.

In[30]:= ModA80, 1, 2, 3, 4<2 , 5E

Out[30]= 80, 1, 4, 4, 1<

Even for a large modulus, the square root can be computed fairly quickly.

In[31]:= PowerModA2, 1 2, 1064 + 57E Timing

Out[31]= 80.01, 876 504 467 496 681 643 735 926 111 996 546 100 401 033 611 976 777 074 909 122 865<

PowerMod@d, nD also works for composite n.

In[32]:= PowerModA3, 1 2, 113 E
Out[32]= 578

There are fHkL distinct Dirichlet characters for a given modulus k, as labeled by the index j.
Different conventions can give different orderings for the possible characters.

DirichletCharacter works for very large numbers.

In[33]:= DirichletCharacter@20 !, 300, 23D
p

Out[33]= 9

The Euler totient function fHnL gives the number of integers less than n that are relatively prime
to n. An important relation (Fermat's little theorem) is that af HnL 1 mod n for all a relatively prime
to n.

The Mbius function mHnL is defined to be H-1Lk if n is a product of k distinct primes, and 0 if n
contains a squared factor (other than 1). An important relation is the Mbius inversion formula,
which states that if gHnL = d n f HdL for all n, then f HnL = d n mHdL gHn dL, where the sums are over all

positive integers d that divide n.

The divisor function sk HnL is the sum of the kth powers of the divisors of n. The function s0 HnL
gives the total number of divisors of n, and is variously denoted dHnL, nHnL and tHnL. The function
402 Mathematics and Algorithms

For prime n, fHnL = n - 1.

In[34]:= EulerPhi@17D
Out[34]= 16

The result is 1, as guaranteed by Fermats little theorem.

In[35]:= PowerMod@3, %, 17D
Out[35]= 1

This gives a list of all the divisors of 24.

In[36]:= Divisors@24D
Out[36]= 81, 2, 3, 4, 6, 8, 12, 24<

s0 HnL gives the total number of distinct divisors of 24.

In[37]:= DivisorSigma@0, 24D
Out[37]= 8

The function DivisorSum@n, formD represents the sum of form@iD for all i that divide n.
DivisorSum@n, form, condD includes only those divisors for which cond@iD gives True.

This gives a list of sums for the divisors of five positive integers.
In[38]:= Table@DivisorSum@n, &D, 8n, 5<D
Out[38]= 81, 3, 4, 7, 6<

This imposes the condition that the value of each divisor i must be less than 6.
In[39]:= Table@DivisorSum@n, &, < 6 &D, 8n, 11, 15<D
Out[39]= 81, 10, 1, 3, 9<

n
The Jacobi symbol JacobiSymbol@n, mD reduces to the Legendre symbol J m N when m is an odd

prime. The Legendre symbol is equal to zero if n is divisible by m; otherwise it is equal to 1 if n is

a quadratic residue modulo the prime m, and to -1 if it is not. An integer n relatively prime to m
is said to be a quadratic residue modulo m if there exists an integer k such that k2 n mod m. The
n
full Jacobi symbol is a product of the Legendre symbols J p N for each of the prime factors pi such
i

that m = i pi .
The Jacobi symbol JacobiSymbol@n, mD reduces to the Legendre symbol J m N when m is an odd

prime. The Legendre symbol is equal to zero if n is divisible by m; otherwise it is equal to 1 if n is

Mathematics and Algorithms 403

is said to be a quadratic residue modulo m if there exists an integer k such that k2 n mod m. The
n
full Jacobi symbol is a product of the Legendre symbols J p N for each of the prime factors pi such
i

that m = i pi .

The extended GCD ExtendedGCD @n1 , n2 , D gives a list 8g, 8r1 , r2 , << where g is the greatest
common divisor of the ni , and the ri are integers such that g = r1 n1 + r2 n2 + . The extended GCD
is important in finding integer solutions to linear Diophantine equations.

The first number in the list is the GCD of 105 and 196.
In[40]:= ExtendedGCD@105, 196D
Out[40]= 87, 8-13, 7<<

The second pair of numbers satisfies g = r m + s n.

In[41]:= - 13 105 + 7 196
Out[41]= 7

The multiplicative order function MultiplicativeOrder@k, nD gives the smallest integer m such
that km 1 mod n. Then m is known as the order of k modulo n. The notation ordn HkL is occasionally
used.

The generalized multiplicative order function MultiplicativeOrder@k, n, 8r1 , r2 , <D gives

the smallest integer m such that km ri mod n for some i. MultiplicativeOrder@k, n, 8- 1, 1<D
is sometimes known as the suborder function of k modulo n, denoted sordn HkL.
MultiplicativeOrder@k, n, 8a<D is sometimes known as the discrete log of a with respect to
the base k modulo n.

The Carmichael function or least universal exponent lHnL gives the smallest integer m such that
km 1 mod n for all integers k relatively prime to n.

ContinuedFraction@x,nD generate the first n terms in the continued fraction represen -

tation of x
FromContinuedFraction@listD reconstruct a number from its continued fraction
representation
Rationalize @x,dxD find a rational approximation to x with tolerance dx

Continued fractions.
404 Mathematics and Algorithms

This generates the first 10 terms in the continued fraction representation for p.
In[42]:= ContinuedFraction@Pi, 10D
Out[42]= 83, 7, 15, 1, 292, 1, 1, 1, 2, 1<

This reconstructs the number represented by the list of continued fraction terms.
In[43]:= FromContinuedFraction@%D
1 146 408
Out[43]=
364 913

The result is close to p.

In[44]:= N@%D
Out[44]= 3.14159

This gives directly a rational approximation to p.

In[45]:= Rationalize@Pi, 1 1000D
201
Out[45]=
64

Continued fractions appear in many number theoretic settings. Rational numbers have terminat-
ing continued fraction representations. Quadratic irrational numbers have continued fraction
representations that become repetitive.

ContinuedFraction@xD the complete continued fraction representation for a

rational or quadratic irrational number
QuadraticIrrationalQ@xD test whether x is a quadratic irrational
RealDigits@xD the complete digit sequence for a rational number
RealDigits@x,bD the complete digit sequence in base b

Complete representations for numbers.

The continued fraction representation of 79 starts with the term 8, then involves a sequence
of terms that repeat forever.
In[46]:= ContinuedFraction@Sqrt@79DD
Out[46]= 88, 81, 7, 1, 16<<
Mathematics and Algorithms 405

This reconstructs 79 from its continued fraction representation.

In[47]:= FromContinuedFraction@%D

Out[47]= 79

This number is a quadratic irrational.

In[48]:= QuadraticIrrationalQ@ Sqrt@79DD
Out[48]= True

This shows the recurring sequence of decimal digits in 3/7.

In[49]:= RealDigits@3 7D
Out[49]= 8884, 2, 8, 5, 7, 1<<, 0<

FromDigits reconstructs the original number.

In[50]:= FromDigits@%D
3
Out[50]=
7

Continued fraction convergents are often used to approximate irrational numbers by rational
ones. Those approximations alternate from above and below, and converge exponentially in the
number of terms. Furthermore, a convergent p q of a simple continued fraction is better than
any other rational approximation with denominator less than or equal to q.

Convergents @xD give a list of rational approximations of x

Convergents @x,nD give only the first n approximations

Continued fraction convergents.

This gives a list of rational approximations of 101/9801, derived from its continued fraction
expansion.
In[51]:= Convergents@ 101 9801 D
1 25 101
Out[51]= :0, , , >
97 2426 9801

This lists only the first 10 convergents.

In[52]:= Convergents@ 10 201 96 059 601, 10 D
1 1 3 16 19 92 203 498 701
Out[52]= :0, , , , , , , , , >
9416 9417 28 250 150 667 178 917 866 335 1 911 587 4 689 509 6 601 096

This lists successive rational approximations to p, until the numerical precision is exhausted.
406 Mathematics and Algorithms

This lists successive rational approximations to p, until the numerical precision is exhausted.
In[53]:= Convergents@ N@ Pi D D
22 333 355 103 993 104 348 208 341
Out[53]= :3, , , , , , ,
7 106 113 33 102 33 215 66 317
312 689 833 719 1 146 408 4 272 943 5 419 351 80 143 857
, , , , , >
99 532 265 381 364 913 1 360 120 1 725 033 25 510 582

With an exact irrational number, you have to explicitly ask for a certain number of terms.
In[54]:= Convergents@ Pi, 10 D
22 333 355 103 993 104 348 208 341 312 689 833 719 1 146 408
Out[54]= :3, , , , , , , , , >
7 106 113 33 102 33 215 66 317 99 532 265 381 364 913

LatticeReduce@8v1 v2 ,<D a reduced lattice basis for the set of integer vectors vi

HermiteDecomposition@8v1 ,v2 ,<D the echelon form for the set of integer vectors vi

Functions for integer lattices.

The lattice reduction function LatticeReduce@8v1 , v2 , <D is used in several kinds of modern
algorithms. The basic idea is to think of the vectors vk of integers as defining a mathematical
lattice. Any vector representing a point in the lattice can be written as a linear combination of
the form ck vk , where the ck are integers. For a particular lattice, there are many possible
choices of the "basis vectors" vk . What LatticeReduce does is to find a reduced set of basis
vectors v k for the lattice, with certain special properties.

Three unit vectors along the three coordinate axes already form a reduced basis.
In[55]:= LatticeReduce@881, 0, 0<, 80, 1, 0<, 80, 0, 1<<D
Out[55]= 881, 0, 0<, 80, 1, 0<, 80, 0, 1<<

This gives the reduced basis for a lattice in four-dimensional space specified by three vectors.
In[56]:= l = LatticeReduce@881, 0, 0, 12 345<, 80, 1, 0, 12 435<, 80, 0, 1, 12 354<<D
Out[56]= 88-1, 0, 1, 9<, 89, 1, -10, 0<, 885, -143, 59, 6<<

Notice that in the last example, LatticeReduce replaces vectors that are nearly parallel by
vectors that are more perpendicular. In the process, it finds some quite short basis vectors.

For a matrix m, HermiteDecomposition gives matrices u and r such that u is unimodular, u.m = r,
and r is in reduced row echelon form. In contrast to RowReduce, pivots may be larger than 1
because there are no fractions in the ring of integers. Entries above a pivot are minimized by
subtracting appropriate multiples of the pivot row.
Mathematics and Algorithms 407

In this case, the original matrix is recovered because it was in row echelon form.
In[57]:= 8u, r< = HermiteDecomposition@lD
Out[57]= 8881371, 143, 1<, 81381, 144, 1<, 81372, 143, 1<<,
881, 0, 0, 12 345<, 80, 1, 0, 12 435<, 80, 0, 1, 12 354<<<

This satisfies the required identities.

In[58]:= 8Abs@Det@uDD 1, u.l r<
Out[58]= 8True, True<

Here the second matrix has some pivots larger than 1, and nonzero entries over pivots.
In[59]:= HermiteDecomposition@88- 2, 1, 1<, 85, 9, 4<, 8- 4, 2, - 11<<D
Out[59]= 8882, 1, 0<, 83, 2, 1<, 82, 0, -1<<, 881, 11, 6<, 80, 23, 0<, 80, 0, 13<<<

DigitCount@n,b,dD the number of d digits in the base-b representation of n

Digit count function.

Here are the digits in the base-2 representation of the number 77.
In[60]:= IntegerDigits@77, 2D
Out[60]= 81, 0, 0, 1, 1, 0, 1<

This directly computes the number of ones in the base-2 representation.

In[61]:= DigitCount@77, 2, 1D
Out[61]= 4

The plot of the digit count function is self-similar.

In[62]:= ListLinePlot@Table@DigitCount@n, 2, 1D, 8n, 128<DD
7

Out[62]= 4
3

20 40 60 80 100 120
408 Mathematics and Algorithms

BitAnd@n1 ,n2 ,D bitwise AND of the integers ni

BitOr@n1 ,n2 ,D bitwise OR of the integers ni

BitXor@n1 ,n2 ,D bitwise XOR of the integers ni

BitNot@nD bitwise NOT of the integer n

BitLength@nD number of binary bits in the integer n
BitSet@n,kD set bit k to 1 in the integer n
BitGet@n,kD get bit k from the integer n
BitClear@n,kD set bit k to 0 in the integer n
BitShiftLeft@n,kD shift the integer n to the left by k bits, padding with zeros
BitShiftRight@n,kD shift to the right, dropping the last k bits

Bitwise operations.

Bitwise operations act on integers represented as binary bits. BitAnd@n1 , n2 , D yields the
integer whose binary bit representation has ones at positions where the binary bit representa-
tions of all of the ni have ones. BitOr@n1 , n2 , D yields the integer with ones at positions where
any of the ni have ones. BitXor@n1 , n2 D yields the integer with ones at positions where n1 or n2
but not both have ones. BitXor@n1 , n2 , D has ones where an odd number of the ni have ones.

This finds the bitwise AND of the numbers 23 and 29 entered in base 2.
In[63]:= BaseForm@BitAnd@2^^10111, 2^^11101D, 2D
Out[63]//BaseForm= 101012

Bitwise operations are used in various combinatorial algorithms. They are also commonly used
in manipulating bitfields in low-level computer languages. In such languages, however, integers
normally have a limited number of digits, typically a multiple of 8. Bitwise operations in Mathe-
matica in effect allow integers to have an unlimited number of digits. When an integer is nega-
tive, it is taken to be represented in two's complement form, with an infinite sequence of ones
on the left. This allows BitNot@nD to be equivalent simply to -1 - n.

SquareFreeQ @nD give True if n does not contain a squared factor, False
otherwise

Testing for a squared factor.

SquareFreeQ @nD checks to see if n has a square prime factor. This is done by computing
MoebiusMu@nD and seeing if the result is zero; if it is, then n is not squarefree, otherwise it is.
Computing MoebiusMu@nD involves finding the smallest prime factor q of n. If n has a small
prime factor (less than or equal to 1223), this is very fast. Otherwise, FactorInteger is used to
find q.
Mathematics and Algorithms 409

This product of primes contains no squared factors.

In[64]:= SquareFreeQ@2 3 5 7D
Out[64]= True

The square number 4 divides 60.

In[65]:= SquareFreeQ@60D
Out[65]= False

SquareFreeQ can handle large integers.

In[66]:= SquareFreeQA2101 - 1E
Out[66]= True

NextPrime@nD give the smallest prime larger than n

RandomPrime @8min,max<D return a random prime number between min and max
RandomPrime @maxD return a random prime number less than or equal to max
RandomPrime @8min,max<,nD return n random prime numbers between min and max
RandomPrime @max,nD return n random prime numbers less than or equal to max

Finding prime numbers.

NextPrime@nD finds the smallest prime p such that p > n. For n less than 20 digits, the algorithm
does a direct search using PrimeQ on the odd numbers greater than n. For n with more than 20
digits, the algorithm builds a small sieve and first checks to see whether the candidate prime is
divisible by a small prime before using PrimeQ. This seems to be slightly faster than a direct
search.

This gives the next prime after 10.

In[67]:= NextPrime@10D
Out[67]= 11
410 Mathematics and Algorithms

Even for large numbers, the next prime can be computed rather quickly.

In[68]:= NextPrimeA10100 E Timing

Out[68]= 80.06,
10 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000
000 000 000 000 000 267<

This gives the largest prime less than 34.

In[69]:= - NextPrime@- 34D
Out[69]= 31

For RandomPrime @8min, max<D and RandomPrime @maxD, a random prime p is obtained by ran-
domly selecting from a prime lookup table if max is small and by a random search of integers in
the range if max is large. If no prime exists in the specified range, the input is returned unevalu-
ated with an error message.

Here is a random prime between 10 and 100.

In[70]:= RandomPrime@810, 100<D
Out[70]= 79

PrimePowerQ @nD determine whether n is a positive integer power of a

rational prime

Testing for involving prime powers.

The algorithm for PrimePowerQ involves first computing the least prime factor p of n and then
attempting division by n until either 1 is obtained, in which case n is a prime power, or until
division is no longer possible, in which case n is not a prime power.

Here is a number that is a power of a single prime.

In[71]:= PrimePowerQ@12 167D
Out[71]= True

Over the GaussianIntegers this is a prime power.

In[72]:= PrimePowerQ@H1 + IL ^ 3D
Out[72]= True
Mathematics and Algorithms 411

ChineseRemainder@list1 ,list2 D give the smallest non-negative integer r with

Mod@r, list2 D == list1

Solving simultaneous congruences.

The Chinese remainder theorem states that a certain class of simultaneous congruences always
has a solution. ChineseRemainder@list1 , list2 D finds the smallest non-negative integer r such
that Mod@r, list2 D is list1 . The solution is unique modulo the least common multiple of the ele-
ments of list2 .

This means that 244 0 mod 4, 244 1 mod 9, and 244 2 mod 121.
In[73]:= ChineseRemainder@80, 1, 2<, 84, 9, 121<D
Out[73]= 244

This confirms the result.

In[74]:= Mod@244, 84, 9, 121<D
Out[74]= 80, 1, 2<

Longer lists are still quite fast.

In[75]:= ChineseRemainder@Range@20D, Prime@Range@20DDD
Out[75]= 169 991 099 649 125 127 278 835 143

PrimitiveRoot@nD give a primitive root of n, where n is a prime power or

twice a prime power

Computing primitive roots.

PrimitiveRoot@nD returns a generator for the group of numbers relatively prime to n under
multiplication mod n. This has a generator if and only if n is 2, 4, a power of an odd prime, or
twice a power of an odd prime. If n is a prime or prime power, the least positive primitive root
will be returned.

Here is a primitive root of 5.

In[76]:= PrimitiveRoot@5D
Out[76]= 2
412 Mathematics and Algorithms

This confirms that it does generate the group.

In[77]:= SortAModA2Range@4D , 5EE

Out[77]= 81, 2, 3, 4<

Here is a primitive root of a prime power.

In[78]:= PrimitiveRootA10933 E
Out[78]= 5

Here is a primitive root of twice a prime power.

In[79]:= PrimitiveRootA2 55 E
Out[79]= 3127

If the argument is composite and not a prime power or twice a prime power, the function does
not evaluate.
In[80]:= PrimitiveRoot@11 13D
Out[80]= PrimitiveRoot@143D

SquaresR@d,nD give the number of representations of an integer n as a

sum of d squares
PowersRepresentations@n,k,pD give the distinct representations of the integer n as a sum
of k non-negative pth integer powers

Representing an integer as a sum of squares or other powers.

Here are the representations of 101 as a sum of 3 squares.

In[81]:= PowersRepresentations@101, 3, 2D
Out[81]= 880, 1, 10<, 81, 6, 8<, 82, 4, 9<, 84, 6, 7<<
Mathematics and Algorithms 413

Combinatorial Functions

n! factorial nHn - 1L Hn - 2L 1
n!! double factorial nHn - 2L Hn - 4L
n
Binomial@n,mD binomial coefficient = n ! @m ! Hn - mL !D
m
Multinomial @n1 ,n2 ,D multinomial coefficient Hn1 + n2 + L ! Hn1 ! n2 ! L
CatalanNumber@nD Catalan number cn

Hyperfactorial@nD hyperfactorial 11 22 ... nn

BarnesG@nD Barnes G-function 1 ! 2 ! ... Hn - 2L !
Subfactorial@nD number of derangements of n objects
Fibonacci@nD Fibonacci number Fn

Fibonacci@n,xD Fibonacci polynomial Fn HxL

LucasL@nD Lucas number Ln

LucasL@n,xD Lucas polynomial Ln HxL

HarmonicNumber@nD harmonic number Hn

HarmonicNumber@n,rD harmonic number Hnr of order r

BernoulliB@nD Bernoulli number Bn

BernoulliB@n,xD Bernoulli polynomial Bn HxL

NorlundB@n,aD Nrlund polynomial BHaL

NorlundB@n,a,xD generalized Bernoulli polynomial BHaL

n HxL

EulerE@nD Euler number En

EulerE@n,xD Euler polynomial En HxL

StirlingS1@n,mD Stirling number of the first kind SnHmL

StirlingS2@n,mD Stirling number of the second kind HmL

BellB@nD Bell number Bn

BellB@n,xD Bell polynomial Bn HxL

PartitionsP @nD the number pHnL of unrestricted partitions of the integer n

IntegerPartitions@nD partitions of an integer
PartitionsQ @nD the number qHnL of partitions of n into distinct parts
Signature@8i1 ,i2 ,<D the signature of a permutation

Combinatorial functions.

The factorial function n ! gives the number of ways of ordering n objects. For non-integer n, the
numerical value of n ! is obtained from the gamma function, discussed in "Special Functions".
414 Mathematics and Algorithms

The factorial function n ! gives the number of ways of ordering n objects. For non-integer n, the
numerical value of n ! is obtained from the gamma function, discussed in "Special Functions".

n
The binomial coefficient Binomial@n, mD can be written as = n ! @m ! Hn - mL !D. It gives the
m
number of ways of choosing m objects from a collection of n objects, without regard to order.
The Catalan numbers, which appear in various tree enumeration problems, are given in terms
2n
of binomial coefficients as cn = Hn + 1L.
n

The subfactorial Subfactorial@nD gives the number of permutations of n objects that leave no
object fixed. Such a permutation is called a derangement. The subfactorial is given by
n ! nk=0 H-1Lk k !.

The multinomial coefficient Multinomial @n1 , n2 , D, denoted HN; n1 , n2 , , nm L = N !

Hn1 ! n2 ! nm !L, gives the number of ways of partitioning N distinct objects into m sets of sizes ni
(with N = m
i=1 ni ).

Mathematica gives the exact integer result for the factorial of an integer.
In[1]:= 30 !
Out[1]= 265 252 859 812 191 058 636 308 480 000 000

For non-integers, Mathematica evaluates factorials using the gamma function.

In[2]:= 3.6 !
Out[2]= 13.3813

Mathematica can give symbolic results for some binomial coefficients.

In[3]:= Binomial@n, 2D
1
Out[3]= H-1 + nL n
2

This gives the number of ways of partitioning 6 + 5 = 11 objects into sets containing 6 and 5
objects.
In[4]:= Multinomial@6, 5D
Out[4]= 462
Mathematics and Algorithms 415

11
The result is the same as .
6
In[5]:= Binomial@11, 6D
Out[5]= 462

The Fibonacci numbers Fibonacci@nD satisfy the recurrence relation Fn = Fn-1 + Fn-2 with
F1 = F2 = 1. They appear in a wide range of discrete mathematical problems. For large n, Fn Fn-1
approaches the golden ratio. The Lucas numbers LucasL@nD satisfy the same recurrence rela-
tion as the Fibonacci numbers do, but with initial conditions L1 = 1 and L2 = 3.

The Fibonacci polynomials Fibonacci@n, xD appear as the coefficients of tn in the expansion of

n=0 Fn HxL t .
t I1 - x t - t2 M = n

The harmonic numbers HarmonicNumber@nD are given by Hn = ni=1 1 i; the harmonic numbers of

order r HarmonicNumber@n, rD are given by HnHrL = ni=1 1 ir . Harmonic numbers appear in many

combinatorial estimation problems, often playing the role of discrete analogs of logarithms.

The Bernoulli polynomials BernoulliB@n, xD satisfy the generating function relation

n=0 Bn HxL t n ! .
= The Bernoulli numbers BernoulliB@nD are given by Bn = Bn H0L. The Bn
xt t n
t e He - 1L

appear as the coefficients of the terms in the Euler|Maclaurin summation formula for approximat-
ing integrals. The Bernoulli numbers are related to the Genocchi numbers by Gn = 2 H1 - 2n L Bn .

Numerical values for Bernoulli numbers are needed in many numerical algorithms. You can
always get these numerical values by first finding exact rational results using BernoulliB@nD,
and then applying N.

The Euler polynomials EulerE@n, xD have generating function 2 ex t Het + 1L =

n=0 En HxL t n ! , and
n

1
the Euler numbers EulerE@nD are given by En = 2n En J 2 N.

The Nrlund polynomials NorlundB@n, aD satisfy the generating function relation

a
n=0 Bn t n ! .
The Nrlund polynomials give the Bernoulli numbers when a = 1. For
a t HaL n
t He - 1L =

other positive integer values of a, the Nrlund polynomials give higher-order Bernoulli numbers.
The generalized Bernoulli polynomials NorlundB@n, a, xD satisfy the generating function relation
ta ex t Het - 1La =
n=0 Bn HxL t n ! .
HaL n
416 Mathematics and Algorithms

This gives the second Bernoulli polynomial B2 HxL.

In[6]:= BernoulliB@2, xD
1
Out[6]= - x + x2
6

You can also get Bernoulli polynomials by explicitly computing the power series for the generat-
ing function.
In[7]:= Series@t Exp@x tD HExp@tD - 1L, 8t, 0, 4<D
1 1 1 1
Out[7]= 1 + - +x t+ I1 - 6 x + 6 x2 M t2 + Ix - 3 x2 + 2 x3 M t3 + I-1 + 30 x2 - 60 x3 + 30 x4 M t4 + O@tD5
2 12 12 720

BernoulliB@nD gives exact rational-number results for Bernoulli numbers.

In[8]:= BernoulliB@20D
174 611
Out[8]= -
330

Stirling numbers show up in many combinatorial enumeration problems. For Stirling numbers of
the first kind StirlingS1@n, mD, H-1Ln-m SnHmL gives the number of permutations of n elements

which contain exactly m cycles. These Stirling numbers satisfy the generating function relation
x Hx - 1L Hx - n + 1L = nm=0 SnHmL xm . Note that some definitions of the SnHmL differ by a factor H-1Ln-m

from what is used in Mathematica.

Stirling numbers of the second kind StirlingS2@n, mD, sometimes denoted HmL
n , give the num-

ber of ways of partitioning a set of n elements into m non-empty subsets. They satisfy the
relation xn = nm=0 HmL
n x Hx - 1L ... Hx - m + 1L.

The Bell numbers BellB@nD give the total number of ways that a set of n elements can be
partitioned into non-empty subsets. The Bell polynomials BellB@n, xD satisfy the generating
tn
function relation eHe -1L x = .
t
n=0 Bn HxL n!

The partition function PartitionsP @nD gives the number of ways of writing the integer n as a
sum of positive integers, without regard to order. PartitionsQ @nD gives the number of ways of
writing n as a sum of positive integers, with the constraint that all the integers in each sum are
distinct.
Mathematics and Algorithms 417

IntegerPartitions@nD gives a list of the partitions of n, with length PartitionsP @nD.

This gives a table of Stirling numbers of the first kind.

In[9]:= Table@StirlingS1@5, iD, 8i, 5<D
Out[9]= 824, -50, 35, -10, 1<

The Stirling numbers appear as coefficients in this product.

In[10]:= Expand@Product@x - i, 8i, 0, 4<DD
2 3 4 5
Out[10]= 24 x - 50 x + 35 x - 10 x + x

Here are the partitions of 4.

In[11]:= IntegerPartitions@4D
Out[11]= 884<, 83, 1<, 82, 2<, 82, 1, 1<, 81, 1, 1, 1<<

The number of partitions is given by PartitionsP @4D.

In[12]:= Length@%D PartitionsP@4D
Out[12]= True

This gives the number of partitions of 100, with and without the constraint that the terms
should be distinct.
In[13]:= 8PartitionsQ@100D, PartitionsP@100D<
Out[13]= 8444 793, 190 569 292<

The partition function pHnL increases asymptotically like e n . Note that you cannot simply use
Plot to generate a plot of a function like PartitionsP because the function can only be
evaluated with integer arguments.
In[14]:= ListPlot@Table@N@Log@PartitionsP@nDDD, 8n, 100<DD

10
Out[14]=

20 40 60 80 100

Most of the functions here allow you to count various kinds of combinatorial objects. Functions
like IntegerPartitions and Permutations allow you instead to generate lists of various combi-
nations of elements.

The signature function Signature@8i1 , i2 , <D gives the signature of a permutation. It is equal
to +1 for even permutations (composed of an even number of transpositions), and to -1 for odd
permutations. The signature function can be thought of as a totally antisymmetric tensor, Levi-
Civita symbol or epsilon symbol.

ClebschGordan@ Clebsch-Gordan coefficient

8 j1 ,m1 <,8 j2 ,m2 <,8 j,m<D
ThreeJSymbol@ Wigner 3-j symbol
8 j1 ,m1 <,8 j2 ,m2 <,8 j3 ,m3 <D
SixJSymbol@8 j1 , j2 , j3 <,8 j4 , j5 , j6 <D Racah 6-j symbol

Rotational coupling coefficients.

Clebsch|Gordan coefficients and n-j symbols arise in the study of angular momenta in quantum
mechanics, and in other applications of the rotation group. The Clebsch|Gordan coefficients
ClebschGordan@8 j1 , m1 <, 8 j2 , m2 <, 8 j, m<D give the coefficients in the expansion of the quan-
tum mechanical angular momentum state j, m\ in terms of products of states j1 , m1 \ j2 , m2 \.

The 3-j symbols or Wigner coefficients ThreeJSymbol@8 j1 , m1 <, 8 j2 , m2 <, 8 j3 , m3 <D are a more
symmetrical form of Clebsch|Gordan coefficients. In Mathematica, the Clebsch|Gordan coeffi-
j j j j1 j2 j3
cients are given in terms of 3-j symbols by Cm11 m2 23m3 = H-1Lm3 + j1 - j2 2 j3 + 1 .
m1 m2 -m3

The 6-j symbols SixJSymbol@8 j1 , j2 , j3 <, 8 j4 , j5 , j6 <D give the couplings of three quantum
mechanical angular momentum states. The Racah coefficients are related by a phase to the 6-j
symbols.

You can give symbolic parameters in 3-j symbols.

In[15]:= ThreeJSymbol@8j, m<, 8j + 1 2, - m - 1 2<, 81 2, 1 2<D

1+j+m
H-1L-j+m
1+3 j+2 j2
Out[15]= -
2
Mathematics and Algorithms 419

Elementary Transcendental Functions

Exp@zD exponential function ez

Log@zD logarithm loge HzL

Log@b,zD logarithm logb HzL to base b

Log2@zD , Log10@zD logarithm to base 2 and 10

Sin@zD , Cos@zD , Tan@zD , Csc@zD , Sec@zD , Cot@zD
trigonometric functions (with arguments in radians)
ArcSin@zD , ArcCos@zD , ArcTan@zD , ArcCsc@zD , ArcSec@zD , ArcCot@zD
inverse trigonometric functions (giving results in radians)
ArcTan@x,yD the argument of x + iy
Sinh@zD , Cosh@zD , Tanh@zD , Csch@zD , Sech@zD , Coth@zD
hyperbolic functions
ArcSinh@zD , ArcCosh@zD , ArcTanh@zD , ArcCsch@zD , ArcSech@zD , ArcCoth@zD
inverse hyperbolic functions
Sinc@zD sinc function sinHzL z
Haversine@zD haversine function havHzL

InverseHaversine@zD inverse haversine function hav-1 HzL

Gudermannian@zD Gudermannian function gdHzL

InverseGudermannian@zD inverse Gudermannian function gd-1 HzL

Elementary transcendental functions.

Mathematica gives exact results for logarithms whenever it can. Here is log2 1024.
In[1]:= Log@2, 1024D
Out[1]= 10

You can find the numerical values of mathematical functions to any precision.
In[2]:= N@Log@2D, 40D
Out[2]= 0.6931471805599453094172321214581765680755

This gives a complex number result.

In[3]:= N@Log@- 2DD
Out[3]= 0.693147 + 3.14159

Mathematica can evaluate logarithms with complex arguments.

420 Mathematics and Algorithms

Mathematica can evaluate logarithms with complex arguments.

In[4]:= N@Log@2 + 8 IDD
Out[4]= 2.10975 + 1.32582

The arguments of trigonometric functions are always given in radians.

In[5]:= Sin@Pi 2D
Out[5]= 1

You can convert from degrees by explicitly multiplying by the constant Degree.
In[6]:= N@Sin@30 DegreeDD
Out[6]= 0.5

Here is a plot of the hyperbolic tangent function. It has a characteristic "sigmoidal" form.
In[7]:= Plot@Tanh@xD, 8x, - 8, 8<D
1.0

0.5

Out[7]=
-5 5

-0.5

-1.0

The haversine function Haversine@zD is defined by sin2 Hz 2L. The inverse haversine function
InverseHaversine@zD is defined by 2 sin-1 I z M. The Gudermannian function Gudermannian@zD is
p
defined as gdHzL = 2 tan-1 Hez L - 2 . The inverse Gudermannian function InverseGudermannian@zD is

defined by gd-1 HzL = log@tanHz 2 + p 4LD. The Gudermannian satisfies such relations as
sinhHzL = tan@gdHxLD. The sinc function Sinc@zD is the Fourier transform of a square signal.

There are a number of additional trigonometric and hyperbolic functions that are sometimes
used. The versine function is sometimes encountered in the literature and simply is
versHzL = 2 havHzL. The coversine function is defined as coversHzL = 1 - sinHzL. The complex exponential
eix is sometimes written as cisHxL.
Mathematics and Algorithms 421

Functions That Do Not Have Unique Values

When you ask for the square root s of a number a, you are effectively asking for the solution to
the equation s2 = a. This equation, however, in general has two different solutions. Both s = 2 and
s = -2 are, for example, solutions to the equation s2 = 4. When you evaluate the "function" 4,
however, you usually want to get a single number, and so you have to choose one of these two
solutions. A standard choice is that x should be positive for x > 0. This is what the Mathemat-
ica function Sqrt@xD does.

The need to make one choice from two solutions means that Sqrt@xD cannot be a true inverse
function for x ^ 2. Taking a number, squaring it, and then taking the square root can give you a
different number than you started with.

4 gives +2, not -2.

In[1]:= Sqrt@4D
Out[1]= 2

Squaring and taking the square root does not necessarily give you the number you started with.
In[2]:= Sqrt@H- 2L ^ 2D
Out[2]= 2

When you evaluate -2 i , there are again two possible answers: -1 + i and 1 - i. In this case,
however, it is less clear which one to choose.

There is in fact no way to choose z so that it is continuous for all complex values of z. There
has to be a "branch cut"~ a line in the complex plane across which the function z is discontinu-
ous. Mathematica adopts the usual convention of taking the branch cut for z to be along the
negative real axis.

This gives 1 - i, not -1 + i.

In[3]:= N@Sqrt@- 2 IDD
Out[3]= 1. - 1.
422 Mathematics and Algorithms

The branch cut in Sqrt along the negative real axis means that values of Sqrt@zD with z just
above and below the axis are very different.
In[4]:= 8Sqrt@- 2 + 0.1 ID, Sqrt@- 2 - 0.1 ID<
Out[4]= 80.0353443 + 1.41466 , 0.0353443 - 1.41466 <

Their squares are nevertheless close.

In[5]:= %^2
Out[5]= 8-2. + 0.1 , -2. - 0.1 <

The discontinuity along the negative real axis is quite clear in this three-dimensional picture of
the imaginary part of the square root function.
In[6]:= Plot3D@Im@Sqrt@x + I yDD, 8x, - 4, 4<, 8y, - 4, 4<D

Out[6]=

When you find an nth root using z1n , there are, in principle, n possible results. To get a single
value, you have to choose a particular principal root. There is absolutely no guarantee that
taking the nth root of an nth power will leave you with the same number.

This takes the tenth power of a complex number. The result is unique.
In[7]:= H2.5 + IL ^ 10
Out[7]= -15 781.2 - 12 335.8

There are 10 possible tenth roots. Mathematica chooses one of them. In this case it is not the
number whose tenth power you took.
In[8]:= % ^ H1 10L
Out[8]= 2.61033 - 0.660446

There are many mathematical functions which, like roots, essentially give solutions to equa-
tions. The logarithm function and the inverse trigonometric functions are examples. In almost
all cases, there are many possible solutions to the equations. Unique "principal" values neverthe-
Mathematics and Algorithms 423

less have to be chosen for the functions. The choices cannot be made continuous over the
whole complex plane. Instead, lines of discontinuity, or branch cuts, must occur. The positions
of these branch cuts are often quite arbitrary. Mathematica makes the most standard mathemati-
cal choices for them.

Sqrt@zD and z^s H-, 0L for Re s > 0, H-, 0D for Re s 0 (s not an integer)
Exp@zD none
Log@zD H-, 0D
trigonometric functions none
ArcSin@zD and ArcCos@zD H-, -1L and H+1, +L
ArcTan@zD H-i, -iD and Hi, iD
ArcCsc@zD and ArcSec@zD H-1, +1L
ArcCot@zD @-i, +iD
hyperbolic functions none
ArcSinh@zD H-i, -iL and H+i, +iL
ArcCosh@zD H-, +1L
ArcTanh@zD H-, -1D and @+1, +L
ArcCsch@zD H-i, iL
ArcSech@zD H-, 0D and H+1, +L
ArcCoth@zD @-1, +1D

Some branch-cut discontinuities in the complex plane.

ArcSin is a multiple-valued function, so there is no guarantee that it always gives the "inverse"
of Sin.
In[9]:= ArcSin@[email protected]
Out[9]= -1.35841

Values of ArcSin@zD on opposite sides of the branch cut can be very different.
In[10]:= 8ArcSin@2 + 0.1 ID, ArcSin@2 - 0.1 ID<
Out[10]= 81.51316 + 1.31888 , 1.51316 - 1.31888 <
424 Mathematics and Algorithms

A three-dimensional picture, showing the two branch cuts for the function sin-1 HzL.
In[11]:= Plot3D@Im@ArcSin@x + I yDD, 8x, - 4, 4<, 8y, - 4, 4<D

Out[11]=

Mathematical Constants

I i= -1
Infinity
Pi p > 3.14159
Degree p 180: degrees to radians conversion factor

GoldenRatio f = J1 + 5 N 2 > 1.61803

E e > 2.71828
EulerGamma Euler's constant g > 0.577216

Catalan Catalan's constant C > 0.915966

Khinchin Khinchin's constant K > 2.68545

Glaisher Glaisher's constant A > 1.28243

Mathematical constants.

1
Eulers constant EulerGamma is given by the limit g = limm Jm
k=1 k
- log mN. It appears in many

integrals, and asymptotic formulas. It is sometimes known as the Euler-Mascheroni constant,

and denoted C.

-2
Catalans constant Catalan is given by the sum
k=0 H-1L H2 k + 1L . It often appears in asymp-
k

totic estimates of combinatorial functions. It is variously denoted by C, K, or G.

Khinchins constant Khinchin (sometimes called Khintchine's constant) is given by

1 log2 s

s=1 J1 + s Hs+2L
N . It gives the geometric mean of the terms in the continued fraction representa-

tion for a typical real number.

Glaishers constant Glaisher A (sometimes called the Glaisher-Kinkelin constant) satisfies

1
log HAL = 12
- z H-1L, where z is the Riemann zeta function. It appears in various sums and inte-
Mathematics and Algorithms 425

Glaishers constant Glaisher A (sometimes called the Glaisher-Kinkelin constant) satisfies

1
log HAL = 12
- z H-1L, where z is the Riemann zeta function. It appears in various sums and inte-

grals, particularly those involving gamma and zeta functions.

Mathematical constants can be evaluated to arbitrary precision.

In[1]:= N@EulerGamma, 40D
Out[1]= 0.5772156649015328606065120900824024310422

Exact computations can also be done with them.

In[2]:= IntegerPart@GoldenRatio ^ 100D
Out[2]= 792 070 839 848 372 253 126

Orthogonal Polynomials

LegendreP@n,xD Legendre polynomials Pn HxL

LegendreP@n,m,xD associated Legendre polynomials Pm

n HxL
m
SphericalHarmonicY@l,m,q,fD spherical harmonics Yl Hq, fL
HmL
GegenbauerC @n,m,xD Gegenbauer polynomials C n HxL

ChebyshevT@n,xD Chebyshev polynomials Tn HxL and Un HxL of the first and

, ChebyshevU@n,xD second kinds
HermiteH@n,xD Hermite polynomials Hn HxL

LaguerreL@n,xD Laguerre polynomials Ln HxL

LaguerreL@n,a,xD generalized Laguerre polynomials Lna HxL

ZernikeR@n,m,xD Zernike radial polynomials RHmL

n HxL

JacobiP@n,a,b,xD Jacobi polynomials PHa,bL

n HxL

Orthogonal polynomials.

Legendre polynomials LegendreP@n, xD arise in studies of systems with three-dimensional

spherical symmetry. They satisfy the differential equation I1 - x2 M y - 2 x y + nHn + 1L y = 0, and the
1
orthogonality relation -1 Pm HxL Pn HxL x = 0 for m n.

The associated Legendre polynomials LegendreP@n, m, xD are obtained from derivatives of the
m2
Legendre polynomials according to Pm m 2
n HxL = H-1L I1 - x M d m @Pn HxLD d xm . Notice that for odd inte-

gers m n, the Pm
n HxL contain powers of 1 - x2 , and are therefore not strictly polynomials. The
m
426 Mathematics and Algorithms

gers m n, the Pm
n HxL contain powers of 1 - x2 , and are therefore not strictly polynomials. The

n HxL reduce to Pn HxL when m = 0.

The spherical harmonics SphericalHarmonicY@l, m, q, fD are related to associated Legendre

m
polynomials. They satisfy the orthogonality relation Yl m Hq, fL Y l Hq, fL w = 0 for l l or m m ,

where d w represents integration over the surface of the unit sphere.

This gives the algebraic form of the Legendre polynomial P8 HxL.

In[1]:= LegendreP@8, xD
1
Out[1]= I35 - 1260 x2 + 6930 x4 - 12 012 x6 + 6435 x8 M
128

1
The integral -1 P7 HxL P8 HxL d x gives zero by virtue of the orthogonality of the Legendre polynomi-
als.
In[2]:= Integrate@LegendreP@7, xD LegendreP@8, xD, 8x, - 1, 1<D
Out[2]= 0

Integrating the square of a single Legendre polynomial gives a nonzero result.

In[3]:= Integrate@LegendreP@8, xD ^ 2, 8x, - 1, 1<D
2
Out[3]=
17

High-degree Legendre polynomials oscillate rapidly.

In[4]:= Plot@LegendreP@10, xD, 8x, - 1, 1<D
1.0
0.8
0.6
Out[4]= 0.4
0.2

-1.0 -0.5 0.5 1.0

-0.2
-0.4

The associated Legendre "polynomials" involve fractional powers.

In[5]:= LegendreP@8, 3, xD
3465
Out[5]= 1 - x2 I-1 + x2 M I3 x - 26 x3 + 39 x5 M
8

"Special Functions" discusses the generalization of Legendre polynomials to Legendre functions,

which can have non-integer degrees.
Mathematics and Algorithms 427

"Special Functions" discusses the generalization of Legendre polynomials to Legendre functions,

which can have non-integer degrees.
In[6]:= [email protected], 0D
Out[6]= 0.268502

Gegenbauer polynomials GegenbauerC @n, m, xD can be viewed as generalizations of the Legen-

dre polynomials to systems with Hm + 2L-dimensional spherical symmetry. They are sometimes
known as ultraspherical polynomials.

GegenbauerC @n, 0, xD is always equal to zero. GegenbauerC @n, xD is however given by the limit
limm0 CnHmL HxL m. This form is sometimes denoted CnH0L HxL.

Series of Chebyshev polynomials are often used in making numerical approximations to func-
tions. The Chebyshev polynomials of the first kind ChebyshevT@n, xD are defined by
Tn Hcos qL = cosHn qL. They are normalized so that Tn H1L = 1. They satisfy the orthogonality relation
1 -12
2
-1 Tm HxL Tn HxL I1 - x M x = 0 for m n. The Tn HxL also satisfy an orthogonality relation under sum-

mation at discrete points in x corresponding to the roots of Tn HxL.

The Chebyshev polynomials of the second kind ChebyshevU@n, zD are defined by

Un Hcos qL = sin@Hn + 1L qD sin q. With this definition, Un H1L = n + 1. The Un satisfy the orthogonality
1 12
relation -1 Um HxL Un HxL I1 - x2 M x = 0 for m n.

The name "Chebyshev" is a transliteration from the Cyrillic alphabet; several other spellings,
such as "Tschebyscheff", are sometimes used.

Hermite polynomials HermiteH@n, xD arise as the quantum-mechanical wave functions for a

harmonic oscillator. They satisfy the differential equation y - 2 x y + 2 n y = 0, and the orthogonal-

ity relation - Hm HxL Hn HxL e-x x = 0 for m n. An alternative form of Hermite polynomials some-
2

times used is Hen HxL = 2-n2 Hn Jx 2 N (a different overall normalization of the Hen HxL is also some-

times used).

The Hermite polynomials are related to the parabolic cylinder functions or Weber functions Dn HxL
2
by Dn HxL = 2-n2 e-x 4
Hn Jx 2 N.
428 Mathematics and Algorithms

This gives the density for an excited state of a quantum-mechanical harmonic oscillator. The
average of the wiggles is roughly the classical physics result.
In[7]:= Plot@HHermiteH@6, xD Exp@- x ^ 2 2DL ^ 2, 8x, - 6, 6<D
25 000

20 000

15 000
Out[7]=
10 000

5000

-6 -4 -2 2 4 6

Generalized Laguerre polynomials LaguerreL@n, a, xD are related to hydrogen atom wave

functions in quantum mechanics. They satisfy the differential equation x y + Ha + 1 - xL y + n y = 0,

and the orthogonality relation 0 Lm
a
HxL Lna HxL xa e-x x = 0 for m n. The Laguerre polynomials

LaguerreL@n, xD correspond to the special case a = 0.

You can get formulas for generalized Laguerre polynomials with arbitrary values of a.
In[8]:= LaguerreL@2, a, xD
1
Out[8]= I2 + 3 a + a2 - 4 x - 2 a x + x2 M
2

Zernike radial polynomials ZernikeR@n, m, xD are used in studies of aberrations in optics. They
HmL 1
n HxL Rk HxL x x = 0 for n k.
satisfy the orthogonality relation 0 RHmL

Jacobi polynomials JacobiP@n, a, b, xD occur in studies of the rotation group, particularly in

1
quantum mechanics. They satisfy the orthogonality relation -1 PHa,bL Ha,bL a b
m HxL Pn HxL H1 - xL H1 + xL x = 0

for m n. Legendre, Gegenbauer, Chebyshev and Zernike polynomials can all be viewed as
special cases of Jacobi polynomials. The Jacobi polynomials are sometimes given in the alterna-
Hp-q,q-1L
tive form Gn Hp, q, xL = n ! G Hn + pL G H2 n + pL Pn H2 x - 1L.

Special Functions
Mathematica includes all the common special functions of mathematical physics found in stan-
dard handbooks. We will discuss each of the various classes of functions in turn.

One point you should realize is that in the technical literature there are often several conflicting
definitions of any particular special function. When you use a special function in Mathematica,
therefore, you should be sure to look at the definition given here to confirm that it is exactly
what you want.
Mathematics and Algorithms 429

Mathematica gives exact results for some values of special functions.

In[1]:= Gamma@15 2D

135 135 p
Out[1]=
128

No exact result is known here.

In[2]:= Gamma@15 7D
15
Out[2]= GammaB F
7

A numerical result, to arbitrary precision, can nevertheless be found.

In[3]:= N@%, 40D
Out[3]= 1.069071500448624397994137689702693267367

You can give complex arguments to special functions.

In[4]:= Gamma@3 + 4 ID N
Out[4]= 0.00522554 - 0.172547

Special functions automatically get applied to each element in a list.

In[5]:= Gamma@83 2, 5 2, 7 2<D

p 3 p 15 p
Out[5]= : , , >
2 4 8

Mathematica knows analytical properties of special functions, such as derivatives.

In[6]:= D@Gamma@xD, 8x, 2<D

Out[6]= Gamma@xD PolyGamma@0, xD2 + Gamma@xD PolyGamma@1, xD

You can use FindRoot to find roots of special functions.

In[7]:= FindRoot@BesselJ@0, xD, 8x, 1<D
Out[7]= 8x 2.40483<

Special functions in Mathematica can usually be evaluated for arbitrary complex values of their
arguments. Often, however, the defining relations given in this tutorial apply only for some
special choices of arguments. In these cases, the full function corresponds to a suitable exten-
sion or "analytic continuation" of these defining relations. Thus, for example, integral representa-
tions of functions are valid only when the integral exists, but the functions themselves can
usually be defined elsewhere by analytic continuation.

As a simple example of how the domain of a function can be extended, consider the function

k=0 x . This sum converges only when

represented by the sum k
x < 1. Nevertheless, it is easy

to show analytically that for any x, the complete function is equal to 1 H1 - xL. Using this form,
you can easily find a value of the function for any x, at least so long as x 1.

Gamma and Related Functions

Beta@a,bD Euler beta function BHa, bL

Beta@z,a,bD incomplete beta function Bz Ha, bL

BetaRegularized@z,a,bD regularized incomplete beta function IHz, a, bL

Gamma@zD Euler gamma function GHzL
Gamma@a,zD incomplete gamma function GHa, zL
Gamma@a,z0 ,z1 D generalized incomplete gamma function GHa, z0 L - GHa, z1 L

GammaRegularized@a,zD regularized incomplete gamma function QHa, zL

InverseBetaRegularized@s,a,bD
inverse beta function
InverseGammaRegularized@a,sD inverse gamma function
Pochhammer@a,nD Pochhammer symbol HaLn

PolyGamma@zD digamma function yHzL

PolyGamma@n,zD nth derivative of the digamma function yHnL HzL

LogGamma@zD Euler log-gamma function log GHzL

LogBarnesG@zD logarithm of Barnes G-function logGHzL

BarnesG@zD Barnes G-function GHzL

Hyperfactorial@nD hyperfactorial function HHnL

Gamma and related functions.

The Euler gamma function Gamma@zD is defined by the integral GHzL = 0 tz-1 e-t d t. For positive

integer n, GHnL = Hn - 1L ! . GHzL can be viewed as a generalization of the factorial function, valid for
complex arguments z.
Mathematics and Algorithms 431

The Euler gamma function Gamma@zD is defined by the integral GHzL = 0 tz-1 e-t d t. For positive

integer n, GHnL = Hn - 1L ! . GHzL can be viewed as a generalization of the factorial function, valid for
complex arguments z.

There are some computations, particularly in number theory, where the logarithm of the
gamma function often appears. For positive real arguments, you can evaluate this simply as
Log@Gamma@zDD. For complex arguments, however, this form yields spurious discontinuities.
Mathematica therefore includes the separate function LogGamma@zD, which yields the logarithm
of the gamma function with a single branch cut along the negative real axis.

1
The Euler beta function Beta@a, bD is BHa, bL = GHaL GHbL GHa + bL = 0 ta-1 H1 - tLb-1 d t.

The Pochhammer symbol or rising factorial Pochhammer@a, nD is

HaLn = aHa + 1L Ha + n - 1L = GHa + nL GHaL. It often appears in series expansions for hypergeometric
functions. Note that the Pochhammer symbol has a definite value even when the gamma func-
tions which appear in its definition are infinite.

The incomplete gamma function Gamma@a, zD is defined by the integral GHa, zL = z ta-1 e-t d t.

Mathematica includes a generalized incomplete gamma function Gamma@a, z0 , z1 D defined as

z 1 a-1 -t
z0 t e d t.

The alternative incomplete gamma function gHa, zL can therefore be obtained in Mathematica as
Gamma@a, 0, zD.
z
The incomplete beta function Beta@z, a, bD is given by Bz Ha, bL = 0 ta-1 H1 - tLb-1 d t. Notice that in

the incomplete beta function, the parameter z is an upper limit of integration, and appears as
the first argument of the function. In the incomplete gamma function, on the other hand, z is a
lower limit of integration, and appears as the second argument of the function.

In certain cases, it is convenient not to compute the incomplete beta and gamma functions on
their own, but instead to compute regularized forms in which these functions are divided by
complete beta and gamma functions. Mathematica includes the regularized incomplete beta
function BetaRegularized@z, a, bD defined for most arguments by IHz, a, bL = BHz, a, bL BHa, bL, but
taking into account singular cases. Mathematica also includes the regularized incomplete
gamma function GammaRegularized@a, zD defined by QHa, zL = GHa, zL GHaL, with singular cases
taken into account.

The incomplete beta and gamma functions, and their inverses, are common in statistics. The
inverse beta function InverseBetaRegularized@s, a, bD is the solution for z in s = IHz, a, bL. The
inverse gamma function InverseGammaRegularized@a, sD is similarly the solution for z in
432 Mathematics and Algorithms

Derivatives of the gamma function often appear in summing rational series. The digamma
function PolyGamma@zD is the logarithmic derivative of the gamma function, given by
yHzL = G HzL GHzL. For integer arguments, the digamma function satisfies the relation
yHnL = -g + Hn-1 , where g is Euler's constant (EulerGamma in Mathematica) and Hn are the har-
monic numbers.

The polygamma functions PolyGamma@n, zD are given by yHnL HzL = d n yHzL d zn . Notice that the
digamma function corresponds to yH0L HzL. The general form yHnL HzL is the Hn + 1Lth , not the nth ,
logarithmic derivative of the gamma function. The polygamma functions satisfy the relation
yHnL HzL = H-1Ln+1 n !
k=0 1 Hz + kL
n+1
. PolyGamma@n, zD is defined for arbitrary complex n by fractional

calculus analytic continuation.

BarnesG@zD is a generalization of the Gamma function and is defined by its functional identity
BarnesG@z + 1D = Gamma@zD BarnesG@zD, where the third derivative of the logarithm of BarnesG
is positive for positive z. BarnesG is an entire function in the complex plane.

LogBarnesG@zD is a holomorphic function with a branch cut along the negative real-axis such
that Exp@LogBarnesG@zDD = BarnesG@zD.

Hyperfactorial@nD is a generalization of nk=1 kk to the complex plane.

Many exact results for gamma and polygamma functions are built into Mathematica.
In[1]:= PolyGamma@6D
137
Out[1]= - EulerGamma
60

Here is a contour plot of the gamma function in the complex plane.

In[2]:= ContourPlot@Abs@Gamma@x + I yDD, 8x, - 3, 3<, 8y, - 2, 2<, PlotPoints -> 50D
Mathematics and Algorithms 433

Zeta and Related Functions

DirichletL@k, j,sD Dirichlet L-function LH c, sL

LerchPhi@z,s,aD Lerch's transcendent FHz, s, aL

PolyLog@n,zD polylogarithm function Lin HzL

PolyLog@n,p,zD Nielsen generalized polylogarithm function Sn,p HzL

RamanujanTau@nD Ramanujan t function t HnL

RamanujanTauL@nD Ramanujan t Dirichlet L function L HsL
RamanujanTauTheta@nD Ramanujan t theta function q HtL
RamanujanTauZ@nD Ramanujan t Z function Z HtL
RiemannSiegelTheta@tD Riemann|Siegel function JHtL
RiemannSiegelZ@tD Riemann|Siegel function ZHtL
StieltjesGamma@nD Stieltjes constants gn

Zeta@sD Riemann zeta function zHsL

Zeta@s,aD generalized Riemann zeta function zHs, aL
HurwitzZeta @s,aD Hurwitz zeta function zHs, aL
HurwitzLerchPhi@z,s,aD Hurwitz|Lerch transcendent FHz, s, aL

Zeta and related functions.

The Dirichlet-L function DirichletL@k, j, sD is implemented as LH c, sL =

n=1 cHnL n
-s
(for Re HsL > 1)

where cHnL is a Dirichlet character with modulus k and index j.

The Riemann zeta function Zeta@sD is defined by the relation z HsL =

k=1 k
-s
(for s > 1). Zeta func-

tions with integer arguments arise in evaluating various sums and integrals. Mathematica gives
exact results when possible for zeta functions with integer arguments.

There is an analytic continuation of zHsL for arbitrary complex s 1. The zeta function for complex
arguments is central to number theoretic studies of the distribution of primes. Of particular
1
importance are the values on the critical line Re HsL = 2 .

1
In studying zJ 2 + i tN, it is often convenient to define the two Riemann|Siegel functions
1
RiemannSiegelZ@tD and RiemannSiegelTheta@tD according to ZHtL = ei J HtL zJ 2 + i tN and
1
JHtL = Im log GJ 4 + i t 2N - t logHpL 2 (for t real). Note that the Riemann|Siegel functions are both real

as long as t is real.

The Stieltjes constants StieltjesGamma@nD are generalizations of Euler's constant which appear
in the series expansion of zHsL around its pole at s = 1; the coefficient of H1 - sLn is gn n !. Euler's
434 Mathematics and Algorithms

-s2
The generalized Riemann zeta function Zeta@s, aD is implemented as zHs, aL = 2
k=0 IHk + aL M ,

where any term with k + a = 0 is excluded.

-s
The Hurwitz zeta function HurwitzZeta @s, aD is implemented as zHs, aL =
k=0 Hk + aL .

tHnL
The Ramanujan t Dirichlet L function RamanujanTauL@sD is defined by L HsL =
n=1 ns
(for

Re HsL > 6), with coefficients RamanujanTau@nD. In analogy with the Riemann zeta function, it is
again convenient to define the functions RamanujanTauZ@tD and RamanujanTauTheta@tD.

Here is the numerical approximation for LH6, 2, 1.0 + L.

In[77]:= DirichletL@6, 2, 1. + D
Out[77]= 0.978008 + 0.0954731

Here is a three-dimensional picture of the real part of a Dirichlet L-function.

In[4]:= Plot3D@Re DirichletL@6, 2, u + vD, 8u, - 5, 5<, 8v, - 10, 10<, Mesh None,
PlotStyle Directive@[email protected], Green, Specularity@10DD, BoxRatios 81, 1, 2<D
105
0
10 0
5

10
Out[4]=
0

Mathematica gives exact results for z H2 nL.

In[1]:= Zeta@6D
p6
Out[1]=
945
Mathematics and Algorithms 435

Here is a three-dimensional picture of the Riemann zeta function in the complex plane.
In[14]:= Plot3D@Abs@Zeta@x + I yDD, 8x, - 3, 3<, 8y, 2, 35<D
Out[14]=

1
This is a plot of the absolute value of the Riemann zeta function on the critical line Re z = . You
2
can see the first few zeros of the zeta function.
In[15]:= Plot@Abs@Zeta@1 2 + I yDD, 8y, 0, 40<D
Out[15]= 3.0

2.5

2.0

1.5

1.0

0.5

10 20 30 40

This is a plot of the absolute value of the Ramanujan t L function on its critical line Re z = 6.
In[4]:= Plot@Abs@RamanujanTauL@6 + I yDD, 8y, 0, 20<D
Out[4]=
2.0

1.5

1.0

0.5

5 10 15 20

The polylogarithm functions PolyLog@n, zD are given by Lin HzL =

k=1 z k . The polylogarithm
k n

function is sometimes known as Jonquire's function. The dilogarithm PolyLog@2, zD satisfies

0
Li2 HzL = z logH1 - tL t t. Sometimes Li2 H1 - zL is known as Spence's integral. The Nielsen general-

ized polylogarithm functions or hyperlogarithms PolyLog@n, p, zD are given by

1
Sn,p HzL = H-1Ln+p-1 HHn - 1L ! p !L 01logn-1 HtL log p H1 - ztL t t. Polylogarithm functions appear in Feynman

diagram integrals in elementary particle physics, as well as in algebraic K-theory.

0
Li2 HzL = z logH1 - tL t t. Sometimes Li2 H1 - zL is known as Spence's integral. The Nielsen general-
436 Mathematics and Algorithms

1
Sn,p HzL = H-1Ln+p-1 HHn - 1L ! p !L 0 logn-1 HtL log p H1 - ztL t t. Polylogarithm functions appear in Feynman

diagram integrals in elementary particle physics, as well as in algebraic K-theory.

The Lerch transcendent LerchPhi@z, s, aD is a generalization of the zeta and polylogarithm

s2
functions, given by FHz, s, aL =
k=0 z IHa + kL M , where any term with a + k = 0 is excluded. Many
k 2

sums of reciprocal powers can be expressed in terms of the Lerch transcendent. For example,
-s 1
the Catalan beta function bHsL = k
k=0 H-1L H2 k + 1L can be obtained as 2-s F J-1, s, 2 N.

The Lerch transcendent is related to integrals of the Fermi|Dirac distribution in statistical

mechanics by 0 ks Iek-m + 1M k = e m GHs + 1L FH-e m , s + 1, 1L.

The Lerch transcendent can also be used to evaluate Dirichlet L-series which appear in number
theory. The basic L-series has the form LHs, cL =
k=1 cHkL k , where the "character" cHkL is an
-s

integer function with period m. L-series of this kind can be written as sums of Lerch functions
with z a power of e2 p im .

s2
LerchPhi@z, s, a, DoublyInfinite -> TrueD gives the doubly infinite sum
k=- z IHa + kL M .
k 2

The Hurwitz|Lerch transcendent HurwitzLerchPhi@z, s, aD generalizes HurwitzZeta @s, aD and

is defined by FHz, s, aL = k s
k=0 z HHa + kL L.

ZetaZero@kD the kth zero of the zeta function zHzL on the critical line

ZetaZero@k,x0 D the kth zero above height x0

Zeros of the zeta function.

ZetaZero@1D represents the first nontrivial zero of z HzL.

In[1]:= Zeta@ZetaZero@1DD
Out[1]= 0

This gives its numerical value.

In[2]:= N@ZetaZero@1DD
Out[2]= 0.5 + 14.1347
Mathematics and Algorithms 437

This gives the first zero with height greater than 15.
In[3]:= N@ZetaZero@1, 15DD
Out[3]= 0.5 + 21.022

Exponential Integral and Related Functions

CosIntegral @zD cosine integral function CiHzL

CoshIntegral@zD hyperbolic cosine integral function ChiHzL
ExpIntegralE@n,zD exponential integral En HzL

ExpIntegralEi@zD exponential integral EiHzL

LogIntegral @zD logarithmic integral liHzL
SinIntegral @zD sine integral function SiHzL
SinhIntegral@zD hyperbolic sine integral function ShiHzL

Exponential integral and related functions.

Mathematica has two forms of exponential integral: ExpIntegralE and ExpIntegralEi.

The exponential integral function ExpIntegralE@n, zD is defined by En HzL = 1 e-z t tn t.

The second exponential integral function ExpIntegralEi@zD is defined by EiHzL = --z e-t t t (for

z > 0), where the principal value of the integral is taken.

z
The logarithmic integral function LogIntegral @zD is given by liHzL = 0 d t log t (for z > 1), where

the principal value of the integral is taken. liHzL is central to the study of the distribution of
primes in number theory. The logarithmic integral function is sometimes also denoted by LiHzL.
z
In some number theoretic applications, liHzL is defined as 2 d t log t, with no principal value taken.

This differs from the definition used in Mathematica by the constant liH2L.

The sine and cosine integral functions SinIntegral @zD and CosIntegral @zD are defined by
z
SiHzL = 0 sinHtL t t and CiHzL = -z cosHtL t t. The hyperbolic sine and cosine integral functions
z
SinhIntegral@zD and CoshIntegral@zD are defined by ShiHzL = 0 sinhHtL t t and
z
ChiHzL = g + logHzL + 0 HcoshHtL - 1L t t.
438 Mathematics and Algorithms

Error Function and Related Functions

Erf@zD error function erfHzL

Erf@z0 ,z1 D generalized error function erfHz1 L - erfHz0 L

Erfc@zD complementary error function erfcHzL

Erfi@zD imaginary error function erfiHzL
FresnelC@zD Fresnel integral C HzL
FresnelS@zD Fresnel integral SHzL
InverseErf@sD inverse error function
InverseErfc @sD inverse complementary error function

Error function and related functions.

The error function Erf@zD is the integral of the Gaussian distribution, given by
z -t2
erfHzL = 2 p 0 e d t. The complementary error function Erfc@zD is given simply by

erfcHzL = 1 - erfHzL. The imaginary error function Erfi@zD is given by erfiHzL = erfHizL i. The generalized
z
error function Erf@z0 , z1 D is defined by the integral 2 p z 1 e-t d t. The error function is central
2

to many calculations in statistics.

The inverse error function InverseErf@sD is defined as the solution for z in the equation
s = erfHzL. The inverse error function appears in computing confidence intervals in statistics as
well as in some algorithms for generating Gaussian random numbers.

Closely related to the error function are the Fresnel integrals FresnelC@zD defined by
z z
CHzL = 0 cosIp t2 2M d t and FresnelS@zD defined by SHzL = 0 sinIp t2 2M d t. Fresnel integrals occur in

diffraction theory.

Bessel and Related Functions

AiryAi@zD and AiryBi@zD Airy functions AiHzL and BiHzL

AiryAiPrime @zD and AiryBiPrime @zD
derivatives of Airy functions Ai HzL and Bi HzL
BesselJ@n,zD and BesselY@n,zD
Bessel functions Jn HzL and Yn HzL
Mathematics and Algorithms 439

BesselI@n,zD and BesselK@n,zD

modified Bessel functions In HzL and Kn HzL

KelvinBer@n,zD and KelvinBei@n,zD

Kelvin functions bern HzL and bein HzL

KelvinKer@n,zD and KelvinKei@n,zD

Kelvin functions kern HzL and kein HzL

HankelH1@n,zD and HankelH2@n,zD

Hankel functions HnH1L HzL and HnH2L HzL

SphericalBesselJ@n,zD and SphericalBesselY@n,zD

spherical Bessel functions jn HzL and yn HzL

SphericalHankelH1@n,zD and SphericalHankelH2@n,zD

spherical Hankel functions hH1L H2L
n HzL and hn HzL

StruveH@n,zD and StruveL@n,zD

Struve function Hn HzL and modified Struve function Ln HzL

Bessel and related functions.

The Bessel functions BesselJ@n, zD and BesselY@n, zD are linearly independent solutions to the
differential equation z2 y + z y + Iz2 - n2 M y = 0. For integer n, the Jn HzL are regular at z = 0, while the

Yn HzL have a logarithmic divergence at z = 0.

Bessel functions arise in solving differential equations for systems with cylindrical symmetry.

Jn HzL is often called the Bessel function of the first kind, or simply the Bessel function. Yn HzL is
referred to as the Bessel function of the second kind, the Weber function, or the Neumann
function (denoted Nn HzL).

The Hankel functions (or Bessel functions of the third kind) HankelH1@n, zD and HankelH2@n, zD
give an alternative pair of solutions to the Bessel differential equation, related according to
HnH1,2L HzL = Jn HzL iYn HzL.

The spherical Bessel functions SphericalBesselJ@n, zD and SphericalBesselY@n, zD, as well

as the spherical Hankel functions SphericalHankelH1@n, zD and SphericalHankelH2@n, zD,
arise in studying wave phenomena with spherical symmetry. These are related to the ordinary

fn HzL = p2 z F 1 HzL, where f and F can be j and J, y and Y, or hi and H i . For

n+
2

integer n, spherical Bessel functions can be expanded in terms of elementary functions by using
FunctionExpand.
The
The spherical
spherical Bessel
Bessel functions
functions SphericalBesselJ@n, zD and
SphericalBesselJ@n, zD and SphericalBesselY@n, zD, as
SphericalBesselY@n, zD, as well
well
as
as the
the spherical
spherical Hankel
Hankel functions
functions SphericalHankelH1@n, zD and
SphericalHankelH1@n, zD and SphericalHankelH2@n,
SphericalHankelH2@n, zD
zD,
440 Mathematics and Algorithms

functions by fn HzL = p2 z F 1 HzL, where f and F can be j and J, y and Y, or hi and H i . For
n+
2

integer n, spherical Bessel functions can be expanded in terms of elementary functions by using
FunctionExpand.

The modified Bessel functions BesselI@n, zD and BesselK@n, zD are solutions to the differential
equation z2 y + z y - Iz2 + n2 M y = 0. For integer n, In HzL is regular at z = 0; Kn HzL always has a logarith-

mic divergence at z = 0. The In HzL are sometimes known as hyperbolic Bessel functions.

Particularly in electrical engineering, one often defines the Kelvin functions KelvinBer@n, zD,
KelvinBei@n, zD, KelvinKer@n, zD and KelvinKei@n, zD. These are related to the ordinary
Bessel functions by bern HzL + i bein HzL = en p i Jn Iz e-p i4 M, kern HzL + i kein HzL = e-n p i2 Kn Iz ep i4 M.

The Airy functions AiryAi@zD and AiryBi@zD are the two independent solutions AiHzL and BiHzL to
the differential equation y - z y = 0. AiHzL tends to zero for large positive z, while BiHzL increases
unboundedly. The Airy functions are related to modified Bessel functions with one-third-integer
orders. The Airy functions often appear as the solutions to boundary value problems in electro-
magnetic theory and quantum mechanics. In many cases the derivatives of the Airy functions
AiryAiPrime @zD and AiryBiPrime @zD also appear.

The Struve function StruveH@n, zD appears in the solution of the inhomogeneous Bessel equa-
2 zn+1
tion which for integer n has the form z2 y + z y + Iz2 - n2 M y = p H2 n-1L!!
; the general solution to this

equation consists of a linear combination of Bessel functions with the Struve function Hn HzL
added. The modified Struve function StruveL@n, zD is given in terms of the ordinary Struve
function by Ln HzL = -i e-i n p2 Hn HzL. Struve functions appear particularly in electromagnetic theory.

Here is a plot of J0 I x M. This is a curve that an idealized chain hanging from one end can form
when you wiggle it.
In[16]:= Plot@BesselJ@0, Sqrt@xDD, 8x, 0, 50<D
Out[16]= 1.0

0.8

0.6

0.4

0.2

10 20 30 40 50
-0.2

-0.4
Mathematics and Algorithms 441

Mathematica generates explicit formulas for half-integer-order Bessel functions.

In[2]:= BesselK@3 2, xD

p 1
-x J1 + N
2 x
Out[2]=
x

The Airy function plotted here gives the quantum-mechanical amplitude for a particle in a
potential that increases linearly from left to right. The amplitude is exponentially damped in the
classically inaccessible region on the right.
In[17]:= Plot@AiryAi@xD, 8x, - 10, 10<D
Out[17]=
0.4

0.2

-10 -5 5 10

-0.2

-0.4

BesselJZero @n,kD the kth zero of the Bessel function Jn HzL

BesselJZero @n,k,x0 D the kth zero greater than x0

BesselYZero @n,kD the kth zero of the Bessel function Yn HzL

BesselYZero @n,k,x0 D the kth zero greater than x0

AiryAiZero@kD the kth zero of the Airy function Ai HzL

AiryAiZero@k,x0 D the kth zero less than x0

AiryBiZero@kD the kth zero of the Airy function Bi HzL

AiryBiZero@k,x0 D the kth zero less than x0

Zeros of Bessel and Airy functions.

BesselJZero @1, 5D represents the fifth zero of J1 HzL.

In[18]:= BesselJ@1, BesselJZero@1, 5DD
Out[18]= 0

This gives its numerical value.

In[19]:= N@BesselJZero@1, 5DD
Out[19]= 16.4706

Legendre and Related Functions

442 Mathematics and Algorithms

Spheroidal Functions

SpheroidalS1@n,m,g,zD and SpheroidalS2@n,m,g,zD

H1L H2L
radial spheroidal functions Sn,m Hg, zL and Sn,m Hg, zL
SpheroidalS1Prime@n,m,g,zD and SpheroidalS2Prime@n,m,g,zD
z derivatives of radial spheroidal functions
SpheroidalPS@n,m,g,zD and SpheroidalQS@n,m,g,zD
angular spheroidal functions PSn,m Hg, zL and QSn,m Hg, zL

SpheroidalPSPrime@n,m,g,zD and SpheroidalQSPrime@n,m,g,zD

z derivatives of angular spheroidal functions
SpheroidalEigenvalue@n,m,gD spheroidal eigenvalue of degree n and order m

Spheroidal functions.

The radial spheroidal functions SpheroidalS1@n, m, g, zD and SpheroidalS2@n, m, g, zD and

angular spheroidal functions SpheroidalPS@n, m, g, zD and SpheroidalQS@n, m, g, zD appear
in solutions to the wave equation in spheroidal regions. Both types of functions are solutions to
m2
the equation I1 - z2 M y - 2 zy + Jl + g2 I1 - z2 M - N y 0. This equation has normalizable solutions
1-z2

only when l is a spheroidal eigenvalue given by SpheroidalEigenvalue@n, m, gD. The

spheroidal functions also appear as eigenfunctions of finite analogs of Fourier transforms.

SpheroidalS1 and SpheroidalS2 are effectively spheroidal analogs of the spherical Bessel
functions jn HzL and yn HzL, while SpheroidalPS and SpheroidalQS are effectively spheroidal
analogs of the Legendre functions Pm
n HzL and Qn HzL. g > 0 corresponds to a prolate spheroidal
m 2

geometry, while g2 < 0 corresponds to an oblate spheroidal geometry.

function g z range name

PSn,m Hg, hL QSn,m Hg, hL g h -1 h 1 angular prolate
H1L H2L
Sn,m Hg, zL Sn,m Hg, zL g z z 1 radial prolate
PSn,m H- g, hL QSn,m H- g, hL - g h -1 h 1 angular oblate
H1L H2L
Sn,m H- g, zL Sn,m H- g, zL - g z z 0 radial oblate

Many different normalizations for spheroidal functions are used in the literature. Mathematica
uses the Meixner|Schfke normalization scheme.
Mathematics and Algorithms 443

Angular spheroidal functions can be viewed as deformations of Legendre functions.

In[1]:= Series@SpheroidalPS@n, 0, g, hD, 8g, 0, 3<D
H-1 + nL n LegendreP@-2 + n, hD H1 + nL H2 + nL LegendreP@2 + n, hD
Out[1]= LegendreP@n, hD + - + g2 + O@gD3
2 H-1 + 2 nL2 H1 + 2 nL 2 H1 + 2 nL H3 + 2 nL2

This plots angular spheroidal functions for various spheroidicity parameters.

In[2]:= Plot@8SpheroidalPS@3, 0, 0, hD,
SpheroidalPS@3, 0, 3, hD, SpheroidalPS@3, 0, 5, hD<, 8h, - 1, 1<D
1.0

0.5

Out[2]=
-1.0 -0.5 0.5 1.0

-0.5

-1.0

Angular spheroidal functions PSn,0 Hg, hL for integers n 0 are eigenfunctions of a band-limited
Fourier transform.
In[3]:= Integrate@SpheroidalPS@3, 0, g, hD Exp@ w g hD, 8h, - 1, 1<D
Out[3]= -2 SpheroidalPS@3, 0, g, wD SpheroidalS1@3, 0, g, 1D

The Mathieu functions are a special case of spheroidal functions.

1
An angular spheroidal function with m = gives Mathieu angular functions.
2
In[4]:= SpheroidalPS@1 2, 1 2, c, zD
c2 c2 14
Out[4]= MathieuCBMathieuCharacteristicAB1, F, , ArcCos@zDF K p I1 - z2 M O
4 4

Legendre and Related Functions

LegendreP@n,zD Legendre functions of the first kind Pn HzL

LegendreP@n,m,zD associated Legendre functions of the first kind Pm

n HzL

LegendreQ@n,zD Legendre functions of the second kind Qn HzL

LegendreQ@n,m,zD associated Legendre functions of the second kind Qm

n HzL

Legendre and related functions.

The Legendre functions and associated Legendre functions satisfy the differential equation
I1 - z2 M y - 2 z y + An Hn + 1L - m2 I1 - z2 ME y = 0. The Legendre functions of the first kind,

LegendreP@n, zD and LegendreP@n, m, zD, reduce to Legendre polynomials when n and m are
integers. The Legendre functions of the second kind LegendreQ@n, zD and LegendreQ@n, m, zD
give the second linearly independent solution to the differential equation. For integer m they
have logarithmic singularities at z = 1. The Pn HzL and Qn HzL solve the differential equation with
m = 0.

Legendre functions arise in studies of quantum-mechanical scattering processes.

LegendreP@n,m,zD or LegendreP@n,m,1,zD
m2
type 1 function containing I1 - z2 M

LegendreP@n,m,2,zD type 2 function containing H1 + zLm2 H1 - zLm2

LegendreP@n,m,3,zD type 3 function containing H1 + zLm2 H-1 + zLm2

Types of Legendre functions. Analogous types exist for LegendreQ.

Legendre functions of type 1 and Legendre functions of type 2 have different symbolic forms,
but the same numerical values. They have branch cuts from - to -1 and from +1 to +. Legen-
dre functions of type 3, sometimes denoted m
n HzL and n HzL, have a single branch cut from - to
m

+1.

Toroidal functions or ring functions, which arise in studying systems with toroidal symmetry,
m m
can be expressed in terms of the Legendre functions P 1 HcoshhL and Q 1 HcoshhL.
n- n-
2 2

m m
Conical functions can be expressed in terms of P 1 HcosqL and Q 1 HcosqL.
- +i p - +i p
2 2

When you use the function LegendreP@n, xD with an integer n, you get a Legendre polynomial.
If you take n to be an arbitrary complex number, you get, in general, a Legendre function.

In the same way, you can use the functions GegenbauerC and so on with arbitrary complex
indices to get Gegenbauer functions, Chebyshev functions, Hermite functions, Jacobi functions
and Laguerre functions. Unlike for associated Legendre functions, however, there is no need to
distinguish different types in such cases.
Mathematics and Algorithms 445

Hypergeometric Functions and Generalizations

Hypergeometric0F1@a,zD hypergeometric function 0 F1 H; a; zL

Hypergeometric0F1Regularized@a,zD
regularized hypergeometric function 0 F1 H; a; zL GHaL
Hypergeometric1F1@a,b,zD Kummer confluent hypergeometric function 1 F1 Ha; b; zL
Hypergeometric1F1Regularized@a,b,zD
regularized confluent hypergeometric function
1 F1 Ha; b; zL GHbL
HypergeometricU@a,b,zD confluent hypergeometric function UHa, b, zL
WhittakerM@k,m,zD and WhittakerW@k,m,zD
Whittaker functions Mk,m HzL and Wk,m HzL

ParabolicCylinderD@n,zD parabolic cylinder function Dn HzL

Confluent hypergeometric functions and related functions.

Many of the special functions that we have discussed so far can be viewed as special cases of
the confluent hypergeometric function Hypergeometric1F1@a, b, zD.

The confluent hypergeometric function can be obtained from the series expansion

k=0 HaLk HbLk z k ! .

F1 Ha; b; zL = 1 + az b + a Ha + 1L b Hb + 1L z2 2 ! + = Some special results are
k
1

obtained when a and b are both integers. If a < 0, and either b > 0 or b < a, the series yields a
polynomial with a finite number of terms.

If b is zero or a negative integer, then 1 F1 Ha; b; zL itself is infinite. But the regularized confluent
hypergeometric function Hypergeometric1F1Regularized@a, b, zD given by 1 F1 Ha; b; zL GHbL has
a finite value in all cases.

Among the functions that can be obtained from 1 F1 are the Bessel functions, error function,
incomplete gamma function, and Hermite and Laguerre polynomials.

The function 1 F1 Ha; b; zL is sometimes denoted FHa; b; zL or MHa, b, zL. It is often known as the
Kummer function.

The 1 F1 function can be written in the integral representation

1
1 F1 Ha; b; zL = GHbL @GHb - aL GHaLD 0 ezt ta-1 H1 - tLb-a-1 d t.

The 1 F1 confluent hypergeometric function is a solution to Kummers differential equation

z y + Hb - zL y - a y = 0, with the boundary conditions 1 F1 Ha; b; 0L = 1 and @1 F1 Ha; b; zLD z

z=0 = a b.
446 Mathematics and Algorithms

The 1 F1 confluent hypergeometric function is a solution to Kummers differential equation

z y + Hb - zL y - a y = 0, with the boundary conditions 1 F1 Ha; b; 0L = 1 and @1 F1 Ha; b; zLD z z=0 = a b.

The function HypergeometricU@a, b, zD gives a second linearly independent solution to Kum-

mer's equation. For Re b > 1 this function behaves like z1-b for small z. It has a branch cut along
the negative real axis in the complex z plane.

The function UHa, b, zL has the integral representation UHa, b, zL = 1 GHaL 0 e-z t ta-1 H1 + tLb-a-1 d t.

UHa, b, zL, like 1 F1 Ha; b; zL, is sometimes known as the Kummer function. The U function is some-
times denoted by Y.

The Whittaker functions WhittakerM@k, m, zD and WhittakerW@k, m, zD give a pair of solutions

to the normalized Kummer differential equation, known as Whittaker's differential equation. The
1
Whittaker function Mk, m is related to 1 F1 by Mk, m HzL = e-z2 z12+m 1 F1 J 2 + m - k; 1 + 2 m; zN. The second

Whittaker function Wk, m obeys the same relation, with 1 F1 replaced by U.

The parabolic cylinder functions ParabolicCylinderD@n, zD are related to the Hermite func-

tions by Dn HzL = 2-n2 e-Hz2L Hn Jz

2
2 N.

The Coulomb wave functions are also special cases of the confluent hypergeometric function.
Coulomb wave functions give solutions to the radial Schrdinger equation in the Coulomb
potential of a point nucleus. The regular Coulomb wave function is given by
FL Hh, rL = CL HhL r L+1 -i r
e 1 F1 HL + 1 - i h; 2 L + 2; 2 i rL, where CL HhL = 2 e L -ph2
GHL + 1 + i hL GH2 L + 2L.

Other special cases of the confluent hypergeometric function include the Toronto functions
THm, n, rL, Poisson|Charlier polynomials rn Hn, xL, Cunningham functions wn,m HxL and Bateman func-
tions kn HxL.

A limiting form of the confluent hypergeometric function which often appears is

Hypergeometric0F1@a, zD. This function is obtained as the limit 0 F1 H; a; zL = limq 1 F1 Hq; a; z qL.

The 0 F1 function has the series expansion 0 k=0 1 HaLk z k ! and satisfies the differential
F1 H; a; zL = k

equation z y + a y - y = 0.

Bessel functions of the first kind can be expressed in terms of the 0 F1 function.
Mathematics and Algorithms 447

Hypergeometric2F1@a,b,c,zD hypergeometric function 2 F1 Ha, b; c; zL

Hypergeometric2F1Regularized@a,b,c,zD
regularized hypergeometric function 2 F1 Ha, b; c; zL GHcL
HypergeometricPFQA9a1 ,,a p =,9b1 ,,bq =,zE
generalized hypergeometric function p Fq Ha; b; zL

HypergeometricPFQRegularizedA9a1 ,,a p =,9b1 ,,bq =,zE

regularized generalized hypergeometric function
MeijerGA98a1 ,,an <,9an+1 ,,a p ==,98b1 ,,bm <,9bm+1 ,,bq ==,zE
Meijer G function
AppellF1@a,b1 ,b2 ,c,x,yD Appell hypergeometric function of two variables
F1 Ha; b1 , b2 ; c; x, yL

Hypergeometric functions and generalizations.

The hypergeometric function Hypergeometric2F1@a, b, c, zD has series expansion

2 F1 Ha, b; c; zL =
k=0 HaLk HbLk HcLk zk k ! . The function is a solution of the hypergeometric differential

equation zH1 - zL y + @c - Ha + b + 1L zD y - a b y = 0.

The hypergeometric function can also be written as an integral: 2 F1 Ha, b; c; zL = GHcL @GHbL GHc - bLD
1 b-1
0 t H1 - tLc-b-1 H1 - t zL-a d t.

The hypergeometric function is also sometimes denoted by F, and is known as the Gauss series
or the Kummer series.

The Legendre functions, and the functions which give generalizations of other orthogonal polyno-
mials, can be expressed in terms of the hypergeometric function. Complete elliptic integrals can
also be expressed in terms of the 2 F1 function.

The Riemann P function, which gives solutions to Riemann's differential equation, is also a 2 F1
function.
448 Mathematics and Algorithms

The generalized hypergeometric function or Barnes extended hypergeometric

function HypergeometricPFQ@8a1 , , a p <, 8b1 , , bq <, zD has series expansion

Ha1 Lk ... Ha p Lk zk
p Fq Ha; b; zL = Hb .
1 Lk ... Hbq Lk k!
k=0

The Meijer G function MeijerGA98a1 , , an <, 9an + 1, , a p ==, 88b1 , , bm <, 8bm + 1, ,

bq <<, zD is defined by the contour integral representation

a1 , , a p 1
Gmp qn z = GH1 - a1 - sL GH1 - an - sL GHb1 + sL GHbm + sL IGHan+1 + sL GIa p + sM
b1 , , bq 2pi

GH1 - bm+1 - sL GI1 - bq - sMM z-s ds, where the contour of integration is set up to lie between the

poles of GH1 - ai - sL and the poles of GHbi + sL. MeijerG is a very general function whose special
cases cover most of the functions discussed in the past few sections.

The Appell hypergeometric function of two variables AppellF1@a, b1 , b2 , c, x, yD has series

expansion F1 Ha; b1 , b2 ; c; x, yL =
m=0 n=0 HaLm+n Hb1 Lm Hb2 Ln Hm ! n ! HcLm+n L x y . This function appears for
m n

example in integrating cubic polynomials to arbitrary powers.

The Product Log Function

ProductLog@zD product log function WHzL

The product log function.

The product log function gives the solution for w in z = w ew . The function can be viewed as a
generalization of a logarithm. It can be used to represent solutions to a variety of transcenden-
tal equations. The tree generating function for counting distinct oriented trees is related to the
product log by THzL = -WH-zL.
Mathematics and Algorithms 449

Elliptic Integrals and Elliptic Functions

Even more so than for other special functions, you need to be very careful about the arguments
you give to elliptic integrals and elliptic functions. There are several incompatible conventions in
common use, and often these conventions are distinguished only by the specific names given to
arguments or by the presence of separators other than commas between arguments.

Amplitude f (used by Mathematica, in radians)

Argument u (used by Mathematica): related to amplitude by f = amHuL

Delta amplitude DHfL: DHfL = 1 - m sin2 HfL

Coordinate x: x = sinHfL
Characteristic n (used by Mathematica in elliptic integrals of the third kind)
Parameter m (used by Mathematica): preceded by , as in IHf mL
Complementary parameter m1 : m1 = 1 - m
Modulus k: preceded by comma, as in IHf, kL; m = k2
Modular angle a: preceded by \ , as in IHf \ aL; m = sin2 HaL
Nome q: preceded by comma in q functions; q = exp@-p KH1 - mL K 8mLD = expHi p w wL
Invariants g2 , g3 (used by Mathematica)
Half-periods w, w : g2 = 60 r, s w-4 , g3 = 140 r, s w-6 , where w = 2 rw + 2 sw
Ratio of periods t: t = w w
3
Discriminant D: D = g2 - 27 g23
Parameters of curve a, b (used by Mathematica)
Coordinate y (used by Mathematica): related by y2 = x3 + a x2 + b x

Common argument conventions for elliptic integrals and elliptic functions.

JacobiAmplitude@u,mD give the amplitude f corresponding to argument u and

parameter m
EllipticNomeQ@mD give the nome q corresponding to parameter m
InverseEllipticNomeQ@qD give the parameter m corresponding to nome q

WeierstrassInvariants@8w,w <D
give the invariants 8g2 , g3 < corresponding to the half-
periods 8w, w <
WeierstrassHalfPeriods@8g2 ,g3 <D
give the half-periods 8w, w < corresponding to the invari-
ants 8g2 , g3 <

Converting between different argument conventions.

Elliptic Integrals
450 Mathematics and Algorithms

Elliptic Integrals

EllipticK@mD complete elliptic integral of the first kind K HmL

EllipticF@f,mD elliptic integral of the first kind FHf mL
EllipticE@mD complete elliptic integral of the second kind E HmL
EllipticE@f,mD elliptic integral of the second kind E Hf mL
EllipticPi@n,mD complete elliptic integral of the third kind P Hn mL
EllipticPi@n,f,mD elliptic integral of the third kind P Hn; f mL
JacobiZeta@f,mD Jacobi zeta function Z Hf mL

Elliptic integrals.

Integrals of the form RHx, yL x, where R is a rational function, and y2 is a cubic or quartic polyno-

mial in x, are known as elliptic integrals. Any elliptic integral can be expressed in terms of the
three standard kinds of Legendre-Jacobi elliptic integrals.

The elliptic integral of the first kind EllipticF@f, mD is given for -p 2 < f < p 2 by
f -12 sin HfL -12
FHf mL = 0 A1 - m sin2 HqLE q= 0 AI1 - t2 M I1 - m t2 ME t. This elliptic integral arises in solving

the equations of motion for a simple pendulum. It is sometimes known as an incomplete elliptic
integral of the first kind.

Note that the arguments of the elliptic integrals are sometimes given in the opposite order from
what is used in Mathematica.
p
The complete elliptic integral of the first kind EllipticK@mD is given by KHmL = F I 2 mM. Note that

K is used to denote the complete elliptic integral of the first kind, while F is used for its incom-
plete form. In many applications, the parameter m is not given explicitly, and KHmL is denoted
simply by K. The complementary complete elliptic integral of the first kind K HmL is given by
KH1 - mL. It is often denoted K . K and i K give the "real" and "imaginary" quarter-periods of the
corresponding Jacobi elliptic functions discussed in "Elliptic Functions".

The elliptic integral of the second kind EllipticE@f, mD is given for -p 2 < f < p 2 by
f 12 sin HfL -12 12
EHf mL = 0 A1 - m sin2 HqLE q= 0 I1 - t2 M I1 - m t2 M t.

p
The complete elliptic integral of the second kind EllipticE@mD is given by EHmL = E I 2 mM. It is

often denoted E. The complementary form is E HmL = EH1 - mL.

The Jacobi zeta function JacobiZeta@f, mD is given by ZHf mL = EHf mL - EHmL FHf mL KHmL.
Mathematics and Algorithms 451

The Jacobi zeta function JacobiZeta@f, mD is given by ZHf mL = EHf mL - EHmL FHf mL KHmL.

2
The Heuman lambda function is given by L0 Hf mL = FHf 1 - mL KH1 - mL + p
KHmL ZHf 1 - mL.

The elliptic integral of the third kind EllipticPi@n, f, mD is given by

f -1 -12
PHn; f mL = 0 I1 - n sin2 HqLM A1 - m sin2 HqLE q.

p
The complete elliptic integral of the third kind EllipticPi@n, mD is given by PHn mL = PIn; 2
mM.

Here is a plot of the complete elliptic integral of the second kind EHmL.
In[1]:= Plot@EllipticE@mD, 8m, 0, 1<D

1.5
1.4
1.3
Out[1]=
1.2
1.1

0.2 0.4 0.6 0.8 1.0

Here is KHaL with a = 30 .

In[2]:= EllipticK@Sin@30 DegreeD ^ 2D N
Out[2]= 1.68575

The elliptic integrals have a complicated structure in the complex plane.

In[1]:= Plot3D@Im@EllipticF@px + I py, 2DD, 8px, 0.5, 2.5<, 8py, - 1, 1<, PlotPoints -> 60D

Out[1]=
452 Mathematics and Algorithms

Elliptic Functions

JacobiAmplitude@u,mD amplitude function amHu mL

JacobiSN@u,mD , JacobiCN@u,mD , etc.
Jacobi elliptic functions snHu mL, etc.
InverseJacobiSN@v,mD , InverseJacobiCN@v,mD , etc.
inverse Jacobi elliptic functions sn-1 Hv mL, etc.
EllipticTheta@a,u,qD theta functions Ja Hu, qL (a = 1, , 4)

EllipticThetaPrime@a,u,qD derivatives of theta functions Ja Hu, qL (a = 1, , 4)

SiegelTheta @t,sD Siegel theta function QHt, sL

SiegelTheta @v,t,sD Siegel theta function Q@vD Ht, sL
WeierstrassP@u,8g2 ,g3 <D Weierstrass elliptic function Hu; g2 , g3 L

WeierstrassPPrime@u,8g2 ,g3 <D

derivative of Weierstrass elliptic function Hu; g2 , g3 L

InverseWeierstrassP@p,8g2 ,g3 <D

inverse Weierstrass elliptic function
WeierstrassSigma@u,8g2 ,g3 <D Weierstrass sigma function sHu; g2 , g3 L

WeierstrassZeta@u,8g2 ,g3 <D Weierstrass zeta function zHu; g2 , g3 L

Elliptic and related functions.

Rational functions involving square roots of quadratic forms can be integrated in terms of
inverse trigonometric functions. The trigonometric functions can thus be defined as inverses of
the functions obtained from these integrals.

By analogy, elliptic functions are defined as inverses of the functions obtained from elliptic
integrals.

The amplitude for Jacobi elliptic functions JacobiAmplitude@u, mD is the inverse of the elliptic
integral of the first kind. If u = FHf mL, then f = amHu mL. In working with Jacobi elliptic functions,
the argument m is often dropped, so amHu mL is written as amHuL.

The Jacobi elliptic functions JacobiSN@u, mD and JacobiCN@u, mD are given respectively by
snHuL = sinHfL and cnHuL = cosHfL, where f = amHu mL. In addition, JacobiDN@u, mD is given by

dnHuL = 1 - m sin2 HfL = DHfL.

Mathematics and Algorithms 453

There are a total of twelve Jacobi elliptic functions JacobiPQ@u, mD, with the letters P and Q
chosen from the set S, C, D and N. Each Jacobi elliptic function JacobiPQ@u, mD satisfies the
relation pqHuL = pnHuL qnHuL, where for these purposes nnHuL = 1.

There are many relations between the Jacobi elliptic functions, somewhat analogous to those
between trigonometric functions. In limiting cases, in fact, the Jacobi elliptic functions reduce to
trigonometric functions. So, for example, snHu 0L = sinHuL, snHu 1L = tanhHuL, cnHu 0L = cosHuL,
cnHu 1L = sechHuL, dnHu 0L = 1 and dnHu 1L = sechHuL.
u
The notation PqHuL is often used for the integrals 0 pq2 HtL t. These integrals can be expressed in

terms of the Jacobi zeta function defined in "Elliptic Integrals".

One of the most important properties of elliptic functions is that they are doubly periodic in the
complex values of their arguments. Ordinary trigonometric functions are singly periodic, in the
sense that f Hz + swL = f HzL for any integer s. The elliptic functions are doubly periodic, so that
f Hz + rw + sw L = f HzL for any pair of integers r and s.

The Jacobi elliptic functions snHu mL, etc. are doubly periodic in the complex u plane. Their peri-
ods include w = 4 KHmL and w = 4 iKH1 - mL, where K is the complete elliptic integral of the first
kind.

The choice of p and q in the notation pqHu mL for Jacobi elliptic functions can be understood in
terms of the values of the functions at the quarter periods K and i K .

This shows two complete periods in each direction of the absolute value of the Jacobi elliptic
1
function snJu N.
3

In[3]:= ContourPlot@Abs@JacobiSN@ux + I uy, 1 3DD,

8ux, 0, 4 EllipticK@1 3D<, 8uy, 0, 4 EllipticK@2 3D<D
8

Out[3]= 4

0
0 1 2 3 4 5 6 7
454 Mathematics and Algorithms

Also built into Mathematica are the inverse Jacobi elliptic functions InverseJacobiSN@v, mD,
InverseJacobiCN@v, mD, etc. The inverse function sn-1 Hv mL, for example, gives the value of u
for which v = snHu mL. The inverse Jacobi elliptic functions are related to elliptic integrals.

The four theta functions Ja Hu, qL are obtained from EllipticTheta@a, u, qD

by taking a to be 1, 2, 3 or 4. The functions are
defined by: J1 Hu, qL = 2 q 14

n=0 H-1L
n
q n Hn+1L
sin @H2 n + 1L uD, 14
J2 Hu, qL = 2 q
n=0 q
n Hn+1L
cos @H2 n + 1L uD,

n=1 q cosH2 n uL, J4 Hu, qL = 1 + 2 n=1 H-1L q cosH2 n uL. The theta functions are often
2 n n 2
J3 Hu, qL = 1 + 2 n

written as Ja HuL with the parameter q not explicitly given. The theta functions are sometimes
written in the form JHu mL, where m is related to q by q = exp @-p KH1 - mL KHmLD. In addition, q is

sometimes replaced by t, given by q = ei pt . All the theta functions satisfy a diffusion-like differen-
tial equation 2 JHu, tL u2 = 4 p i JHu, tL t.

The Siegel theta function SiegelTheta @t, sD with Riemann square modular matrix t of dimen-
sion p and vector s generalizes the elliptic theta functions to complex dimension p. It is defined
by QHt, sL = n expH pHn.t.n + 2 n.sLL, where n runs over all p-dimensional integer vectors. The Siegel

theta function with characteristic SiegelTheta @n, t, sD is defined by

QHn, t, sL = n expH pHHn + aL.t.Hn + aL + 2 Hn + aL.Hs + bLLL, where the characteristic n is a pair of p-dimen-

sional vectors 8a, b<.

The Jacobi elliptic functions can be expressed as ratios of the theta functions.

An alternative notation for theta functions is QHu mL = J4 Hv mL, Q1 Hu mL = J3 Hv mL, HHu mL = J1 HvL,
H1 Hu mL = J2 HvL, where v = p u 2 KHmL.

The Neville theta functions can be defined in terms of the theta functions as Js HuL = 2 KHmL J1 Hv mL
pJ1 H0 mL, Jc HuL = J2 Hv mL J2 H0 mL, Jd HuL = J3 Hv mL J3 H0 mL, Jn HuL = J4 Hv mL J4 H0 mL, where v = p u 2 KHmL.
The Jacobi elliptic functions can be represented as ratios of the Neville theta functions.

The Weierstrass elliptic function WeierstrassP@u, 8g2 , g3 <D can be considered as the inverse of
an elliptic integral. The Weierstrass function Hu; g2 , g3 L gives the value of x for which
x -12
u = I4 t3 - g2 t - g3 M t. The function WeierstrassPPrime@u, 8g2 , g3 <D is given by

Hu; g2 , g3 L = u
Hu; g2 , g3 L.

The Weierstrass functions are also sometimes written in terms of their fundamental half-periods
w and w , obtained from the invariants g2 and g3 using WeierstrassHalfPeriods@8u, 8g2 , g3 <D.

The function InverseWeierstrassP@p, 8g2 , g3 <D finds one of the two values of u for which
p = Hu; g2 , g3 L. This value always lies in the parallelogram defined by the complex number half-
Mathematics and Algorithms 455

InverseWeierstrassP@8p, q<, 8g2 , g3 <D finds the unique value of u for which p = Hu; g2 , g3 L and
q = Hu; g2 , g3 L. In order for any such value of u to exist, p and q must be related by
q2 = 4 p3 - g2 p - g3 .

The Weierstrass zeta function WeierstrassZeta@u, 8g2 , g3 <D and Weierstrass sigma function
WeierstrassSigma@u, 8g2 , g3 <D are related to the Weierstrass elliptic functions by
z Hz; g2 , g3 L = -Hz; g2 , g3 L and s Hz; g2 , g3 L sHz; g2 , g3 L = zHz; g2 , g3 L.

The Weierstrass zeta and sigma functions are not strictly elliptic functions since they are not
periodic.

Elliptic Modular Functions

DedekindEta @tD Dedekind eta function hHtL

KleinInvariantJ@tD Klein invariant modular function JHtL

ModularLambda@tD modular lambda function lHtL

Elliptic modular functions.

The modular lambda function ModularLambda@tD relates the ratio of half-periods t = w w to the
parameter according to m = lHtL.

The Klein invariant modular function KleinInvariantJ@tD and the Dedekind eta function
DedekindEta @tD satisfy the relations D = g32 JHtL = H2 pL12 h24 HtL.

Modular elliptic functions are defined to be invariant under certain fractional linear transforma-
tions of their arguments. Thus for example lHtL is invariant under any combination of the trans-
formations t t + 2 and t t H1 - 2 tL.
456 Mathematics and Algorithms

Generalized Elliptic Integrals and Functions

ArithmeticGeometricMean@a,bD the arithmetic-geometric mean of a and b

EllipticExp @u,8a,b<D generalized exponential associated with the elliptic curve
y2 = x3 + a x2 + bx
EllipticLog @8x,y<,8a,b<D generalized logarithm associated with the elliptic curve
y2 = x3 + a x2 + bx

Generalized elliptic integrals and functions.

The definitions for elliptic integrals and functions given above are based on traditional usage.
For modern algebraic geometry, it is convenient to use slightly more general definitions.

The function EllipticLog @8x, y<, 8a, b<D is defined as the value of the integral
1 x 3 -12
2
It + a t2 + b tM t, where the sign of the square root is specified by giving the value of y such

x -12
that y = x3 + a x2 + b x . Integrals of the form It2 + a tM t can be expressed in terms of the

ordinary logarithm (and inverse trigonometric functions). You can think of EllipticLog as
giving a generalization of this, where the polynomial under the square root is now of degree
three.

The function EllipticExp @u, 8a, b<D is the inverse of EllipticLog . It returns the list 8x, y<
that appears in EllipticLog . EllipticExp is an elliptic function, doubly periodic in the com-
plex u plane.

ArithmeticGeometricMean@a, bD gives the arithmetic-geometric mean (AGM) of two numbers a

and b. This quantity is central to many numerical algorithms for computing elliptic integrals and
other functions. For positive reals a and b the AGM is obtained by starting with a0 = a, b0 = b, then
1
iterating the transformation an+1 = 2
Han + bn L, bn+1 = an bn until an = bn to the precision required.
Mathematics and Algorithms 457

Mathieu and Related Functions

MathieuC@a,q,zD even Mathieu functions with characteristic value a and

parameter q
MathieuS@b,q,zD odd Mathieu functions with characteristic value b and
parameter q
MathieuCPrime@a,q,zD z derivatives of Mathieu functions
and MathieuSPrime@b,q,zD
MathieuCharacteristicA@r,qD characteristic value ar for even Mathieu functions with
characteristic exponent r and parameter q
MathieuCharacteristicB@r,qD characteristic value br for odd Mathieu functions with
characteristic exponent r and parameter q
MathieuCharacteristicExpone-@ characteristic exponent r for Mathieu functions with charac -
nt teristic value a and parameter q
a,qD

Mathieu and related functions.

The Mathieu functions MathieuC@a, q, zD and MathieuS@a, q, zD are solutions to the equation
y + @a - 2 q cos H2 zLD y = 0. This equation appears in many physical situations that involve elliptical
shapes or periodic potentials. The function MathieuC is defined to be even in z, while MathieuS
is odd.

When q = 0 the Mathieu functions are simply cos I a zM and sin I a zM. For nonzero q, the Mathieu

functions are only periodic in z for certain values of a. Such Mathieu characteristic values are
given by MathieuCharacteristicA@r, qD and MathieuCharacteristicB@r, qD with r an integer
or rational number. These values are often denoted by ar and br .

For integer r, the even and odd Mathieu functions with characteristic values ar and br are often
denoted c er Hz, qL and s er Hz, qL, respectively. Note the reversed order of the arguments z and q.

According to Floquets theorem any Mathieu function can be written in the form ei rz
f HzL, where
f HzL has period 2p and r is the Mathieu characteristic exponent
MathieuCharacteristicExponent@a, qD. When the characteristic exponent r is an integer or
rational number, the Mathieu function is therefore periodic. In general, however, when r is not a
real integer, ar and br turn out to be equal.
458 Mathematics and Algorithms

This shows the first five characteristic values ar as functions of q.

In[1]:= Plot@Evaluate@Table@MathieuCharacteristicA@r, qD, 8r, 0, 4<DD, 8q, 0, 15<D

10
Out[1]=
2 4 6 8 10 12 14
-10

-20

Working with Special Functions

automatic evaluation exact results for specific arguments

N@expr,nD numerical approximations to any precision
D@expr,xD exact results for derivatives
N@D@expr,xDD numerical approximations to derivatives
Series@expr,8x,x0 ,n<D series expansions

Integrate@expr,xD exact results for integrals

NIntegrate@expr,xD numerical approximations to integrals
FindRoot@expr==0,8x,x0 <D numerical approximations to roots

Some common operations on special functions.

Most special functions have simpler forms when given certain specific arguments. Mathematica
will automatically simplify special functions in such cases.

Mathematica automatically writes this in terms of standard mathematical constants.

In[1]:= PolyLog@2, 1 2D
p2 Log@2D2
Out[1]= -
12 2

Here again Mathematica reduces a special case of the Airy function to an expression involving
gamma functions.
In[2]:= AiryAi@0D
1
Out[2]=
2
323 GammaB F
3

For most choices of arguments, no exact reductions of special functions are possible. But in
such cases, Mathematica allows you to find numerical approximations to any degree of preci-
sion. The algorithms that are built into Mathematica cover essentially all values of parameters~
real and complex~for which the special functions are defined.
Mathematics and Algorithms 459

There is no exact result known here.

In[3]:= AiryAi@1D
Out[3]= AiryAi@1D

This gives a numerical approximation to 40 digits of precision.

In[4]:= N@AiryAi@1D, 40D
Out[4]= 0.1352924163128814155241474235154663061749

The result here is a huge complex number, but Mathematica can still find it.
In[5]:= N@AiryAi@1000 IDD
6472
Out[5]= -4.780266637767027 10 + 3.674920907226875 106472

Most special functions have derivatives that can be expressed in terms of elementary functions
or other special functions. But even in cases where this is not so, you can still use N to find
numerical approximations to derivatives.

This derivative comes out in terms of elementary functions.

In[6]:= D@FresnelS@xD, xD
p x2
Out[6]= SinB F
2

This evaluates the derivative of the gamma function at the point 3.

In[7]:= Gamma @3D
3
Out[7]= 2 - EulerGamma
2

There is no exact formula for this derivative of the zeta function.

In[8]:= Zeta @PiD

Out[8]= Zeta @pD
460 Mathematics and Algorithms

Applying N gives a numerical approximation.

In[9]:= N@%D
Out[9]= -0.167603

Mathematica incorporates a vast amount of knowledge about special functions~including essen-

tially all the results that have been derived over the years. You access this knowledge whenever
you do operations on special functions in Mathematica.

Here is a series expansion for a Fresnel function.

In[10]:= Series@FresnelS@xD, 8x, 0, 15<D
p x3 p3 x7 p5 x11 p7 x15
Out[10]= - + - + O@xD16
6 336 42 240 9 676 800

Mathematica knows how to do a vast range of integrals involving special functions.

In[11]:= Integrate@AiryAi@xD ^ 2, 8x, 0, Infinity<D
1
Out[11]=
1 2
323 GammaB F
3

One feature of working with special functions is that there are a large number of relations
between different functions, and these relations can often be used in simplifying expressions.

FullSimplify@exprD try to simplify expr using a range of transformation rules

Simplifying expressions involving special functions.

This uses the reflection formula for the gamma function.

In[12]:= FullSimplify@Gamma@xD Gamma@1 - xDD
Out[12]= p Csc@p xD

This makes use of a representation for Chebyshev polynomials.

In[13]:= FullSimplify@ChebyshevT@n, zD - k Cos@n ArcCos@zDDD
Out[13]= -H-1 + kL Cos@n ArcCos@zDD

The Airy functions are related to Bessel functions.

In[14]:= FullSimplify@3 AiryAi@1D + Sqrt@3D AiryBi@1DD
1 2
Out[14]= 2 BesselIB- , F
3 3
Mathematics and Algorithms 461

FunctionExpand@exprD try to expand out special functions

Manipulating expressions involving special functions.

This expands the Gauss hypergeometric function into simpler functions.

In[15]:= FunctionExpand@Hypergeometric2F1@1 2, 3 2, 3, xDD
16 H2 - xL EllipticE@xD 16 H-2 + 2 xL EllipticK@xD
Out[15]= +
3 p x2 3 p x2

Here is an example involving Bessel functions.

In[16]:= FunctionExpand@BesselY@n, I xDD
2 H xL-n xn BesselK@n, xD
Out[16]= - + BesselI@n, xD I-H xL-n xn + H xLn x-n Cos@n pDM Csc@n pD
p

In this case the final result does not even involve PolyGamma.
In[17]:= FunctionExpand@Im@PolyGamma@0, 3 IDDD
1 1
Out[17]= + p Coth@3 pD
6 2

This finds an expression for a derivative of the Hurwitz zeta function.

In[18]:= FunctionExpand@Derivative@1, 0D@ZetaD@- 1, 4DD
1
Out[18]= + 2 Log@2D + 3 Log@3D - Log@GlaisherD
12

TCP2101 Algorithm Design and Analysis Lab07 - Divide-and-Conquer
No ratings yet
TCP2101 Algorithm Design and Analysis Lab07 - Divide-and-Conquer
5 pages
Combinatorial Optimization: Alexander Schrijver
No ratings yet
Combinatorial Optimization: Alexander Schrijver
34 pages
Structured Analysis and System Specification - DeMarco
No ratings yet
Structured Analysis and System Specification - DeMarco
6 pages
Chapter 1 Answers
100% (1)
Chapter 1 Answers
8 pages
Leetcode Preparation
No ratings yet
Leetcode Preparation
130 pages
Back Tracking
No ratings yet
Back Tracking
36 pages
Algorithms On String Trees and Sequences
No ratings yet
Algorithms On String Trees and Sequences
326 pages
NNG Reference Manual, Second Edition
From Everand
NNG Reference Manual, Second Edition
Garrett D'Amore
No ratings yet
Algorithms: 2006 S. Dasgupta, C. H. Papadimitriou, and U. V. Vazirani July 18, 2006
No ratings yet
Algorithms: 2006 S. Dasgupta, C. H. Papadimitriou, and U. V. Vazirani July 18, 2006
8 pages
Amazon Questions
No ratings yet
Amazon Questions
2 pages
Algorithm
No ratings yet
Algorithm
91 pages
Bit Manipulation Notes by Kapil Yadav
No ratings yet
Bit Manipulation Notes by Kapil Yadav
79 pages
Fundamentals of Algorithm
No ratings yet
Fundamentals of Algorithm
14 pages
Data Stru by Chapman and Application
No ratings yet
Data Stru by Chapman and Application
1,321 pages
20 Coding Patterns To Master MAANG Interviews
No ratings yet
20 Coding Patterns To Master MAANG Interviews
22 pages
5 Designing Dropbox - Grokking The System Design Interview
No ratings yet
5 Designing Dropbox - Grokking The System Design Interview
10 pages
7 Data Structure Stack Questions
100% (1)
7 Data Structure Stack Questions
22 pages
Data Structures and Algorithms: Lecture Notes For
No ratings yet
Data Structures and Algorithms: Lecture Notes For
126 pages
Computer Arithmetic - Algorith
No ratings yet
Computer Arithmetic - Algorith
510 pages
Algorithm Essentials Java
No ratings yet
Algorithm Essentials Java
526 pages
Pointers and Dynamic Objects: COMP171 Fall 2006
No ratings yet
Pointers and Dynamic Objects: COMP171 Fall 2006
55 pages
Dynamic Programming
No ratings yet
Dynamic Programming
24 pages
Knapsack Algorithm
No ratings yet
Knapsack Algorithm
9 pages
Don't Overprep Coding Interviews.: For Your
No ratings yet
Don't Overprep Coding Interviews.: For Your
24 pages
Dokumen - Pub Introduction To The Design and Analysis of Algorithms 0071243461 9780071243469
No ratings yet
Dokumen - Pub Introduction To The Design and Analysis of Algorithms 0071243461 9780071243469
750 pages
bookAdomaitisAppliedMath4ChE 19utf81 PDF
No ratings yet
bookAdomaitisAppliedMath4ChE 19utf81 PDF
216 pages
Notes
No ratings yet
Notes
422 pages
Design Analysis and Algorithm
100% (1)
Design Analysis and Algorithm
78 pages
Introduction To Graph Theory
No ratings yet
Introduction To Graph Theory
22 pages
33 - BD - Data Structures and Algorithms - Narasimha Karumanchi
No ratings yet
33 - BD - Data Structures and Algorithms - Narasimha Karumanchi
12 pages
Dynamic Programming
No ratings yet
Dynamic Programming
69 pages
An Introduction To The Usa Computing Olympiad: Darren Yao
No ratings yet
An Introduction To The Usa Computing Olympiad: Darren Yao
87 pages
Puzzles As Programmer Interview Question
No ratings yet
Puzzles As Programmer Interview Question
32 pages
Graph Algorithms
No ratings yet
Graph Algorithms
45 pages
Backtracking Algorithms
100% (2)
Backtracking Algorithms
42 pages
Ds and Algo
No ratings yet
Ds and Algo
2 pages
Programming Interview Questions
No ratings yet
Programming Interview Questions
6 pages
C++ Programming (Mastering Programming Languages Series) by Theophilus Edet
No ratings yet
C++ Programming (Mastering Programming Languages Series) by Theophilus Edet
336 pages
Graph Theory: Graph of A Function Graph (Disambiguation)
No ratings yet
Graph Theory: Graph of A Function Graph (Disambiguation)
8 pages
Fundamentals of Data Structure in C - S. Sahni, S. Anderson-Freed and E. Horowitz
100% (1)
Fundamentals of Data Structure in C - S. Sahni, S. Anderson-Freed and E. Horowitz
82 pages
('Christos Papadimitriou', 'Midterm 2', ' (Solution) ') Fall 2009 PDF
No ratings yet
('Christos Papadimitriou', 'Midterm 2', ' (Solution) ') Fall 2009 PDF
4 pages
A2oj Ladders Abcde
No ratings yet
A2oj Ladders Abcde
24 pages
Linked List
No ratings yet
Linked List
258 pages
Introduction To Computer Science in C Sharp Release 2
No ratings yet
Introduction To Computer Science in C Sharp Release 2
252 pages
CSC101 Computer Science and Programming Ezeani Majesty Ignatius
No ratings yet
CSC101 Computer Science and Programming Ezeani Majesty Ignatius
114 pages
Pytest
No ratings yet
Pytest
487 pages
Dynamic Programming
No ratings yet
Dynamic Programming
14 pages
Latest Data Structures
No ratings yet
Latest Data Structures
202 pages
Tree Data Structure - GeeksforGeeks
No ratings yet
Tree Data Structure - GeeksforGeeks
1 page
ف1
No ratings yet
ف1
4 pages
Floating Point Numbers
No ratings yet
Floating Point Numbers
26 pages
Data Structures and Algorithm Analysis 1005
100% (1)
Data Structures and Algorithm Analysis 1005
22 pages
Coding Problems
0% (1)
Coding Problems
75 pages
Exam Data Structure
No ratings yet
Exam Data Structure
2 pages
Dynamic programming The Ultimate Step-By-Step Guide
From Everand
Dynamic programming The Ultimate Step-By-Step Guide
Gerardus Blokdyk
No ratings yet
Logic synthesis Standard Requirements
From Everand
Logic synthesis Standard Requirements
Gerardus Blokdyk
No ratings yet
Computational complexity theory A Complete Guide
From Everand
Computational complexity theory A Complete Guide
Gerardus Blokdyk
No ratings yet
Opt Art: From Mathematical Optimization to Visual Design
From Everand
Opt Art: From Mathematical Optimization to Visual Design
Robert Bosch
No ratings yet
Learn Digital and Microprocessor Techniques On Your Smartphone: Portable Learning, Reference and Revision Tools.
From Everand
Learn Digital and Microprocessor Techniques On Your Smartphone: Portable Learning, Reference and Revision Tools.
Clive W. Humphris
No ratings yet
Tizen Cookbook
From Everand
Tizen Cookbook
Leon Anavi
No ratings yet
VLSI Career ICE Breaker
From Everand
VLSI Career ICE Breaker
Yogesh Soni
3/5 (1)
Guc 19 59 29459 2023-03-04T10 01 39
No ratings yet
Guc 19 59 29459 2023-03-04T10 01 39
8 pages
Binary Systems: Atty. Manuel O. Diaz JR
No ratings yet
Binary Systems: Atty. Manuel O. Diaz JR
44 pages
Computer Programming MCQ With Answers
0% (1)
Computer Programming MCQ With Answers
2 pages
Itpc2 Platform 1S 2022 2023
No ratings yet
Itpc2 Platform 1S 2022 2023
18 pages
Assignment Four 1
No ratings yet
Assignment Four 1
21 pages
Computer Fundamentals - 15 QnA
No ratings yet
Computer Fundamentals - 15 QnA
4 pages
Electronic Data and Instruction
No ratings yet
Electronic Data and Instruction
1 page
4-Bit Binary To Decimal Decoder (BDD)
100% (2)
4-Bit Binary To Decimal Decoder (BDD)
6 pages
Exploding Dots ™ Teaching Guide: Insight
No ratings yet
Exploding Dots ™ Teaching Guide: Insight
12 pages
How To Convert Octal To Binary Step by Step
No ratings yet
How To Convert Octal To Binary Step by Step
1 page
Eric C. R. Hehner R. N. S. Horspool: D D D N D B
No ratings yet
Eric C. R. Hehner R. N. S. Horspool: D D D N D B
16 pages
AL 2025 ICT - NUMBER SYSTEM ANSWER SCHEME - RIYAS RUSHARD - Learn with Maestro - 078 8584316
No ratings yet
AL 2025 ICT - NUMBER SYSTEM ANSWER SCHEME - RIYAS RUSHARD - Learn with Maestro - 078 8584316
32 pages
Combinational Logic Circuits
No ratings yet
Combinational Logic Circuits
53 pages
SELF CHECK 1.5 (Buenconsejo)
No ratings yet
SELF CHECK 1.5 (Buenconsejo)
2 pages
Z-80 Assembly Language Programming 1979 Leventhal Text
100% (1)
Z-80 Assembly Language Programming 1979 Leventhal Text
642 pages
Grade 8 Lesson Plan Number Base System
No ratings yet
Grade 8 Lesson Plan Number Base System
14 pages
Lec - 1 Micro Must
No ratings yet
Lec - 1 Micro Must
19 pages
Module 2 Computer Programming
No ratings yet
Module 2 Computer Programming
43 pages
MCA 454 Java Lab Assignment 1
No ratings yet
MCA 454 Java Lab Assignment 1
7 pages
CVE 154 - Statics of Rigid Bodies: Lesson 2: Errors in Numerical Computation
No ratings yet
CVE 154 - Statics of Rigid Bodies: Lesson 2: Errors in Numerical Computation
37 pages
Binary Subtraction: St. Paul University Surigao
No ratings yet
Binary Subtraction: St. Paul University Surigao
7 pages
Programming Lab File
No ratings yet
Programming Lab File
71 pages
Environment - Part 3 - Updates On RAF Bentwaters Mystery - IR Anomalies and Telepathic Download of Binary Code
No ratings yet
Environment - Part 3 - Updates On RAF Bentwaters Mystery - IR Anomalies and Telepathic Download of Binary Code
15 pages
Deschamps - Sutter - Canto Guide To FPGA Implementation of Algorithms
No ratings yet
Deschamps - Sutter - Canto Guide To FPGA Implementation of Algorithms
5 pages
Computer Systems Fundamentals H175 34: Numbers in Base 16
No ratings yet
Computer Systems Fundamentals H175 34: Numbers in Base 16
25 pages
SLG CS1 8.1 Number System
No ratings yet
SLG CS1 8.1 Number System
6 pages
Lesson Exemplar
No ratings yet
Lesson Exemplar
3 pages
Harvard CS50 - Full Computer Science University Course - YouTube - English
No ratings yet
Harvard CS50 - Full Computer Science University Course - YouTube - English
966 pages
EECS 452 Digital Signal Processing Lab: Lecture 3: Number Representation
No ratings yet
EECS 452 Digital Signal Processing Lab: Lecture 3: Number Representation
46 pages