Topology: A Very Short Introduction

VERY SHORT INTRODUCTIONS are for anyone wanting a stimulating and

accessible way into a new subject. They are written by experts, and have been
translated into more than 45 different languages.
The series began in 1995, and now covers a wide variety of topics in every
discipline. The VSI library currently contains over 600 volumes—a Very Short
Introduction to everything from Psychology and Philosophy of Science to American
History and Relativity—and continues to grow in every subject area.

Very Short Introductions available now:

ABOLITIONISM Richard S. Newman

THE ABRAHAMIC RELIGIONS Charles L. Cohen
ACCOUNTING Christopher Nobes
ADAM SMITH Christopher J. Berry
ADOLESCENCE Peter K. Smith
ADVERTISING Winston Fletcher
AESTHETICS Bence Nanay
AFRICAN AMERICAN RELIGION Eddie S. Glaude Jr
AFRICAN HISTORY John Parker and Richard Rathbone
AFRICAN POLITICS Ian Taylor
AFRICAN RELIGIONS Jacob K. Olupona
AGEING Nancy A. Pachana
AGNOSTICISM Robin Le Poidevin
AGRICULTURE Paul Brassley and Richard Soffe
ALEXANDER THE GREAT Hugh Bowden
ALGEBRA Peter M. Higgins
AMERICAN CULTURAL HISTORY Eric Avila
AMERICAN FOREIGN RELATIONS Andrew Preston
AMERICAN HISTORY Paul S. Boyer
AMERICAN IMMIGRATION David A. Gerber
AMERICAN LEGAL HISTORY G. Edward White
AMERICAN NAVAL HISTORY Craig L. Symonds
AMERICAN POLITICAL HISTORY Donald Critchlow
AMERICAN POLITICAL PARTIES AND ELECTIONS L. Sandy Maisel
AMERICAN POLITICS Richard M. Valelly
THE AMERICAN PRESIDENCY Charles O. Jones
THE AMERICAN REVOLUTION Robert J. Allison
AMERICAN SLAVERY Heather Andrea Williams
THE AMERICAN WEST Stephen Aron
AMERICAN WOMEN’S HISTORY Susan Ware
ANAESTHESIA Aidan O’Donnell
ANALYTIC PHILOSOPHY Michael Beaney
ANARCHISM Colin Ward
ANCIENT ASSYRIA Karen Radner
ANCIENT EGYPT Ian Shaw
ANCIENT EGYPTIAN ART AND ARCHITECTURE Christina Riggs
ANCIENT GREECE Paul Cartledge
THE ANCIENT NEAR EAST Amanda H. Podany
ANCIENT PHILOSOPHY Julia Annas
ANCIENT WARFARE Harry Sidebottom
ANGELS David Albert Jones
ANGLICANISM Mark Chapman
THE ANGLO-SAXON AGE John Blair
ANIMAL BEHAVIOUR Tristram D. Wyatt
THE ANIMAL KINGDOM Peter Holland
ANIMAL RIGHTS David DeGrazia
THE ANTARCTIC Klaus Dodds
ANTHROPOCENE Erle C. Ellis
ANTISEMITISM Steven Beller
ANXIETY Daniel Freeman and Jason Freeman
THE APOCRYPHAL GOSPELS Paul Foster
APPLIED MATHEMATICS Alain Goriely
ARCHAEOLOGY Paul Bahn
ARCHITECTURE Andrew Ballantyne
ARISTOCRACY William Doyle
ARISTOTLE Jonathan Barnes
ART HISTORY Dana Arnold
ART THEORY Cynthia Freeland
ARTIFICIAL INTELLIGENCE Margaret A. Boden
ASIAN AMERICAN HISTORY Madeline Y. Hsu
ASTROBIOLOGY David C. Catling
ASTROPHYSICS James Binney
ATHEISM Julian Baggini
THE ATMOSPHERE Paul I. Palmer
AUGUSTINE Henry Chadwick
AUSTRALIA Kenneth Morgan
AUTISM Uta Frith
AUTOBIOGRAPHY Laura Marcus
THE AVANT GARDE David Cottington
THE AZTECS Davíd Carrasco
BABYLONIA Trevor Bryce
BACTERIA Sebastian G. B. Amyes
BANKING John Goddard and John O. S. Wilson
BARTHES Jonathan Culler
THE BEATS David Sterritt
BEAUTY Roger Scruton
BEHAVIOURAL ECONOMICS Michelle Baddeley
BESTSELLERS John Sutherland
THE BIBLE John Riches
BIBLICAL ARCHAEOLOGY Eric H. Cline
BIG DATA Dawn E. Holmes
BIOGRAPHY Hermione Lee
BIOMETRICS Michael Fairhurst
BLACK HOLES Katherine Blundell
BLOOD Chris Cooper
THE BLUES Elijah Wald
THE BODY Chris Shilling
THE BOOK OF COMMON PRAYER Brian Cummings
THE BOOK OF MORMON Terryl Givens
BORDERS Alexander C. Diener and Joshua Hagen
THE BRAIN Michael O’Shea
BRANDING Robert Jones
THE BRICS Andrew F. Cooper
THE BRITISH CONSTITUTION Martin Loughlin
THE BRITISH EMPIRE Ashley Jackson
BRITISH POLITICS Anthony Wright
BUDDHA Michael Carrithers
BUDDHISM Damien Keown
BUDDHIST ETHICS Damien Keown
BYZANTIUM Peter Sarris
C. S. LEWIS James Como
CALVINISM Jon Balserak
CANCER Nicholas James
CAPITALISM James Fulcher
CATHOLICISM Gerald O’Collins
CAUSATION Stephen Mumford and Rani Lill Anjum
THE CELL Terence Allen and Graham Cowling
THE CELTS Barry Cunliffe
CHAOS Leonard Smith
CHARLES DICKENS Jenny Hartley
CHEMISTRY Peter Atkins
CHILD PSYCHOLOGY Usha Goswami
CHILDREN’S LITERATURE Kimberley Reynolds
CHINESE LITERATURE Sabina Knight
CHOICE THEORY Michael Allingham
CHRISTIAN ART Beth Williamson
CHRISTIAN ETHICS D. Stephen Long
CHRISTIANITY Linda Woodhead
CIRCADIAN RHYTHMS Russell Foster and Leon Kreitzman
CITIZENSHIP Richard Bellamy
CIVIL ENGINEERING David Muir Wood
CLASSICAL LITERATURE William Allan
CLASSICAL MYTHOLOGY Helen Morales
CLASSICS Mary Beard and John Henderson
CLAUSEWITZ Michael Howard
CLIMATE Mark Maslin
CLIMATE CHANGE Mark Maslin
CLINICAL PSYCHOLOGY Susan Llewelyn and Katie Aafjes-van Doorn
COGNITIVE NEUROSCIENCE Richard Passingham
THE COLD WAR Robert McMahon
COLONIAL AMERICA Alan Taylor
COLONIAL LATIN AMERICAN LITERATURE Rolena Adorno
COMBINATORICS Robin Wilson
COMEDY Matthew Bevis
COMMUNISM Leslie Holmes
COMPARATIVE LITERATURE Ben Hutchinson
COMPLEXITY John H. Holland
THE COMPUTER Darrel Ince
COMPUTER SCIENCE Subrata Dasgupta
CONCENTRATION CAMPS Dan Stone
CONFUCIANISM Daniel K. Gardner
THE CONQUISTADORS Matthew Restall and Felipe Fernández-Armesto
CONSCIENCE Paul Strohm
CONSCIOUSNESS Susan Blackmore
CONTEMPORARY ART Julian Stallabrass
CONTEMPORARY FICTION Robert Eaglestone
CONTINENTAL PHILOSOPHY Simon Critchley
COPERNICUS Owen Gingerich
CORAL REEFS Charles Sheppard
CORPORATE SOCIAL RESPONSIBILITY Jeremy Moon
CORRUPTION Leslie Holmes
COSMOLOGY Peter Coles
COUNTRY MUSIC Richard Carlin
CRIME FICTION Richard Bradford
CRIMINAL JUSTICE Julian V. Roberts
CRIMINOLOGY Tim Newburn
CRITICAL THEORY Stephen Eric Bronner
THE CRUSADES Christopher Tyerman
CRYPTOGRAPHY Fred Piper and Sean Murphy
CRYSTALLOGRAPHY A. M. Glazer
THE CULTURAL REVOLUTION Richard Curt Kraus
DADA AND SURREALISM David Hopkins
DANTE Peter Hainsworth and David Robey
DARWIN Jonathan Howard
THE DEAD SEA SCROLLS Timothy H. Lim
DECADENCE David Weir
DECOLONIZATION Dane Kennedy
DEMOCRACY Bernard Crick
DEMOGRAPHY Sarah Harper
DEPRESSION Jan Scott and Mary Jane Tacchi
DERRIDA Simon Glendinning
DESCARTES Tom Sorell
DESERTS Nick Middleton
DESIGN John Heskett
DEVELOPMENT Ian Goldin
DEVELOPMENTAL BIOLOGY Lewis Wolpert
THE DEVIL Darren Oldridge
DIASPORA Kevin Kenny
DICTIONARIES Lynda Mugglestone
DINOSAURS David Norman
DIPLOMACY Joseph M. Siracusa
DOCUMENTARY FILM Patricia Aufderheide
DREAMING J. Allan Hobson
DRUGS Les Iversen
DRUIDS Barry Cunliffe
DYNASTY Jeroen Duindam
DYSLEXIA Margaret J. Snowling
EARLY MUSIC Thomas Forrest Kelly
THE EARTH Martin Redfern
EARTH SYSTEM SCIENCE Tim Lenton
ECONOMICS Partha Dasgupta
EDUCATION Gary Thomas
EGYPTIAN MYTH Geraldine Pinch
EIGHTEENTH‐CENTURY BRITAIN Paul Langford
THE ELEMENTS Philip Ball
EMOTION Dylan Evans
EMPIRE Stephen Howe
ENERGY SYSTEMS Nick Jenkins
ENGELS Terrell Carver
ENGINEERING David Blockley
THE ENGLISH LANGUAGE Simon Horobin
ENGLISH LITERATURE Jonathan Bate
THE ENLIGHTENMENT John Robertson
ENTREPRENEURSHIP Paul Westhead and Mike Wright
ENVIRONMENTAL ECONOMICS Stephen Smith
ENVIRONMENTAL ETHICS Robin Attfield
ENVIRONMENTAL LAW Elizabeth Fisher
ENVIRONMENTAL POLITICS Andrew Dobson
EPICUREANISM Catherine Wilson
EPIDEMIOLOGY Rodolfo Saracci
ETHICS Simon Blackburn
ETHNOMUSICOLOGY Timothy Rice
THE ETRUSCANS Christopher Smith
EUGENICS Philippa Levine
THE EUROPEAN UNION Simon Usherwood and John Pinder
EUROPEAN UNION LAW Anthony Arnull
EVOLUTION Brian and Deborah Charlesworth
EXISTENTIALISM Thomas Flynn
EXPLORATION Stewart A. Weaver
EXTINCTION Paul B. Wignall
THE EYE Michael Land
FAIRY TALE Marina Warner
FAMILY LAW Jonathan Herring
FASCISM Kevin Passmore
FASHION Rebecca Arnold
FEDERALISM Mark J. Rozell and Clyde Wilcox
FEMINISM Margaret Walters
FILM Michael Wood
FILM MUSIC Kathryn Kalinak
FILM NOIR James Naremore
THE FIRST WORLD WAR Michael Howard
FOLK MUSIC Mark Slobin
FOOD John Krebs
FORENSIC PSYCHOLOGY David Canter
FORENSIC SCIENCE Jim Fraser
FORESTS Jaboury Ghazoul
FOSSILS Keith Thomson
FOUCAULT Gary Gutting
THE FOUNDING FATHERS R. B. Bernstein
FRACTALS Kenneth Falconer
FREE SPEECH Nigel Warburton
FREE WILL Thomas Pink
FREEMASONRY Andreas Önnerfors
FRENCH LITERATURE John D. Lyons
THE FRENCH REVOLUTION William Doyle
FREUD Anthony Storr
FUNDAMENTALISM Malise Ruthven
FUNGI Nicholas P. Money
THE FUTURE Jennifer M. Gidley
GALAXIES John Gribbin
GALILEO Stillman Drake
GAME THEORY Ken Binmore
GANDHI Bhikhu Parekh
GARDEN HISTORY Gordon Campbell
GENES Jonathan Slack
GENIUS Andrew Robinson
GENOMICS John Archibald
GEOFFREY CHAUCER David Wallace
GEOGRAPHY JOHN Matthews and David Herbert
GEOLOGY Jan Zalasiewicz
GEOPHYSICS William Lowrie
GEOPOLITICS Klaus Dodds
GERMAN LITERATURE Nicholas Boyle
GERMAN PHILOSOPHY Andrew Bowie
GLACIATION David J. A. Evans
GLOBAL CATASTROPHES Bill McGuire
GLOBAL ECONOMIC HISTORY Robert C. Allen
GLOBALIZATION Manfred Steger
GOD John Bowker
GOETHE Ritchie Robertson
THE GOTHIC Nick Groom
GOVERNANCE Mark Bevir
GRAVITY Timothy Clifton
THE GREAT DEPRESSION AND THE NEW DEAL Eric Rauchway
HABERMAS James Gordon Finlayson
THE HABSBURG EMPIRE Martyn Rady
HAPPINESS Daniel M. Haybron
THE HARLEM RENAISSANCE Cheryl A. Wall
THE HEBREW BIBLE AS LITERATURE Tod Linafelt
HEGEL Peter Singer
HEIDEGGER Michael Inwood
THE HELLENISTIC AGE Peter Thonemann
HEREDITY John Waller
HERMENEUTICS Jens Zimmermann
HERODOTUS Jennifer T. Roberts
HIEROGLYPHS Penelope Wilson
HINDUISM Kim Knott
HISTORY John H. Arnold
THE HISTORY OF ASTRONOMY Michael Hoskin
THE HISTORY OF CHEMISTRY William H. Brock
THE HISTORY OF CHILDHOOD James Marten
THE HISTORY OF CINEMA Geoffrey Nowell-Smith
THE HISTORY OF LIFE Michael Benton
THE HISTORY OF MATHEMATICS Jacqueline Stedall
THE HISTORY OF MEDICINE William Bynum
THE HISTORY OF PHYSICS J. L. Heilbron
THE HISTORY OF TIME Leofranc Holford‐Strevens
HIV AND AIDS Alan Whiteside
HOBBES Richard Tuck
HOLLYWOOD Peter Decherney
THE HOLY ROMAN EMPIRE Joachim Whaley
HOME Michael Allen Fox
HOMER Barbara Graziosi
HORMONES Martin Luck
HUMAN ANATOMY Leslie Klenerman
HUMAN EVOLUTION Bernard Wood
HUMAN RIGHTS Andrew Clapham
HUMANISM Stephen Law
HUME A. J. Ayer
HUMOUR Noël Carroll
THE ICE AGE Jamie Woodward
IDENTITY Florian Coulmas
IDEOLOGY Michael Freeden
THE IMMUNE SYSTEM Paul Klenerman
INDIAN CINEMA Ashish Rajadhyaksha
INDIAN PHILOSOPHY Sue Hamilton
THE INDUSTRIAL REVOLUTION Robert C. Allen
INFECTIOUS DISEASE Marta L. Wayne and Benjamin M. Bolker
INFINITY Ian Stewart
INFORMATION Luciano Floridi
INNOVATION Mark Dodgson and David Gann
INTELLECTUAL PROPERTY Siva Vaidhyanathan
INTELLIGENCE Ian J. Deary
INTERNATIONAL LAW Vaughan Lowe
INTERNATIONAL MIGRATION Khalid Koser
INTERNATIONAL RELATIONS Paul Wilkinson
INTERNATIONAL SECURITY Christopher S. Browning
IRAN Ali M. Ansari
ISLAM Malise Ruthven
ISLAMIC HISTORY Adam Silverstein
ISOTOPES Rob Ellam
ITALIAN LITERATURE Peter Hainsworth and David Robey
JESUS Richard Bauckham
JEWISH HISTORY David N. Myers
JOURNALISM Ian Hargreaves
JUDAISM Norman Solomon
JUNG Anthony Stevens
KABBALAH Joseph Dan
KAFKA Ritchie Robertson
KANT Roger Scruton
KEYNES Robert Skidelsky
KIERKEGAARD Patrick Gardiner
KNOWLEDGE Jennifer Nagel
THE KORAN Michael Cook
LAKES Warwick F. Vincent
LANDSCAPE ARCHITECTURE Ian H. Thompson
LANDSCAPES AND GEOMORPHOLOGY Andrew Goudie and Heather Viles
LANGUAGES Stephen R. Anderson
LATE ANTIQUITY Gillian Clark
LAW Raymond Wacks
THE LAWS OF THERMODYNAMICS Peter Atkins
LEADERSHIP Keith Grint
LEARNING Mark Haselgrove
LEIBNIZ Maria Rosa Antognazza
LEO TOLSTOY Liza Knapp
LIBERALISM Michael Freeden
LIGHT Ian Walmsley
LINCOLN Allen C. Guelzo
LINGUISTICS Peter Matthews
LITERARY THEORY Jonathan Culler
LOCKE John Dunn
LOGIC Graham Priest
LOVE Ronald de Sousa
MACHIAVELLI Quentin Skinner
MADNESS Andrew Scull
MAGIC Owen Davies
MAGNA CARTA Nicholas Vincent
MAGNETISM Stephen Blundell
MALTHUS Donald Winch
MAMMALS T. S. Kemp
MANAGEMENT John Hendry
MAO Delia Davin
MARINE BIOLOGY Philip V. Mladenov
THE MARQUIS DE SADE John Phillips
MARTIN LUTHER Scott H. Hendrix
MARTYRDOM Jolyon Mitchell
MARX Peter Singer
MATERIALS Christopher Hall
MATHEMATICAL FINANCE Mark H. A. Davis
MATHEMATICS Timothy Gowers
MATTER Geoff Cottrell
THE MEANING OF LIFE Terry Eagleton
MEASUREMENT David Hand
MEDICAL ETHICS Michael Dunn and Tony Hope
MEDICAL LAW Charles Foster
MEDIEVAL BRITAIN John Gillingham and Ralph A. Griffiths
MEDIEVAL LITERATURE Elaine Treharne
MEDIEVAL PHILOSOPHY John Marenbon
MEMORY Jonathan K. Foster
METAPHYSICS Stephen Mumford
METHODISM William J. Abraham
THE MEXICAN REVOLUTION Alan Knight
MICHAEL FARADAY Frank A. J. L. James
MICROBIOLOGY Nicholas P. Money
MICROECONOMICS Avinash Dixit
MICROSCOPY Terence Allen
THE MIDDLE AGES Miri Rubin
MILITARY JUSTICE Eugene R. Fidell
MILITARY STRATEGY Antulio J. Echevarria II
MINERALS David Vaughan
MIRACLES Yujin Nagasawa
MODERN ARCHITECTURE Adam Sharr
MODERN ART David Cottington
MODERN CHINA Rana Mitter
MODERN DRAMA Kirsten E. Shepherd-Barr
MODERN FRANCE Vanessa R. Schwartz
MODERN INDIA Craig Jeffrey
MODERN IRELAND Senia Pašeta
MODERN ITALY Anna Cento Bull
MODERN JAPAN Christopher Goto-Jones
MODERN LATIN AMERICAN LITERATURE Roberto González Echevarría
MODERN WAR Richard English
MODERNISM Christopher Butler
MOLECULAR BIOLOGY Aysha Divan and Janice A. Royds
MOLECULES Philip Ball
MONASTICISM Stephen J. Davis
THE MONGOLS Morris Rossabi
MOONS David A. Rothery
MORMONISM Richard Lyman Bushman
MOUNTAINS Martin F. Price
MUHAMMAD Jonathan A. C. Brown
MULTICULTURALISM Ali Rattansi
MULTILINGUALISM John C. Maher
MUSIC Nicholas Cook
MYTH Robert A. Segal
NAPOLEON David Bell
THE NAPOLEONIC WARS Mike Rapport
NATIONALISM Steven Grosby
NATIVE AMERICAN LITERATURE Sean Teuton
NAVIGATION Jim Bennett
NAZI GERMANY Jane Caplan
NELSON MANDELA Elleke Boehmer
NEOLIBERALISM Manfred Steger and Ravi Roy
NETWORKS Guido Caldarelli and Michele Catanzaro
THE NEW TESTAMENT Luke Timothy Johnson
THE NEW TESTAMENT AS LITERATURE Kyle Keefer
NEWTON Robert Iliffe
NIETZSCHE Michael Tanner
NINETEENTH‐CENTURY BRITAIN Christopher Harvie and H. C. G. Matthew
THE NORMAN CONQUEST George Garnett
NORTH AMERICAN INDIANS Theda Perdue and Michael D. Green
NORTHERN IRELAND Marc Mulholland
NOTHING Frank Close
NUCLEAR PHYSICS Frank Close
NUCLEAR POWER Maxwell Irvine
NUCLEAR WEAPONS Joseph M. Siracusa
NUMBERS Peter M. Higgins
NUTRITION David A. Bender
OBJECTIVITY Stephen Gaukroger
OCEANS Dorrik Stow
THE OLD TESTAMENT Michael D. Coogan
THE ORCHESTRA D. Kern Holoman
ORGANIC CHEMISTRY Graham Patrick
ORGANIZATIONS Mary Jo Hatch
ORGANIZED CRIME Georgios A. Antonopoulos and Georgios Papanicolaou
ORTHODOX CHRISTIANITY A. Edward Siecienski
PAGANISM Owen Davies
PAIN Rob Boddice
THE PALESTINIAN-ISRAELI CONFLICT Martin Bunton
PANDEMICS Christian W. McMillen
PARTICLE PHYSICS Frank Close
PAUL E. P. Sanders
PEACE Oliver P. Richmond
PENTECOSTALISM William K. Kay
PERCEPTION Brian Rogers
THE PERIODIC TABLE Eric R. Scerri
PHILOSOPHY Edward Craig
PHILOSOPHY IN THE ISLAMIC WORLD Peter Adamson
PHILOSOPHY OF BIOLOGY Samir Okasha
PHILOSOPHY OF LAW Raymond Wacks
PHILOSOPHY OF SCIENCE Samir Okasha
PHILOSOPHY OF RELIGION Tim Bayne
PHOTOGRAPHY Steve Edwards
PHYSICAL CHEMISTRY Peter Atkins
PHYSICS Sidney Perkowitz
PILGRIMAGE Ian Reader
PLAGUE Paul Slack
PLANETS David A. Rothery
PLANTS Timothy Walker
PLATE TECTONICS Peter Molnar
PLATO Julia Annas
POETRY Bernard O’Donoghue
POLITICAL PHILOSOPHY David Miller
POLITICS Kenneth Minogue
POPULISM Cas Mudde and Cristóbal Rovira Kaltwasser
POSTCOLONIALISM Robert Young
POSTMODERNISM Christopher Butler
POSTSTRUCTURALISM Catherine Belsey
POVERTY Philip N. Jefferson
PREHISTORY Chris Gosden
PRESOCRATIC PHILOSOPHY Catherine Osborne
PRIVACY Raymond Wacks
PROBABILITY John Haigh
PROGRESSIVISM Walter Nugent
PROJECTS Andrew Davies
PROTESTANTISM Mark A. Noll
PSYCHIATRY Tom Burns
PSYCHOANALYSIS Daniel Pick
PSYCHOLOGY Gillian Butler and Freda McManus
PSYCHOLOGY OF MUSIC Elizabeth Hellmuth Margulis
PSYCHOPATHY Essi Viding
PSYCHOTHERAPY Tom Burns and Eva Burns-Lundgren
PUBLIC ADMINISTRATION Stella Z. Theodoulou and Ravi K. Roy
PUBLIC HEALTH Virginia Berridge
PURITANISM Francis J. Bremer
THE QUAKERS Pink Dandelion
QUANTUM THEORY John Polkinghorne
RACISM Ali Rattansi
RADIOACTIVITY Claudio Tuniz
RASTAFARI Ennis B. Edmonds
READING Belinda Jack
THE REAGAN REVOLUTION Gil Troy
REALITY Jan Westerhoff
THE REFORMATION Peter Marshall
RELATIVITY Russell Stannard
RELIGION IN AMERICA Timothy Beal
THE RENAISSANCE Jerry Brotton
RENAISSANCE ART Geraldine A. Johnson
REPTILES T. S. Kemp
REVOLUTIONS Jack A. Goldstone
RHETORIC Richard Toye
RISK Baruch Fischhoff and John Kadvany
RITUAL Barry Stephenson
RIVERS Nick Middleton
ROBOTICS Alan Winfield
ROCKS Jan Zalasiewicz
ROMAN BRITAIN Peter Salway
THE ROMAN EMPIRE Christopher Kelly
THE ROMAN REPUBLIC David M. Gwynn
ROMANTICISM Michael Ferber
ROUSSEAU Robert Wokler
RUSSELL A. C. Grayling
RUSSIAN HISTORY Geoffrey Hosking
RUSSIAN LITERATURE Catriona Kelly
THE RUSSIAN REVOLUTION S. A. Smith
THE SAINTS Simon Yarrow
SAVANNAS Peter A. Furley
SCEPTICISM Duncan Pritchard
SCHIZOPHRENIA Chris Frith and Eve Johnstone
SCHOPENHAUER Christopher Janaway
SCIENCE AND RELIGION Thomas Dixon
SCIENCE FICTION David Seed
THE SCIENTIFIC REVOLUTION Lawrence M. Principe
SCOTLAND Rab Houston
SECULARISM Andrew Copson
SEXUAL SELECTION Marlene Zuk and Leigh W. Simmons
SEXUALITY Véronique Mottier
SHAKESPEARE’S COMEDIES Bart van Es
SHAKESPEARE’S SONNETS AND POEMS Jonathan F. S. Post
SHAKESPEARE’S TRAGEDIES Stanley Wells
SIKHISM Eleanor Nesbitt
THE SILK ROAD James A. Millward
SLANG Jonathon Green
SLEEP Steven W. Lockley and Russell G. Foster
SOCIAL AND CULTURAL ANTHROPOLOGY John Monaghan and Peter Just
SOCIAL PSYCHOLOGY Richard J. Crisp
SOCIAL WORK Sally Holland and Jonathan Scourfield
SOCIALISM Michael Newman
SOCIOLINGUISTICS John Edwards
SOCIOLOGY Steve Bruce
SOCRATES C. C. W. Taylor
SOUND Mike Goldsmith
SOUTHEAST ASIA James R. Rush
THE SOVIET UNION Stephen Lovell
THE SPANISH CIVIL WAR Helen Graham
SPANISH LITERATURE Jo Labanyi
SPINOZA Roger Scruton
SPIRITUALITY Philip Sheldrake
SPORT Mike Cronin
STARS Andrew King
Statistics David J. Hand
STEM CELLS Jonathan Slack
STOICISM Brad Inwood
STRUCTURAL ENGINEERING David Blockley
STUART BRITAIN John Morrill
SUPERCONDUCTIVITY Stephen Blundell
SYMMETRY Ian Stewart
SYNAESTHESIA Julia Simner
SYNTHETIC BIOLOGY Jamie A. Davies
TAXATION Stephen Smith
TEETH Peter S. Ungar
TELESCOPES Geoff Cottrell
TERRORISM Charles Townshend
THEATRE Marvin Carlson
THEOLOGY David F. Ford
THINKING AND REASONING Jonathan St B. T. Evans
THOMAS AQUINAS Fergus Kerr
THOUGHT Tim Bayne
TIBETAN BUDDHISM Matthew T. Kapstein
TIDES David George Bowers and Emyr Martyn Roberts
TOCQUEVILLE Harvey C. Mansfield
TOPOLOGY Richard Earl
TRAGEDY Adrian Poole
TRANSLATION Matthew Reynolds
THE TREATY OF VERSAILLES Michael S. Neiberg
THE TROJAN WAR Eric H. Cline
TRUST Katherine Hawley
THE TUDORS John Guy
TWENTIETH‐CENTURY BRITAIN Kenneth O. Morgan
TYPOGRAPHY Paul Luna
THE UNITED NATIONS Jussi M. Hanhimäki
UNIVERSITIES AND COLLEGES David Palfreyman and Paul Temple
THE U.S. CONGRESS Donald A. Ritchie
THE U.S. CONSTITUTION David J. Bodenhamer
THE U.S. SUPREME COURT Linda Greenhouse
UTILITARIANISM Katarzyna de Lazari-Radek and Peter Singer
UTOPIANISM Lyman Tower Sargent
VETERINARY SCIENCE James Yeates
THE VIKINGS Julian D. Richards
VIRUSES Dorothy H. Crawford
VOLTAIRE Nicholas Cronk
WAR AND TECHNOLOGY Alex Roland
WATER John Finney
WAVES Mike Goldsmith
WEATHER Storm Dunlop
THE WELFARE STATE David Garland
WILLIAM SHAKESPEARE Stanley Wells
WITCHCRAFT Malcolm Gaskill
WITTGENSTEIN A. C. Grayling
WORK Stephen Fineman
WORLD MUSIC Philip Bohlman
THE WORLD TRADE ORGANIZATION Amrita Narlikar
WORLD WAR II Gerhard L. Weinberg
WRITING AND SCRIPT Andrew Robinson
ZIONISM Michael Stanislawski

Available soon:
KOREA Michael J. Seth
TRIGONOMETRY Glen Van Brummelen
THE SUN Philip Judge
AERIAL WARFARE Frank Ledwidge
RENEWABLE ENERGY Nick Jelley

For more information visit our website

www.oup.com/vsi/
 Richard Earl

TOPOLOGY
A Very Short Introduction
 Great Clarendon Street, Oxford, OX2 6DP, United Kingdom
Oxford University Press is a department of the University of Oxford. It furthers the University’s
objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a
registered trade mark of Oxford University Press in the UK and in certain other countries
© Richard Earl 2019
The moral rights of the author have been asserted
First edition published in 2019
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, without the prior permission in writing of Oxford
University Press, or as expressly permitted by law, by licence or under terms agreed with the
appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of
the above should be sent to the Rights Department, Oxford University Press, at the address above
You must not circulate this work in any other form and you must impose this same condition on any
acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2019949429
ISBN 978–0–19–883268–3
ebook ISBN 978–0–19–256899–1
Printed in Great Britain by Ashford Colour Press Ltd, Gosport, Hampshire
Links to third party websites are provided by Oxford in good faith and for information only. Oxford
disclaims any responsibility for the materials contained in any third party website referenced in this
work.
 In memory of
Dan Lunn.
Friend, colleague, tutor.
Contents

Acknowledgements

List of illustrations

1 What is topology?

2 Making surfaces

3 Thinking continuously

4 The plane and other spaces

5 Flavours of topology

6 Unknot or knot to be?

Epilogue

Appendix

Historical timeline

Further reading

Index
Acknowledgements

Thanks go to Martin Galpin, Andy Krasun, Natalie Lane, Marc Lackenby,

Kevin McGerty for their comments on and help with draft chapters. Thanks
especially to Marc for his encouragement to write the book.
List of illustrations

1 London underground maps

(a) Archive PL / Alamy Stock Photo(b), © TfL from the London Transport Museum collection.

2 Examples of polyhedra

3 Manipulations of a cube to find its Euler number

4 The Platonic solids

5 Matt Parker and his football

Matt Parker / YouTube.

6 Diagonals in a square must intersect

7 Two non-planar graphs

8 Complicated Jordan curves

(a) Get Drawings
(b) David Eppstein/ Wikimedia Commons / Public Domain.

9 Original figure from Flatland showing how A Sphere is perceived by A Square

The History Collection / Alamy Stock Photo.

10 The unknot and the trefoil

11 The torus

12 Making a cylinder

13 Making a torus

14 Gluing instructions for a triangle, pentagon, and square

15 Valid and invalid subdivisions of a sphere

16 Connected sums with tori

17 The Möbius strip

(b) Dotted Yeti / Shutterstock.com.

18 Moving an oriented loop around a Möbius strip

19 The Klein bottle and projective plane

20 The real line and complex plane

21 Visualizing Riemann surfaces

22 Distance, speed, and acceleration on a journey

23 Examples of graphs

24 Continuous and discontinuous functions

25 The rigorous definition of continuous and discontinuous

26 The intermediate value theorem

27 The blancmange function

28 Visualizing different metrics

29 Graphs of functions of two variables

30 Open balls in the plane

31 Open sets in the plane

32 Visualizing convergence

33 Relating to a disconnected subspace

34 Relating to path- connectedness

35 Examples of a minimum, maximum, and saddle point

36 Critical points of functions on a torus and sphere

37 Vector fields on the sphere and torus

38 Examples calculating indices of vector fields

39 Loops on a torus

40 Functions on the disc

41 The unknot and trefoils

42 The three Reidemeister moves

43 The granny and reef knots

44 Prime knots with seven or fewer crossings

45 The knot group of a trefoil

46 Links involved in the skein relation

47 Simple examples of links

48 Calculating the Alexander polynomial of a trefoil

Chapter 1
What is topology?

As you read this, passengers the world over are travelling on metro (or
subway or underground) trains. There are around 60 billion individual
journeys made annually on such metro systems. But whether this be in
Tokyo, London, São Paolo, New York, Shanghai, Paris, Cairo, Moscow,
those travellers are perusing maps for their journeys that are crucially
different from maps in atlases or seen on geography classroom walls.
Foremost in the minds of those passengers are the connections they need to
make—getting out at the right station and changing to the correct new line.
They are not interested in whether the map’s left–right lines do indeed run
west–east, or whether they really did make a right angle turn when they
changed lines, as depicted on the metro map.

The oldest metro network in the world is the London Underground. When
first produced, the underground maps superimposed the different train lines
onto an actual (geographically accurate) map of London, as shown in
Figure 1(a). A first version of the current map was designed by Harry Beck
in 1931 as in Figure 1(b). Beck’s map, and the current underground map,
are not wrong. Rather they transparently show information important to
travellers—for example, the various connections between lines and the
number of stops between stations. It is an early example of a topological
map and demonstrates the different focus of topology—which is all about
shape, connection, relative position—compared with that of geometry (or
geography) which is about more rigid notions such as distance, angle, and
area.
1. London underground maps (a) Geographically accurate 1908 map, (b) Beck’s topological
1931 map.
Topology is now a major area of modern mathematics, so you may be
surprised to learn that an appreciation of topology came late in the history
of mathematics. The word topology—meaning ‘the study of place’—wasn’t
even coined until 1836. (‘Geometry’ by comparison is an ancient Greek
word and ‘algebra’ is an Arabic word, with its mathematical meaning
dating back to the 9th century.) Just why this was the case is not a simple
question to address, though we will see some aspects of topology developed
as mathematicians sought to put their subject on a more rigorous footing.
Topology is a highly visual subject that lends itself to an informal treatment
and this book will give you a sense of topology’s ideas and its technical
vocabulary.

A topologist’s alphabet
As a first example, to convey how differently topologists and geometers see
objects, consider what capital letters a topologist would deem to be the
‘same’. Using the sans serif font, the four letters

EFTY

are all topologically the same. They are not congruent, meaning that none
of the letters can be picked up, and rotated or reflected, and then put down
as one of the other letters. But I hope you can envisage, if allowed to bend,
stretch, or shrink the letters, how any of them might be transformed into one
of the others.

To a topologist these four letters are homeomorphic to one another. The

geometer would notice that the angle made by the arms of the Y is different
from any angle found in the other letters. The topologist, on the other hand,
would be happy to flatten the arms of the Y, and stretch its body a little, to
give the T shape. Likewise the E could have its bottom rung bent around to
the vertical, and then shortened somewhat, to make the F. Finally doing the
same to the top of the F would make a T on its side. These four letters can
be continuously deformed into one another and back again. Broadly
speaking this is what it is to be homeomorphic, to be topologically the
same.
But what is it about these letters that makes them topologically different
from other letters? Another collection of letters that are topologically the
same as one another is:

C G I J L M N S U V W Z.

Topologically all of these are equivalent to a line segment and it’s not hard
to imagine how each might be formed by bending and stretching a suitably
mutable letter I. So hopefully you’re convinced that the letters in the second
list are all homeomorphic to one another, but what makes this second
collection topologically different from the first list?

Note, for each letter E, F, T, Y, that every point lies on a distorted bit of line
with one exception. In each of these letters there is a single point that might
be described as a T-junction. These T-junctions are highlighted below.

One way in which these T-junctions are special is that, if removed, the
remainder of the letter is disconnected into three parts; the removal of any
other point would leave just two parts remaining. In whatever ways we
might bend and deform an E the deformed version would still include a
single T-junction. As none of the second set C … Z has such a T-junction
then none of them can be a deformed version of an E (or an F, T, or Y ).

This gives a genuine sense of how mathematicians resolve the question: are
two shapes the same topologically? This either amounts to finding some
means of continuously deforming one into the other, or involves finding
some topological invariant of one that does not apply to the other. The
word invariant is used in different contexts in mathematics: for example, if
you shuffle a pack of cards, there will still remain fifty-two cards afterwards
and four suits, these are invariants; but the top card may have changed and
the jack of clubs may no longer come before the eight of diamonds, and so
such facts aren’t invariants of a shuffle. A topological invariant is
something immutable about a shape, no matter how we stretch and deform
it. In the above example we used the presence of a T-junction as our
topological invariant. You might note that an E includes four right angles
whilst an F contains only three. The presence of four right angles is a
geometric invariant and so shows that E and F are not congruent (i.e. not
geometrically the same), but—working topologically—we are permitted to
unbend these right angles and so right angles are not important from a
topological point of view. Rather they’re mutable aspects of a shape and not
topological invariants.

The remaining twenty-six letters, grouped topologically, break down as:

DO, KX, AR, B, PQ, H.

You might want to take a moment thinking about what makes an A different
from a P or O different from Q. In fact, the O introduces an important
topological invariant that separates it from both I and E. The shape of the O
is different as it makes a loop. Technically O is not simply connected, a
topic we will discuss more in Chapter 5.

Euler’s formula
One of the first topological results was due to Leonhard Euler (pronounced
‘oil-er’), a titan of 18th-century mathematics and one of the most prolific
mathematicians ever; his formula dates to around 1750. The result relates—
at first glance—to polyhedra, three-dimensional objects such as cubes and
pyramids (Figure 2). It is also so fundamental—a straightforward
observation at least—that it is surprising ancient Greek mathematicians
missed it.
2. Examples of polyhedra (a) A cube, (b) A Square-based pyramid.

Looking at the cube, we can see that it is made up of vertices (the corners
of the cube—the singular is ‘vertex’), these vertices being connected by
edges and that these edges then bound (square) faces. For the cube the
number of vertices V equals 8, there are E = 12 edges and F = 6 faces. For
the (square-based) pyramid we have V = 5, E = 8, and F = 5. No pattern
may be evident immediately but if we include the four other so-called
Platonic solids—tetrahedron, octahedron, dodecahedron, icosahedron
(Figure 4)—and other familiar polyhedra, we create Table 1.

Table 1. Vertices, edges, faces for various polyhedra

4. The Platonic solids (a) Tetrahedron, (b) Cube, (c) Octahedron, (d) Dodecahedron, (e)
Icosahedron.

(We shall see soon that the truncated icosahedron is familiar to us—just not
by that name!)

For all the geometry that the ancient Greeks knew, it seems striking that this
pattern eluded them, but we will prove now—or more honestly sketch a
proof of—Euler’s formula which states, for a polyhedron with V vertices,
E edges, and F faces, that

In the proof our aim will be to begin with a polyhedron and manipulate it in
certain ways—for example, we might remove or subdivide faces—but in all
cases we will carefully track the effect our manipulation has (if any) on the
number . If, after such manipulations, we arrive at a simplified
situation where we know what equals, and we know the effects
our manipulations had on that number, then we may be able to work
backwards to find what was originally.

We begin then with a polyhedron, and first remove one of the faces. This
has the effect of reducing by 1 as F has decreased by 1. Now
that the polyhedron has a missing face—effectively the polyhedron has
been punctured—it can be flattened into the plane, taking care that all the
vertices, edges, faces present on the punctured polyhedron remain and are
connected in the plane in the same manner they were on the punctured
polyhedron. For example, if we removed one face from a cube and flattened
the remaining cube then we would have something like Figure 3(a).
3. Manipulations of a cube to find its Euler number (a) A flattened, punctured cube, (b) With
flattened faces triangulated, (c) Removing a triangle, (d) Removing a triangle.

The next manipulation is to subdivide each of the flattened faces into

triangles—as has been done to the flattened cube in Figure 3(b).
Introducing a single triangle has the effect of increasing F by 1—what was
one face becomes split into two—of increasing E by 1—the new edge,
introduced to make a triangle—and doesn’t change V. So there is no overall
effect to as we keep introducing triangles; the increase of 1 to
F, a term that is added in the formula, is precisely balanced by the increase
in E which is a term we subtract. When this has been done for each
flattened face (as in Figure 3(b)) then is still just one less than
it was originally.

We now remove the triangles one at a time. For example, if we remove the
bottom triangle from Figure 3(b) to make Figure 3(c), then we remove one
edge and one face and, by the same reasoning as before, this has no overall
effect on .

Similarly, we might then remove the right-most triangle to create Figure

3(d), the manipulation again having no effect on . But the
bottom triangle of Figure 3(d) is connected differently. If we remove that
triangle then we remove 2 edges, 1 face, and 1 vertex. The algebra is a little
more complicated this time, but again removing 1 vertex and 1 face means
goes down 2 but this is countered by removing 2 edges as E is
a term we subtract. Or if you prefer more formal algebraic reasoning, we
are just saying

Let’s summarize what’s happened so far:

• We removed a face and decreased by 1.

• We flattened the polyhedron into the plane—all V, E, F remain, so no
change to .
• We subdivided the flattened faces into triangles—this had no effect on
.
• We kept removing triangles from the edge of the flattened polygon—each
removal having no effect on

Eventually only a single triangle will remain, having removed all others. A
triangle has a single face, three vertices, and three edges, so that
equals . This is the value of that we
finish with. The only manipulation that ever changed was that
very first removal of a face which reduced it by one; initially then it was the
case that

This ‘sketch proof’ was given by Augustin-Louis Cauchy in 1811. It’s

worth highlighting there are several ‘i’s still to be dotted to make a proof
with which a professional mathematician would be happy, but also noting
how much of the idea of the proof is genuinely here. We didn’t take care
describing how we removed triangles from the boundary of the flattened
polygon; if we’d been careless we might have removed a triangle that
disconnected the flattened polygon into two separate polygons, and we
should have taken time to make sure such an occasion can always be
avoided. Other issues will become more apparent in Chapter 2 but these ‘i’s
can indeed be dotted. In Proofs and Refutations, the Hungarian philosopher
Imre Lakatos used the specific example of Euler’s formula, and historical
efforts to prove it, to highlight how hard it can sometimes be to generate a
watertight proof and to also raise the question of when a theorem properly
becomes part of mathematics or has mathematical content.

It’s also worth mentioning that René Descartes had, over a century before
Euler, demonstrated a theorem for polyhedra that is equivalent to Euler’s
formula; his theorem was in terms of ‘angular defects’ at vertices. All of a
sudden we are back in the geometrical world and it’s less than clear that
there is a genuinely new subject, an importantly different way of
mathematical thinking, that Euler’s fingertips were brushing against.
Euler’s formulation encourages appreciation of the result as something a
little new—in Euler’s terminology, rather than Descartes’s, it’s much clearer
that the connection of the vertices, edges, and faces is what counts, but
historically we are still a long time from a deeper appreciation of topology
as a fundamental mode of mathematical thinking.

There are five Platonic solids

Platonic solids are polyhedra with regular faces that are all congruent
(geometrically the same) and which meet in the same manner at each
vertex. There are infinitely many regular polygons—equilateral triangles,
squares, regular pentagons, etc.—but in 3D it turns out that there are just
five regular solids which have been known since antiquity. These are shown
in Figure 4 and with Euler’s formula we can show there are just these five.

Consider a regular polyhedron with V vertices, E edges, and F faces. As the

solid is regular then each face is bounded by the same number of edges;
let’s call this number n. Likewise there is a common number m for how
many edges meet at each vertex. So, with the cube, (the faces are
squares) and (three edges meet at each vertex). Continuing with the
cube as our example for now, think about how we can make a cube by
gluing together the edges of six squares.
We begin with 6 separate squares so that, before any gluing happens, there
are 6 squares, 24 edges, and 24 vertices. Note that to make the cube it takes
two ‘unglued’ edges to make each edge of the cube (which agrees with
there being 24/2 = 12 edges) and it takes three ‘unglued’ vertices to make a
single vertex of the cube (again there are 24/3 = 8 vertices). More generally,
when we have F faces each with n edges, we would have nF edges before
any gluing. It takes two of these unglued edges to make a single edge of the
polyhedron which then has edges. There are as many unglued
vertices as unglued edges, namely , and these will be glued
together to make vertices on the polyhedron as it takes m
unglued vertices to make one glued vertex on the solid.

Putting these expressions for V and F into Euler’s formula we get

(The equation has been rearranged to make F the subject of

the equation so that .) We can then divide both sides of the
above equation by 2E and rearrange to find

As 1/E is positive this means that

So m and n can’t both be very large as then and would be very

small and their sum would not exceed . Also recall m and n are positive
whole numbers, so there aren’t many options and it’s not hard to find all
their possible values.
It’s impossible for both m and n to exceed 4 as then would be
less than . So either or or or
(with perhaps more than one of these being true). For example, if ,
the only n for which the inequality is true are . If then
equals or less and the inequality is not
true. For the three cases of and we have

A similar calculation for leads to , , and when

we find , . In all these five cases we can use the previous
formulas to work out the numbers of vertices and faces
. We can put the full details into Table 2.

Table 2. Possible m and n values for the Platonic solids

If we are seeking to be rigorous here, we should really point out that the
previous calculations show that there are at most five possible pairs of
values that m, n can take. Those calculations limit the possibilities, but do
not necessarily mean that there is a Platonic solid for each of these cases,
nor preclude there being more than one Platonic solid for permitted m and n
—it might be that there are two different Platonic solids with three
pentagons meeting at each vertex. Listed in Table 2 are the five Platonic
solids, and so we can see that there is at least one solid for permitted m, n.
And it’s not hard to appreciate why there can be at most one. In the case
where , then three squares meet at each vertex; seemingly this
only tells us something about parts of the solid, but if we follow this recipe
of attaching three squares at each vertex then there is only one way to
progress building up the solid—it’s not clear that this recipe will actually
lead to a complete solid, but it does show that there can be at most one
Platonic solid for each allowed m, n.

(As an aside, you may have noticed that and , and

, are solutions if we permit E to be infinite. These ‘solutions’
correspond to tessellations of the plane where four squares meet at a vertex,
where three regular hexagons meet at a vertex (as with honeycombs), and
where six equilateral triangles meet at a vertex. There are also some
patterns apparent in Table 2 for the values of V, E, F for the cube and
octahedron, and likewise the dodecahedron and icosahedron. This is
because these solids are dual to one another—this means that the midpoints
of the faces of a cube make an octahedron, and vice versa; likewise the
dodecahedron is dual to the icosahedron and the tetrahedron is self-dual in
this sense.)

Footballs
In 2017 the mathematics popularizer Matt Parker began a petition seeking
to get road signs to football stadia corrected.

You may not have noticed the inaccuracy of such signs in the past, perhaps
being happy just to know you’re travelling the right way for the game. But
it’s clear (Figure 5) that the sign’s football does not resemble the actual
football that Matt is carrying. A football’s surface is made from pentagons
and hexagons and the everyday football is more formally known as a
truncated icosahedron. (It can be created from an icosahedron by planing
down, around each vertex, the five edges meeting there. If we plane down
one-third of each of those five edges, we create a new pentagonal face and
continued planing eventually shrinks all the triangular faces to hexagons.) I
think though the irksome principle for Matt was not that the sign’s football
was badly drawn, it was in fact impossibly drawn.

5. Matt Parker and his football.

There is no way that a sphere can be made by stitching together hexagons

as shown on the sign. That would be an example where and ,
using the previous notation, and whilst we can cover the plane in this way
—which may be why the sign looks plausible at first glance—making a
football this way is mathematically impossible.

In fact, Euler’s formula shows us how many pentagons and hexagons there
are on a football. Recalling how to truncate an icosahedron we see there are
as many pentagons as original vertices (12) and hexagons as original faces
(20), but Euler’s formula can show this is the only way to construct such a
football. Say a football has P pentagonal faces and H hexagonal faces.
Then, before gluing these together, we have 5P + 6H unglued edges and the
same number of unglued vertices. Looking at Matt’s ball we can see that (i)
two unglued edges are needed to make an edge on the football and (ii) three
unglued vertices make a vertex with (iii) two hexagons and one pentagon
meeting at a vertex; from (iii) we see there are twice as many ‘unglued’
vertices collectively on the hexagons as on the pentagons. So
If we put these values into Euler’s formula we find that

which simplifies and rearranges to P =12 and the equation 10P = 6H yields
H =20. This, then, is the only way to make a football if we follow the rules
(i), (ii), (iii).

Graph theory
Let’s change tack a little and consider the following problem. The square
PQRS in Figure 6 has diagonally opposite vertices P and R, Q and S. If we
were to draw curves from P to R, and from Q to S, curves which remain
within the square as in Figure 6, then surely those curves would have to
cross at least once. (In Figure 6 there are three intersections.) This seems
obvious—and is true—but how would you go about proving this?

6. Diagonals in a square must intersect Curves PR and QS in the square PQRS.

Before more is said, it might be worth stressing how characteristic of a

topological question this is. The curves PR and QS need to connect their
end points. Those curves don’t need to be polygonal, or have well-defined
gradients, or be defined by specific functions. They need to connect the end
points in some continuous sense—fuller details in Chapter 3—but they are
otherwise general paths from P to R and from Q to S that remain in the
square.

At first glance, this problem might seem quite removed from the polyhedra
we were just discussing. However, Figure 6 doesn’t look that different from
Figures 3(a)–(d). We have vertices (P, Q, R, S and any points where the
curves PR and QS meet), edges running between these vertices (though
admittedly they’re now curved), and we have faces bounded by those edges.
It was crucial to the proof of Euler’s formula that for each
of Figures 3(a)–(d). If we also include the outside region as a face—
essentially the one removed so we could flatten the polyhedron—then we
arrive back at Euler’s formula . (By this reckoning ,
, in Figure 6.)

So suppose, somehow, we could draw curves PR and QS in the square

PQRS which don’t intersect. We’d then find

• the four corners P, Q, R, S.

• the square’s four sides and the curves PR, QS.
• the outside of the square, above PS, below QR, right of PQ, left of
SR.

But this leaves us with and so such a scenario is

impossible by Euler’s formula.

Graph theory is an area of mathematics that models networks in a wide

sense: physical, biological, and social systems, variously representing
transport networks, computer networks, website structure, evolution of
words across languages and time in philology, migrations in biology, etc. A
graph is a collection of points called vertices, with these vertices connected
by edges. We will also assume that graphs are connected, meaning that
there is a walk between any two vertices along the edges. This definition
may be extended to include one-way edges—directed graphs or digraphs—
and weights might be introduced to edges representing the difficulty—in
terms of time, distance, or cost—of travelling along a particular edge.

Some graphs are planar, meaning that they can be drawn in the plane
without their edges crossing (at points that aren’t vertices). The two graphs
K5 and K3,3 in Figure 7 are importantly not planar. The complete graph on 5
vertices, denoted K5, has a single edge between each pair of the 5 vertices
making 10 edges. You might think that K5 is planar as it’s drawn in Figure
7(a)—the point is that, so drawn, many of the edges’ crossings don’t occur
at vertices and to deem these crossings as vertices would mean we were no
longer considering K5 which has only 5 vertices. If we could properly draw
K5 in the plane there would be 10 triangular faces v1v2v3, v1v2v4, through to
v3v4v5. We’d then have that equals and
so K5 is not planar.

7. Two non-planar graphs (a) The complete graph K5, (b) The complete bipartite graph K3,3.

K3,3 is the complete bipartite graph between two trios of vertices. A

somewhat subtler argument shows K3,3 is not planar. Note that K3,3 has
vertices and edges. If drawn in the plane this would mean
. But a face of K3,3 would have at least four edges as
its perimeter necessarily runs from a v to a w to a different v and to a w and
only then may return to the original v. So, counting the edges by going
around all the faces, we would get a total of at least edges.
However, as an edge can bound at most two faces this would mean we’d
have at least edges, which is our required contradiction.

The Polish mathematician Kazimierz Kuratowski proved in 1930 that a

graph is planar precisely when neither a copy of K5 nor K3,3 can be found
within the graph. We will in due course see that K3,3 can be drawn on other
surfaces such as a torus (Figure 13(c)).

Nasty surprises
Euler arrived at his formula a century before the word topology was coined.
His formula is characteristic of a visual side of topology naturally aligned
with geometry. But topology, as a subject, would develop along various
themes and in particular had an important role in the foundational work
mathematicians were doing around the start of the 20th century. As I hinted
earlier, topology’s rise may have been hampered by a traditional mindset
that some of its questions had obvious answers. For example Camille
Jordan, as late as the 19th century, proved the following: a curve in the
plane, which does not cross itself and which finishes back where it began—
a curve which we would now call a Jordan curve—splits the plane into
two regions, the technical phrase for these regions being connected
components. One of these regions is bounded, the inside, and the other is
unbounded, the outside, and this is the Jordan curve theorem. Earlier
mathematicians would have happily thought this obvious and the first
rigorous proof didn’t appear until 1887. You may agree with those earlier
mathematicians that the result can be safely assumed. Maybe even the
Pollock-like Jordan curve in Figure 8(a) does not sway your view of the
intuitiveness of the result.
8. Complicated Jordan curves (a) A more complicated Jordan curve, (b) A Jordan curve with
positive area.
In Figure 8(b) is the Knopp–Osgood curve which, for all its fractal-like
appearance, is a Jordan curve. Astonishingly it has a positive area—that is
the curve itself has positive area, we’re not referring to some region that it
bounds. Would you have said a moment ago that it’s obvious that curves
can’t themselves have area?

You shouldn’t worry too much in the sense that most things those early
mathematicians thought to be true turned out to be true, once properly
understood and qualified, but mathematicians towards the end of the 19th
century were getting nervous about the rules and assumptions that
mathematics relied on.

A related problem within topology at that time was rigorously defining

what dimension means. Again this had previously been treated as an
intuitive concept, only for mathematicians to begin finding space-filling
curves that pass through every point in the plane or other weird-and-
wonderful spaces that can reasonably be assigned dimensions that are not
whole numbers—spaces that would now be called fractals.

An early theme of topology was this general topology or point-set

topology seeking to address what it means to be a set, to be a space, etc.
Metric and topological spaces were introduced—to be discussed in Chapter
4—each being attempts to describe general structures where continuity
could be defined. Set theory deals with collections that are essentially just
things-in-a-bag. This general topology sought to define ways in which
objects might be considered ‘close’ to one another, with the aim being to
define continuity in a broad setting.

A Flatland mindset
The novella Flatland: A Romance of Many Dimensions, written in 1884 by
Edwin Abbott, is a satire on Victorian mores. The narrator is ‘A Square’, an
inhabitant of Flatland, a planar world having just two dimensions. The
culture of Flatland and the logistics of living in two dimensions are fully
described, implicitly highlighting some of the narrow-mindedness of
Victorian culture—for example, women are one- rather than two-
dimensional beings. The story doesn’t explicitly discuss topology, but in its
description of worlds with different dimensions and implications for the
inhabitants, it provides a useful metaphor for understanding certain aspects
of topology.

For example, A Square is visited at one point by A Sphere. Being a three-

dimensional object, A Sphere can only be perceived by Flatlanders as a
circular cross-section (Figure 9). By moving up and down—relative to
Flatland’s plane—A Sphere can grow, shrink, and even disappear entirely.
In a similar manner, to truly understand the topology of a space, we have to
begin thinking like inhabitants of that space.

9. Original figure from Flatland showing how A Sphere is perceived by A Square.

Topology is often characterized as rubber-sheet geometry. It’s a somewhat

clichéd metaphor, but it’s also slightly inaccurate. It gives a correct sense of
topology being more about shape and less rigid than geometry in its focus.
On the other hand, in Chapter 6 we discuss knots, and as a (genuine) knot—
like the trefoil—and the (unknotted) circle (Figure 10) cannot be
continuously deformed into one another in 3D then you might be tempted to
say the circle and trefoil are not homeomorphic, but they are.
10. The unknot and the trefoil (a) The unknot, (b) The trefoil knot.

The knottedness of the trefoil says something about its position in 3D. In
fact, all knots are homeomorphic to a circle. To better appreciate this, you
might imagine life as an ant living on either the circle or trefoil. As the ant
moves around either the unknot or trefoil it has a sense of being on a loop,
but the ant has no notion of whether it is living on a knot. It is only by being
able to view things from outside the two loops, and looking on from a
position in the ambient space, that we are able to recognize one loop as
knotted as compared with the other. This Flatland mindset will prove useful
again later when we meet subspaces.

Topology would advance on various fronts in the 19th and 20th centuries.
In particular, Bernhard Riemann would early on show the usefulness of a
‘topological mindset’, introducing Riemann surfaces into the study of
polynomial equations and demonstrating some deep connections between
topology and many other areas of mathematics.
Chapter 2
Making surfaces

The shape of surfaces

Recall Euler’s formula states for a polyhedron. Various
details of the proof were brushed under the carpet, the most significant of
these being the claim that, once a face is removed from a polyhedron, the
remaining polyhedron can be flattened into the plane. This was true for the
polyhedra we were considering, but the claim says something important
about the shape of the remaining polyhedron that was perhaps
unintentional. In any case, the next example will either make us question
what we mean by a polyhedron or have us looking to generalize Euler’s
formula.

For Figure 11(a)’s ‘polyhedron’, a count of vertices, edges, and faces shows
that , , , giving which seems to
disprove Euler’s formula. We are left with a few alternatives: either the
object in Figure 11(a) should not be considered a polyhedron, or we need to
restrict Euler’s formula to a certain type of polyhedron, or we need to adapt
and generalize Euler’s formula into a version that remains true for a broader
family of polyhedra.
11. The torus (a) A polyhedron with one hole, (b) A torus.

The most obvious issue with this new ‘polyhedron’ is the hole through its
middle. This is not immediately reason enough to exclude it as a
polyhedron, but this shape, once a face is removed, does not leave a
remainder that can be flattened into the plane, making our earlier proof
invalid. We need either to restrict Euler’s formula to polyhedra without
holes, or we need to work out the correct values for polyhedra
with holes.

Recalling the rubbery nature of topology, we might recognize that the

polyhedra of Chapter 1 all had the same underlying spherical shape. If
allowed to smooth out those polyhedra—the pointy vertices and the ridgy
edges—we could transform each of those polyhedra to a sphere, covered
with a patchwork of curved faces, just like Matt Parker’s football (Figure 5)
was covered with curved pentagons and hexagons. But however we smooth
down our new polyhedron we can’t make a sphere, rather we would make a
torus, the shape of a doughnut with a hole through it (Figure 11(b)).

Perhaps then all of the examples of Chapter 1—including the proof of

Euler’s formula—point to the Euler number of the sphere being 2. And
Figure 11(a) is a first example suggesting the Euler number of the torus is 0;
this would mean the number equals 0 however we divide up
the torus. All this could become quite involved unless we have a way of
efficiently describing surfaces—including more complicated ones than the
sphere or torus—and for systematically calculating their Euler numbers,
that is the value common to all surfaces of a certain underlying
shape.

Gluing surfaces together

A useful way of constructing surfaces is to begin with a polygon and
pairwise glue together the edges of the polygon, the way a model kit might
direct you to ‘glue tab A to tab B’. How might we make a torus in this
manner? If we begin with a (suitably elastic) square (Figure 12(a)), bend it
around (Figure 12(b)), and glue the opposite edges e1 and e3 so that the
vertices v1, v2 get glued respectively to v4, v3, and likewise for all other
opposite points of e1 and e3—as signified by the two arrows—then we will
make the cylinder drawn in Figure 12(c). Note that the edges e2 and e4 on
the original square have become the two circular ends of this cylinder. We
can then glue together these circular ends; if we do this so that the opposite
points of the original e2 and e4 are glued together, then we make a torus as
in Figure 13(a).

12. Making a cylinder (a) A square with identified edges, (b) Making the cylinder, (c) A
cylinder.
13. Making a torus, (a) Torus with v and e1, e2 drawn, (b) Square with gluing instructions, (c)
K3,3 on the torus.

Note that the four corners of the original square—denoted v1, v2, v3, v4—
have all been glued together to make a single point v on the torus. Similarly,
the edges e1 and e3 have been glued together to form a circle going around
the outside of the torus and e2 and e4 have been glued together to form a
different circle going through the hole of the torus, these two circles
meeting at the point v.

Importantly, the torus’s shape is fully described by a square with directions

for how the edges are to be glued together. Mathematicians would draw this
square-with-gluing-instructions as shown in Figure 13(b). The single arrows
—and importantly the directions in which they’re drawn—show how those
two edges are glued, and the double arrows (again noting directions) tell us
how the other pair of edges is glued.

As an aside, the graph K3,3, which we saw can’t be drawn in the plane
(Figure 7(b)), can be drawn on the torus (Figure 13(c)). The reason K3,3 can
be drawn on the torus is because a torus has a lower Euler number of 0 and,
arguing as before, it can then be shown that F = 3. If you look carefully at
Figure 13(c) you will see that there are indeed three faces (quadrilaterals
v1w2v2w1 and v3w3v2w2 and a single hexagonal face v1w2v3w1v2w3).

How does all this help with determining Euler numbers? With a single
square and gluing directions, we have been able to make an object with the
shape of a torus and this is certainly a conciser description of a torus than
Figure 11(a) for which , , . But is this glued square
enough to work out the Euler number of a torus? The answer is yes if we’re
careful when thinking about just how the original vertices and edges glue
together. On the original square there was just one face—the square’s
interior—four unglued edges (labelled e1, e2, e3, e4), and four unglued
vertices (labelled v1, v2, v3, v4). However, once we’ve followed the gluing
directions, those four edges have become the two circular edges on the torus
and the four vertices have become one point v, as in Figure 13(a). And there
is still one ‘face’ on the torus—the square’s interior has been stretched to
become all of the torus except those circles and v. So when we use this
glued square to calculate the Euler number of the torus we get the answer

which agrees with our calculation from Figure 11(a).

If you prefer Figure 13(a) showing the torus with the four glued vertices
becoming one, and the four edges become two loops on the torus, then
Figure 13(b) will only appear as an unfinished DIY job. But the single
surface obtained from gluing the triangle, pentagon, and square in Figure
14, following all the gluing directions a, b, … f according to the arrows,
might reasonably start stretching your visualization skills. But we can still
work out the Euler number of this chimera and seek to understand just what
surface we are looking at. This time there are three faces (the triangle,
pentagon, and square) and the twelve unglued edges of the polygons make
six glued edges a, b, c, … f on the surface. How many vertices will we
ultimately have? There were twelve unglued vertices originally but various
of these get glued together as we make the surface. For example, v1 and v5
are glued together as they are both at the rear end of the edge marked a. In
fact, we can chase around these gluings to see just how many vertices we
have:
14. Gluing instructions for a triangle, pentagon, and square.

v1 and v5 are glued (rear end of a)

v5 and v9 are glued (rear end of f )
v9 and v4 are glued (rear end of d)
v4 and v12 are glued (front end of f )
v12 and v2 are glued (rear end of c)
v2 and v7 are glued (rear end of b)
v7 and v11 are glued (front end of e)
v11 and v1 are glued (front end of c)

So eight different (unglued) vertices v1, v2, v4, v5, v7, v9, v11, v12 all get
glued together as a single vertex on the surface. In a similar fashion we can
see that the remaining four vertices v3, v6, v10, v8 get glued together (in that
order). So once made, the surface has 2 vertices, 6 edges, and 3 faces giving
an Euler number of . Just what surface have
we made?

Getting the right answer: subdivisions

We need now to make clear just what surfaces we are considering—closed
surfaces—and how they can be divided up into vertices, edges, and faces. A
closed surface is one without a boundary, such as a torus or sphere, but not
the cylinder of Figure 12(c). Our process of making a torus begins with a
square and at that point our surface has a boundary consisting of its four
edges; when we glue two edges to make a cylinder then the surface still has
a boundary, namely its top and bottom circles. Once the torus has been
made, no boundary points remain unglued.

Secondly, we can only calculate the correct Euler number of a closed

surface if we are careful dividing it up. Cubes, footballs, dodecahedra, and
pyramids are all valid ways of ‘subdividing’ the sphere into vertices, edges,
and faces. In each of these cases we obtained an Euler number
of 2. However here are other ways we might subdivide the sphere that
seemingly produce differing Euler numbers (Figure 15).

15. Valid and invalid subdivisions of a sphere (a) V = 0, E = 0, F = 1, (b) V = 0, E = 1, F = 2, (c)

V = 1, E = 0, F = 1, (d) V = 1, E = 1, F = 2, (e) V = 1, E = 2, F = 3, (f ) V = 2, E = 0, F = 1.
In order the values of for the six spheres in Figure 15 are 1, 1,
2, 2, 2, 3 and we know the correct Euler number equals 2. So any old
subdivision will not lead to a correct calculation of the sphere’s Euler
number. For a collection of vertices, edges, and faces to make a permissible
subdivision the following must be true:

• an edge must start and finish in a vertex;

• when two edges meet, they must meet in a vertex;
• faces must be (distorted) polygons.

Looking at these so-called six subdivisions, only two of these are in fact
permissible, 15(c) and 15(d). In 15(a), 15(e), 15(f), there is a face that is not
a distorted polygon; neither the whole sphere (15(a)) nor the punctured
cummerbund (15(e)) nor the twice-punctured sphere (15(f)) are
topologically the same as a polygon and so not permissible faces. In 15(b)
and 15(e), the edges do not begin and end in a vertex. 15(e) is deliberately
given to show that the correct Euler number can be incorrectly calculated.

Looking back at the torus in Figure 13(a) and the surface in Figure 14, we
calculated their Euler numbers using subdivisions consistent with the above
three rules. Therefore, we correctly calculated their Euler numbers as 0 and
–1 respectively.

Connected sums
Given two closed surfaces S1 and S2, then we can create their connected
sum S1#S2. This is a way to glue surfaces together and a useful means of
making new surfaces from the few we have so far met. Say S1 and S2 each
has a subdivision that includes a ‘triangular’ face bounded by three edges.
(The inverted commas here hint that, this being topology, the faces may not
be that recognizably triangular in terms of having straight edges.) The
connected sum S1#S2 is then created by removing these two triangular
faces, so making two holes in the surfaces, and then gluing the two surfaces
together along the boundaries of the holes, pairing up the three vertices and
three edges with those on the boundary of the second removed face as, for
example, in Figure 16.

16. Connected sums with tori (a) One torus with a “triangle” missing, (b) A torus with two
holes as a connected sum.

Helpfully there is a formula for the Euler number of S1#S2. In making the
connected sum, we remove two triangular faces, the six different vertices on
these triangles are glued to make three vertices on the connected sum, and
likewise six edges are glued to make three. So the total number of faces has
gone down by 2 and the total numbers of edges and vertices have each gone
down by 3. As V and F are added in the formula for the Euler number, and
E is subtracted, overall we have

Or, if you prefer a more careful algebraic proof, say the original subdivision
of S1 has V1 vertices, E1 edges, and F1 faces and define V2, E2, F2 similarly
for S2. The number of vertices V#, edges E#, and faces F# on the connected
sum is given by

Finally
Thinking in terms of connected sums helps us work out the Euler numbers
of some more complicated surfaces. We know that a torus has an Euler
number of 0. The connected sum is a torus with two holes (Figure
16(b)) and we see

and similarly the torus with three holes, , has Euler number

In fact, we can see that every time we make a connected sum with the
surface gains one more hole and the Euler number reduces by 2. So the
torus with g holes—which can be considered as , the connected sum of
g copies of the torus —has Euler number

The number g of holes in the surface is called the genus of the surface.

One-sided surfaces
At this point, we still can’t identify the peculiar surface from Figure 14
which has an Euler number of –1. So far we’ve only constructed surfaces
with even Euler numbers and –1 is odd. In fact, with the tori , we’ve
only met half the story and half of the closed surfaces. Recall how in Figure
12 we made a cylinder by gluing two edges of a square. We could, instead,
have glued those two sides using reversed arrows (Figure 17(a)),
introducing a single twist. So the points near v2 on e1 are glued to the points
near v4 on e3 and those near v1 on e1 are glued to the points near v3 on e3.
This would have created a Möbius strip, named after August Möbius who
discovered it in 1858.

17. The Möbius strip (a) A square with identified edges, (b) Runners on a Möbius strip.

The Möbius strip is unusual in only having one side—this is apparent in

Figure 17(b) as the runners cover the entirety of the strip rather than just
one side of it as they would if running around just the outside (or inside) of
a cylinder. Or you can imagine painting the outside of a cylinder black and
the inside white, but should you begin painting a Möbius strip one colour
you would find yourself covering the entire strip in that colour. The Möbius
strip is an example of a non-orientable surface. Like the cylinder it is a
surface with boundary, but note its boundary is a single circle rather than
two separate ones as with the cylinder.

In Figures 18(a)–(d) we see an oriented loop—here a circle—moving

around a Möbius strip. By an oriented loop I mean a loop with a given
sense of direction, here initially (18(a)) appearing as clockwise to the
reader. But as this loop moves around the strip (or equivalently moves left
in the square) we see that when the circle returns to its original position
(18(d)) that sense has now reversed and appears anti-clockwise. If you are
having a little trouble visualizing what’s happening to the loop, note in
18(b) and 18(c) how the points labelled P are glued together and likewise
the Qs. In 18(b) most of the loop (on the left) looks to be clockwise running
from P to Q, but as the loop appears on the right and continues from Q to P
that sense is beginning to appear as anti-clockwise.
18. Moving an oriented loop around a Möbius strip.

Any surface on which it is possible to reverse the sense of an oriented loop

is called non-orientable. If it is impossible to reverse a loop’s sense, then
the surface is called orientable. Any surface that contains a Möbius strip is
non-orientable as we could just send an oriented loop once around that strip
to reverse its sense. A surface with an inside and an outside is orientable. To
appreciate this, imagine walking around the outside of such a surface.
Looking down to your feet on the surface you could draw a circle in a
clockwise manner. As you wander around the outside of the surface you can
consistently take your notion of clockwise across the whole surface. This
means, in particular, that the tori , which we met earlier and which each
have an inside and outside, are all examples of orientable surfaces.

Returning to Figure 17(a), a partly glued square making a Möbius strip,

there remain two unglued edges e2 and e4. We could glue these together as
in Figure 19(a), but what surface would we make? Certainly a non-
orientable one as it contains a Möbius strip (the shaded region). If instead
we make this surface by gluing e2 and e4 first, we first create a cylinder
with e1 and e3 as its circular ends. But to complete the surface, rather than
bringing those circular ends together as with a torus, one circular end has to
be glued backwards on to the other circular end—this is because of the
reverse arrows on e1 and e3. Figure 19(b) shows how we might try to do
this; we could take one circular end back into the cylinder and glue it to the
other end from inside, and this way the reverse arrows line up properly. The
surface made is called a Klein bottle, after Felix Klein who first described
it in 1882. Being non-orientable, the Klein bottle does not have an inside
and outside.

19. The Klein bottle and projective plane (a) A Klein bottle, (b) 3D depiction of a Klein Bottle,
(c) A projective plane.

There is a subtle problem with the Klein bottle in Figure 19(b). When we
take the cylinder back into itself, some single points in space actually
represent two distinct points on the Klein bottle. So this image is not a
proper representation or embedding of the Klein bottle in 3D. In fact, it is
impossible to construct a Klein bottle in 3D without such self-intersections
as occur where the cylinder cuts back into itself. The relevant result
demonstrating this impossibility can be viewed as a generalization of the
Jordan curve theorem. That theorem concerned embedding circles in the
plane with a Jordan curve having an inside and an outside. In a like manner
when a closed surface is embedded in 3D, the surface again divides the
remaining space into an inside and an outside and so the closed surface
must be orientable. As the Klein bottle is non-orientable, it cannot be
embedded in 3D.

However, the Klein bottle can be embedded in 4D and this isn’t too hard to
imagine if we treat the fourth dimension as time. The Klein bottle is two-
dimensional (as surfaces are) and so from this 4D viewpoint it is important
to consider the Klein bottle as only existing for an instant, a certain ‘now’;
for it to have a past or future would give it a third dimension. So when
faced with bringing the cylinder back into itself—which would normally
cause self-intersections—we can instead move that bit of cylinder gradually
into the future (the fourth dimension), where the remainder of the Klein
bottle doesn’t exist and then, once the cylinder has passed through the space
its present self occupies, we can gradually bring that bit of the cylinder back
into the present. The self-intersections no longer occur, as the distinct points
of the Klein bottle that became merged in Figure 19(b) instead sit in the
same point of space but crucially at different times.

We can also determine the Euler number of the Klein bottle, again being
careful to note how edges and vertices are glued together. The square is our
only face; e1 and e3 are glued together, as are e2 and e4, making two rather
than four edges; finally v1 is glued to v2 which is glued to v4 which is glued
to v3 and so we have just one vertex, giving ,
the Euler number of the Klein bottle. Unfortunately, 0 is also the Euler
number of the torus, so any hope we might have had that the Euler number
alone is information enough to recognize the shape of a surface was
simplistic. The torus and Klein bottle are different surfaces—the former is
orientable (two-sided), the latter not—and yet they both have the same
Euler number.
Another important non-orientable surface, which can be formed from
gluing a square’s edges together, is the projective plane ℙ. In Figure 19(c)
we assign e2 and e4 reverse arrows (in contrast to 19(a)). The surface
formed is non-orientable, as it again contains a Möbius strip (the shaded
region), and we can calculate the Euler number as before: again and
but this time v1 and v3 are glued together and separately v2 and v4
are glued, so that . Hence ℙ has Euler number
.

The classification theorem

Classification is an important theme in mathematics. A mathematical theory
often begins with definitions and rules about certain mathematical objects
or structures (say functions or curves) and seeks to prove results about them
using those rules. It’s natural to search for examples satisfying those rules,
preferably producing a complete list or classification of such objects.

We are now close to classifying closed surfaces. Explicitly, we are seeking

to give a complete list of all the closed surfaces, so that every closed surface
is homeomorphic to (i.e. topologically the same as) one of the surfaces on
the list, and the list contains no duplicates—each surface on the list can be
shown to be topologically different from all others on the list.

It turns out that the Euler number goes a long way to separating out the
different surfaces, but we have seen that this cannot be the whole story as
the torus and Klein bottle have the same Euler number whilst being
different surfaces—the first is orientable, the second not. The only missing
ingredient in the classification is that notion of orientability.

So the first half of the classification theorem for two-sided surfaces states:

• An orientable closed surface is homeomorphic to precisely one of the tori

where These tori are not topologically the same as one
another as they have different Euler numbers—the Euler number of
is 2–2g.
A similar result holds for one-sided closed surfaces. Just as the torus is a
building block for the orientable surfaces, so can the projective plane ℙ be
used to make the non-orientable surfaces. Recall that the projective plane ℙ
has Euler number 1. So the connected sums ℙ#ℙ and ℙ#ℙ#ℙ have

and more generally k copies of ℙ in a connected sum, a surface denoted

ℙ#k, has Euler number 2–k.

And the second half of the classification theorem for one-sided surfaces
states:

• A non-orientable closed surface is homeomorphic to precisely one of ℙ#k

where These surfaces are not topologically the same as they
have different Euler numbers—the Euler number of ℙ#k is 2–k.

Making a connected sum with ℙ is equivalent to sewing a Möbius strip into

the surface. ℙ itself can be made by introducing a Möbius strip into a
sphere; to do this we might make a tear in the sphere and then, rather than
gluing the tear back together, we could instead assign reverse arrows to the
two sides of the tear, thus introducing a Möbius strip. So the surface ℙ#k can
be thought of as a sphere with k Möbius strips sewed in.

Overall then, the classification theorem says that if we know the Euler
number of a closed surface and whether it is one- or two-sided, then we
know its topological shape. If you were wondering, where the Klein bottle
is on this list, we know its Euler number to be 0 and we know it to be one-
sided. The only surface in the classification matching these facts is the
surface ℙ#ℙ and this is topologically the same as the Klein bottle.
We might create a yet more complicated connected sum such as
which at first glance is not on our list. This surface is
one-sided and its Euler number equals
so topologically it’s the same surface as ℙ#7. And at long last we are able to
identify the surface we formed in Figure 14. That surface had Euler number
–1 and so the surface is ℙ#ℙ#ℙ, this being the only surface on our list with
that Euler number.

Complex numbers
Surfaces are a natural two-dimensional extension of one-dimensional
curves which mathematicians had long been interested in but, historically,
surfaces and their topology became of particular importance because of the
work of 19th-century mathematicians, most notably Bernhard Riemann.

To understand Riemann’s motivation for studying surfaces, we need to take

a brief foray into the world of complex numbers. Complex numbers have, at
first glance, nothing to do with topology, but the need to introduce them
here is a consequence of the deep interconnectedness of mathematics. In the
mid-19th century mathematicians found worthwhile reasons to think about
older mathematics in new topological ways. It might then seem as though
topology was somehow born of practical necessity for addressing these
older problems. However I’d like to suggest a rosier picture of how
mathematicians think: nothing will put a glint in the eyes of a generation of
mathematicians, an itch to be thinking hard about the essence of
mathematics, so much as a sense of there being something profound just
around the corner and a deeper understanding of their subject tantalizingly
beyond their fingertips. And so it was to prove.

These so-called ‘complex’ numbers arose—somewhat uncertainly—from

the work of Italian mathematicians during the Renaissance. For a long time
mathematicians had been interested in the solutions of polynomial
equations. These are equations involving powers and multiples of an
unknown quantity, say x, such as
This is a degree 3 equation, that being the highest power of x. A solution of
an equation is a value of x which makes both sides equal. We can see that
solves this equation because

You might check that is a solution and so is . And that’s all of

them! Three solutions . Other polynomials, though, seem to
have no solutions. For example, the degree 2 equation

has no real numbers as solutions. If you take a positive number x then its
square x2 is also positive (and so cannot equal –1); if you take a negative
number then its square is also positive; finally . So there are no
solutions. If you prefer a more pictorial approach then you might draw the
graphs of and , and the fact that these graphs don’t meet
(Figure 20(a)) is again another way of showing that no number x solves the
equation . Basically the problem is that negative numbers don’t
have real square roots.

20. The real line and complex plane (a) Graphs of y = x2 and y = –1, (b) The real line, (c) The
complex plane.
And there the story might have ended except those Renaissance
mathematicians found good reasons to ‘imagine’ that does have
solutions, denoting a solution as i. This may seem somewhat ludicrous at
first, but around 1530 a method was found for solving degree 3 equations.
One problem was that this method necessitated calculations with square
roots of negative numbers, even when all the equation’s solutions were real
numbers. The worth of the number i became truly apparent with the proof
of the fundamental theorem of algebra in 1799 by Carl Gauss. This
theorem shows all the solutions of any polynomial equation have the form
where a and b are real numbers. For example, the number
solves the equation

as shown by the calculation

Numbers of this form, where a and b are real numbers and ,

are called complex numbers and the fundamental theorem of algebra says
that a polynomial of degree n has (counting possible repeats) n solutions
amongst the complex numbers.

In the same way that real numbers are commonly represented on the real
line (Figure 20(b)) the complex numbers can be represented as a plane, the
complex plane (Figure 20(c)). A complex number such as can then
naturally be identified with the point (1, 2) as shown. The real numbers
occupy the horizontal axis—denoted ‘Re’—and the vertical axis ‘Im’ is
called the imaginary axis.
Complex numbers have a rich theory of their own which, for
mathematicians at least, is reason enough to warrant their study. You may,
though, be surprised to find that quantum theory, the physical theory that
successfully models subatomic physics, is naturally described using the
language of complex numbers and so physicists, chemists, and engineers all
need to be well versed in the use of complex numbers.

Riemann surfaces
The introduction of complex numbers led to a much richer theory
connecting algebra and geometry. In Figure 20(a) we see that the curves
and don’t meet; if they did meet at a point (x,y) in the real
xy-plane then we’d have and no such x exists. But using complex
numbers they do intersect at two points, at and at
. The fact that a degree 2 curve and a degree 1 curve meet
in points in this case is not entirely coincidental. More generally
it is the case that, if properly counted, a degree m curve and a degree n
curve intersect in points. Multiple contacts need to be counted
properly—so a line tangentially meeting the curve would count as a
double contact, so that there are still intersections. The final
finesse, when counting intersections, is to include points at infinity. For
example, two parallel lines—each degree 1 curves—are still deemed to
meet at a point at infinity so that there is intersection as expected.

Using real numbers, the graph of is a one-dimensional curve lying

in the two-dimensional xy-plane. In this case x and y are everyday real
numbers and the curve consists of all points (x, x2) where x is a real number.
We might instead consider the same equation where x and y can now be
complex numbers. Again all the points satisfying are of the form (x,
x2) but this time x can be any complex number. When using real numbers,
the input x represents some point of the x-axis and the corresponding output
x2 can be plotted distance x2 above the point (x, 0) in the xy-plane (Figure
20(a)). However, when it comes to using complex numbers, the input
is itself two-dimensional. The x-‘axis’ is a version of the
complex plane, the y-‘axis’ a second version, and the complex xy-‘plane’ is
in fact four-dimensional. ‘Above’ the point (x, 0) is a point (x, x2), and
together the points (x, x2) make a two-dimensional surface situated in the
four-dimensional complex xy-space. All the points such as (2, 4) that were
on the original real curve are still present, and make up a cross-section of
the complex surface; present too now are points like (i, –1) and (2+i, 3+4i).
If we separate out these complex numbers into their real and imaginary
dimensions, then we might instead represent these points as

and their four-dimensional nature is a little clearer.

The curve sits in the real xy-plane as a curved version of the x-axis
(Figure 20(a)); the curve and axis are topologically the same with a
homeomorphism between the two just pushing each point (x, 0) up to the
point (x, x2). If we include also the curve’s point at infinity bringing
together the curve’s ‘ends’ then the curve topologically becomes a circle.
When using complex numbers, the curve sits in complex xy-space
as a curved version of the x-axis which, remember, is itself a two-
dimensional complex plane. When we include the point at infinity this
brings together this curved plane as a sphere. (This is the reverse process of
puncturing a sphere to get the plane that we met earlier in the proof of
Euler’s formula.)

So, the complex version of , if we include its point at infinity, is

topologically a sphere, a surface; this is called the Riemann surface of
. We might similarly consider the Riemann surfaces of higher degree
equations.

In Figure 21(a) we have the real cross-section described by the degree 3

equation . On the left is a loop, and when we add
the point at infinity to the curve on the right then topologically this real
cross-section becomes two loops; so it might not be surprising that the
whole complex version, the Riemann surface, in this case is a torus with
Figure 21(a) just being a cross-section of that torus. In Figure 21(b) we
have a real cross-section with a singular point where the curve crosses
itself. Topologically the complex version of this curve is a pinched torus as
in Figure 21(c).

21. Visualizing Riemann surfaces (a) Graph of y2 = x(x – 1)(x – 2), (b) Graph of y2 = x(x – 1)2,
(c) A pinched torus.

Provided there are no singular points, then a degree d equation defines a

Riemann surface which is topologically a torus with g holes. There is a
profound but easily described connection between the degree of a curve’s
equation d and the genus g of its Riemann surface. This is given by the
degree-genus formula which states that

where g is the genus of the Riemann surface and d is the degree of the
curve’s equation. Remembering the examples we have met, note that
for gives , a sphere, and for
gives , a torus. For curves with singular points, the formula can be
generalized including a correction term for each singularity, as shown by
Max Noether in 1884.

This is a first glimpse at some of the deep connections between topology,

algebra, geometry, and calculus that would be uncovered in mathematics.
Quoting the French mathematician Jean Dieudonné, ‘In the history of
mathematics the twentieth century will remain as the century of topology.’
The century would see a fledgling subject, intuitively but informally
understood, go on to become one of the central pillars of mathematics.

It is worth noting that those early topologists—Möbius, Klein, Riemann—

did not in their time have available the rigorous definitions necessary to
prove their results to modern standards. In 1861 Möbius gave an early
sketch proof of the classification theorem for orientable surfaces, and
Walther Von Dyck gave a sketch proof for all closed surfaces in 1888. But
without having any formal definition of what a surface is, these proofs can
at best be considered incomplete. This is not to relegate such proofs to the
dustbin, nor to consider them simply wrong, as such proofs often contain
most or all of the crucial ideas of a proof. Somewhat differently expressed
rigorous versions of the classification theorem would be proved by Max
Dehn and Poul Heegaard in 1907 and by Roy Brahana in 1921.

Curves and surfaces are one- and two-dimensional examples of manifolds,

spaces that look ‘up close’ like the real line, the plane, or some higher
dimensional equivalent. It wasn’t until 1936 that Hassler Whitney gave the
modern definition of a manifold, and proved an important theorem showing
when manifolds can be embedding in space (recall, for example, how the
Klein bottle cannot be made in 3D space without self-intersections but can
be made in 4D). An important aspect of modern geometry concerns the
different types of mathematical structure—continuous, smooth, complex,
metric—that can be put on these manifolds and I will say a little more on
this in Chapter 5 when we discuss differential topology.
Chapter 3
Thinking continuously

Given just one sentence for the task, many topologists might choose to
describe their subject as the study of continuity. The word ‘continuous’
appeared a few times in Chapter 1 but it was left to the reader’s intuition as
to quite what the word entailed. In many ways this reflects how
mathematicians used to regard continuity—historically it was just
considered evident what was meant by ‘continuous’ and as many (but not
all) of the results about continuous functions that are ‘obvious’ also happen
to be true, relatively little effort was spent providing further clarity. A
rigorous definition of continuity did not appear until the 19th century.

In your everyday routine there are continuous and discontinuous functions

around you. For example, if you drive to work, the distance you have
travelled after a certain time will be a continuous function of time—for this
not to be the case would mean that at one moment your car was in a certain
place only for it to immediately afterwards be at another place some
distance away. Your speed on the journey will similarly be continuous.
However, the acceleration need not be; if you were sat at rest (say at traffic
lights) the acceleration would be zero but then would jump to a certain
value once your foot was on the accelerator. The graphs in Figure 22 give a
plausible (if simplistic) model for someone’s drive to work.
22. Distance, speed, and acceleration on a journey (a) Distance, (b) Speed, (c) Acceleration.

From Figure 22(b) we can see that the car stops at t3—where the speed s(t)
becomes zero—and after t4 increases to the speed limit. The distance
travelled d(t) in Figure 22(a) is a continuous function of time t. Historically
this would have been understood as meaning its graph could be drawn
without taking pen from paper, but we will seek to provide a fuller
understanding. But the acceleration function a(t) is not continuous because
of the jumps in the graph in Figure 22(c). The times t1, t2, … t6 of
discontinuity in the acceleration relate to the driver’s foot coming off the
accelerator, being put on the brake, coming off the brake, and then the
pattern repeats again.

My aim in this chapter is to provide a more rigorous sense of just what

continuity entails for real-valued functions of a real variable. This means
we will focus on functions having a single numerical input and a single
numerical output.

Functions
The idea of a function is a central one to mathematics, though this has only
been true since around the 17th century. Once Descartes and Fermat
independently introduced the idea of Cartesian coordinates x and y to
describe position in a plane, a curve could just as easily be described by an
equation as by its geometry. For example, the curve is a parabola, a
curve the ancient Greeks would have investigated solely using geometry. A
sketch of this curve is given in Figure 23(a)—the curve’s equation gives a
rule for plotting, above each point (x, 0) of the x-axis, a point (x, x2). Note
how certain algebraic properties of the function are represented in the shape
and position of the curve—as x2 ⩾ 0 for all x, the curve lies entirely on or
above the x-axis; as , the curve is symmetric about the y-axis.
23. Examples of graphs (a) Graph of , (b) Graph of y =.

For some functions, we might naturally have to limit the allowed inputs—or
we might choose to do so anyway. For example, if then we at
least need to ensure that x is non-zero as division by zero is meaningless;
for the function , then we cannot permit x to be negative, as no
real number has a negative square (Figure 23(b)).

More generally, a function comes with a set of inputs, known as the

domain, and there is likewise the codomain, a set containing the outputs. It
is an important, if subtle, point to appreciate that a function is this whole
package: the domain, the codomain, and the rule assigning values.

Some first thoughts about continuity

Let’s first try to understand what it means for a function, with real inputs
and outputs, to be continuous. Currently we sort of intuitively know
continuity when we see it. Certainly, looking at two functions in Figures
24(a) and 24(b), it seems reasonable to say f(x) is continuous and g(x) is not
continuous, and further that g(x) is discontinuous only at . (The full
disc on the graph shows where the function takes its value, so that
.) But what does intuition say about Figure 24(c)? Is h(x) continuous or not?
It seems that, if h(x) is discontinuous, the only point of discontinuity is
, but the function oscillates so wildly there, we may now be thinking
that our intuition didn’t have all the answers.
24. Continuous and discontinuous functions (a) y = f (x) = sin x, (b) , (c)
.
Back to Figure 24(b), what is it about the function’s behaviour at that
makes us think g(x) is discontinuous? For input x a little more than 1, then
g(x) has much the same value as g(1); however for input x a little less than 1
then g(x) is noticeably different from g(1). It is this jump in output at 1 that
is crucial to g(x) being discontinuous at .

At first, we might be tempted to think this is because g(1) is different from

the value of g(x) achieved immediately before we get to x equalling 1. But
there are all sorts of problems with this thinking. First, there is no real
number x that is ‘immediately before’ 1. Given a number like 0.999, close
to 1, then we can always improve on that and see 0.9999 is a little closer. Or
we might suggest using 0.999 … (where the ellipsis means that there are
infinitely many recurring 9s) but this is just another decimal expansion for
1. More rigorously, for any input x < 1 then (1+x)/2 is less than 1 but closer
to 1 than x is. Instead we might be tempted to talk about an input that is
infinitesimally close to 1 but then—whatever we mean by this—we are no
longer talking about the real numbers and have just replaced resolving one
definition with resolving a different one.

We need another approach that can be comfortably expressed entirely in

terms of real numbers. This problem was independently resolved in the 19th
century by Bernard Bolzano and Karl Weierstrass. We feel that g(x) is
discontinuous at because g(x) is noticeably different from g(1) for
some inputs x nearby to 1.

There is still quite a bit of subtlety needed to fully capture what this means.
In our example, g(x) has a jump of 1 from output values near 2 (just before
) to output values near 1 (just after ). The size of that jump was
unimportant, the presence of any jump at all was sufficient. And the notion
of ‘nearby inputs’ should not be interpreted as several inputs that are in
some sense close to 1; rather we mean there are inputs x arbitrarily close to
1 such that g(x) is noticeably different from g(1). Necessarily this means
that we are talking about infinitely many such inputs x, not just several x.
By way of example, it is enough to note that:
This rigorously shows that g(x) is discontinuous at . The sequence of
inputs 0.9, 0.99, 0.999, 0.9999, … gets arbitrarily close to 1. What this
means is: however demanding ‘nearby to 1’ is required to be, there are
inputs from this sequence that are at least that close.

Whilst we still haven’t quite defined just what we mean by discontinuous,

we have made some progress with regard to the function h(x) (Figure
24(c)). This function does not appear to have any noticeable ‘jump’ in
outputs, but it does seem to meet the definition

h(x) is noticeably different from h(0) for some inputs x arbitrarily near
to 0.

Near the function h(x) is varying crazily. From the graph we can see
that there are inputs x, arbitrarily close to 0, where whilst we have
. It now seems clear by our emerging sense of continuity that h(x)
is discontinuous at .

An example in detail
We still need to be careful turning these nascent thoughts into a rigorous
definition. We’ll consider in detail the function which is
continuous for all inputs x. If f(x) is continuous at an input then—
based on our previous thoughts—we need that
 f(x) is not noticeably different from f(a) for all inputs x suitably near to
a.

Take a moment to appreciate why we need all inputs x suitably near to a to

produce not noticeably different outputs f(x) to f(a). If some—but only one
—nearby input x to a resulted in noticeably different outputs f(x) and f(a),
then we could just tighten our notion of ‘suitably near’ to exclude the
problem input x. In fact, if we can never get ‘suitably near’ with our inputs,
then this means that there were arbitrarily close problematic inputs x to a
where f(x) was noticeably different from f(a)—so, f(x) would be
discontinuous at .

To begin, what does it mean for to be continuous at ? Is it

true that

x2 is not noticeably different from for all inputs x suitably near

to 0?

We try out some values in Table 3.

Table 3. Sample input and output values for

It seems—admittedly only on the basis of five choices of x—that x2 is

closer to 0 than x is to 0, and some quick algebra checks that small numbers
generally square to smaller numbers (in magnitude). We cannot find inputs
x close to 0 where the outputs x2 and 0 are noticeably different.
Now we can hang some rigorous mathematics on these initial thoughts:
whatever potential ‘noticeable difference’ in the outputs x2 and we
consider, represented by a positive number e, then there need to be ‘suitably
close’ inputs x to 0, represented by a positive number d, such that

if inputs x and 0 differ by less than d then outputs x2 and 0 differ by less
than e.

As the outputs here are closer to one another than the inputs are—that is, as
x2 is closer to 0 than x is to 0—then we can just choose d to equal e. So if
inputs differ by e or less so do the outputs. We have then shown that
is continuous at .

What about continuity at a different input, say ? We can create a

similar table to Table 3 (see Table 4).

Table 4. More sample input and output values for

The function is growing much more rapidly at than

it is at . A change of around 0.1 in the inputs leads to a change in the
outputs of around 200; a change of 0.01 in the inputs still leads to a
difference of around 20 in the outputs. This may lead you to think that the
outputs are ‘noticeably different’ here, but a more careful check of other
inputs would show that these large differences have been incrementally
achieved. All this is a consequence of the function changing more rapidly
near , and what needs tightening is our notion of ‘suitably near’.
As the function is growing more rapidly, small changes in the input will
lead to relatively large changes, but still in a continuous fashion. If we
consider the input , a little larger than the input 1000, then
the difference in the outputs equals

as d2 < d when d < 1. So a shift in inputs by d results in a shift of outputs

roughly 2000 times larger. (Note similar behaviour in Table 4.) This is, in
itself, not a problem but it does mean that if we want the outputs to differ by
no more than e then we should only allow the inputs to differ by no more
that e/2001. This still shows the continuity of at , we
just needed a tighter sense of ‘suitably near’ with the inputs as the function
was growing so fast. For continuity at yet larger inputs that notion would
have to become yet more stringent, but we would always be able to find
some small wiggle room about an input for which the outputs don’t differ
beyond the desired amount e.

A rigorous definition
Putting all this thinking together gives us a rigorous definition of continuity.
I’d suggest reading the definition and seeking to understand how this means
that the function in Figure 25(a) is continuous and the one in Figure 25(b)
isn’t, but if you find the generality of the definition and the technical level
of the language difficult then move on to the next section on the properties
of continuous functions. And be reassured, as it took generations of
mathematicians to finally get this definition right, and current and past
generations of mathematics undergraduates still wrestle with proofs
involving this definition in their analysis courses.
25. The rigorous definition of continuous and discontinuous (a) A continuous function, (b) A
discontinuous function.

Formally, then, a function with real inputs x and real outputs f(x) is
continuous at an input if:

for any positive e there is some positive d

such that the difference between the outputs f(x) and f(a) is less than e

when the difference between the inputs x and a is less than d.

In Figure 25(a), we are focusing on demonstrating the continuity of f(x) at

input . A particular choice of e > 0 has been made and our task now is
to make sure that the outputs don’t differ from f(a) by more than this e. So
the outputs have to remain below f(a) + e and above f(a) – e (as shown on
the y-axis). And this has to happen for inputs x in some range a – d < x <
a+d. We can see from Figure 25(a) that some such interval has been found,
as shown on the x-axis—the range of outputs on this interval are bounded
by the dashed lines and these fall within the permitted range for the outputs.
To show continuity of f(x) at an input we’d have to show that this can
be done for all e > 0, however small; to show continuity of the function f(x)
we’d have to do this for all inputs x.

There are several important points to note here:

• we require that the outputs can be constrained in a certain way if the
inputs are appropriately constrained;
• we need to be able to do this for all constraints e in the outputs; for each
choice of e we will need a choice of d that meets the requirement;
• for a smaller choice of e then d will usually need to be smaller as well;
• given a positive e, then any positive d that meets the requirement is fine
—we’re not looking for a largest such d, say;
• the faster the function f(x) is changing at a, the smaller d will need to be
relative to e.

A function is then said to be continuous if it is continuous at all its inputs.

And for a function f(x) to be discontinuous at an input means:

there is some positive e such that for any positive d

the difference between the outputs f(x) and f(a) is greater than e
for some input x where the difference between x and a is less than d.

Note how this captures there being inputs x arbitrarily close to a (as they
can be found within any distance d of a) for which the inputs f(x) and f(a)
are ‘noticeably different’ (here meaning differing by more than e). In Figure
25(b) if we choose e to be smaller than the jump in the output that occurs at
then outputs f(a + d) will be greater than f(a) + e and so outside the
permitted range—no matter how small we make d.

Properties of continuous functions

There would be limited reason to be interested in continuous functions if
there wasn’t some payback in what can be guaranteed about continuous
functions, when compared with what can be said of functions in general.
The intermediate value theorem, as you might guess from the name,
states: if we have a continuous function f(x) that is negative at some input
and positive at some later input , there is some input
between a and b—possibly more than one—where . Zero here is
the eponymous ‘intermediate value’ between the negative starting output
f(a) and the final positive output f(b).
If f(x) is continuous, then this seems like something that just has to be true:
if we were to draw the graph of between the point (a, f(a))
below the x-axis and (b, f(b)) above the x-axis, without taking pen off paper,
we surely must cross the x-axis at least once. But, now that we have a
formal definition of continuity, is it particularly clear how we would go
about proving this result? It’s not hard to come up with a counter-example
to the result when the function is discontinuous. Such a function (with
, , say) is

which is sketched in Figure 26(b). Note that its only outputs are –1 and 1,
so that there are definitely no solutions to . But to prove the
intermediate value theorem, we would need to show that a function f(x)
satisfying the theorem’s requirements takes the value zero somewhere—
importantly we would know very little specifically about f(x) save that it is
continuous, begins negatively, and finishes positively, so any approach to a
proof would have to be similarly general. Visualizing drawing the graph
from beneath to above the x-axis, it seems as though there would have to be
a first time that we cross the x-axis and this is indeed the case. That input
value c where we first cross would be

26. The intermediate value theorem (a) Intermediate value theorem, (b) A discontinuous
function.
But writing down this definition of c doesn’t itself constitute a proof and
carefully proving the intermediate value theorem is well beyond the aims of
this book. The above line is the proof’s starting point, we have a candidate
input c where we think the function is zero, but it remains to carefully show
that f(c) = 0. The intermediate value theorem was first proved by
Bolzano in 1817.

Another important theorem is the boundedness theorem. A function f(x) is

said to be bounded if there are bounds M and N such that M ⩽ f(x) ⩽ N for
all x. Continuous functions—like —can be unbounded (here no
such N exists) when we consider all possible inputs x, but when we restrict
the inputs to a domain such as the interval a ⩽ x ⩽ b then the outputs f(x)
will be bounded. More than that, there will be a maximum output and a
minimum output which are achieved at some inputs. For example, the
function on the interval –1 ⩽ x ⩽ 2 has a maximum output of 4
achieved at and a minimum output of 0 achieved at . On the
other hand a function like

is not bounded on the interval 0 ⩽ x ⩽ 2. As x becomes small, but remains

non-zero, then becomes arbitrarily large. The crucial point
here is that g(x) is not continuous, specifically at , and so the
boundedness theorem does not apply in this case. The boundedness theorem
was first proved by Weierstrass during the 1840s, though the result did not
become widely known until he began lecturing in Berlin in 1859.

The continuous functions form an important part of mathematics because of

powerful results, like the intermediate value theorem and boundedness
theorem, guaranteeing certain properties for continuous functions. There are
also theorems that guarantee the continuity of functions that can be
constructed from other functions: if f(x) and g(x) are continuous functions
then so are
Consequently, the continuous functions include many of the functions that
mathematicians and scientists routinely meet and work with. For example,
any function with a defined gradient will also be continuous. However, be
aware that there are still some nasty, pathological examples amongst the
continuous functions—for example, the blancmange function (Figure 27(d))
is a continuous function that has a defined gradient at no input at all.

27. The blancmange function (a) y = f1(x), (b) y = f2(x), (c) y = f3(x), (d) Blancmange function.

The blancmange function can be defined by adding together an infinite list

of functions f1(x), f2(x), f3(x), … The first three functions appear in Figures
27(a), 27(b), 27(c). Note that f1(x) has a defined gradient/slope at each point
except its two peaks and the trough in the middle. There are more inputs—
but still finitely many—where the second and third functions don’t have a
defined gradient. Rather astonishing though if we add the whole list of
functions f1(x), f2(x), f3(x), … then we arrive at the blancmange-shaped
function drawn in Figure 27(d) and this hasn’t a defined gradient at any
point of its graph.

To conclude, recall the earlier comment that a function is the whole package
of domain, codomain, and the assignment rule. We cannot simply say
whether x2 is a bounded function. The assignment x2 is part of an
unbounded function when the domain and codomain are both the real line,
but when we restrict the domain to the interval a ⩽ x ⩽ b then the
assignment x2 yields a bounded function. This pre-empts somewhat a more
detailed discussion for Chapter 4: what is it about the domain a ⩽ x ⩽ b
that means continuous functions are bounded on that domain or satisfy the
intermediate value theorem? The answer to the first question is that the
interval is compact and to the second question is that the interval is
connected. We will see in Chapter 4 what these terms mean more generally.
Chapter 4
The plane and other spaces

More on functions
In Chapter 3 we discussed the continuity of functions that take one
numerical input and produce one numerical output. Most functions are not
of such a simple form. As you read this, the density of matter in the room
around you is a function of three spatial coordinates (needed to describe a
point of the room you’re in) and one coordinate describing time. If you are
reading the paperback version, that density function takes a roughly
constant value (the density of paper) at points in the book you’re reading
but that value changes (discontinuously) at the book’s edges and takes on a
new value (the density of air) at points outside the book. If you’re reading
this outside, the wind’s velocity is an output with three components
measuring to what extent the wind is blowing ahead/behind, up/down,
left/right; each of these three components is again a function of one
temporal and three spatial inputs. We’d expect wind velocity to be a
continuous function, even if it may sometimes change quite quickly. To
have you thinking a little harder, is it reasonable to say that the distance a
car has travelled is a continuous function of its speed? This really is a subtle
question as neither the input nor output are numbers, but rather functions of
time, with input the speed function s(t) and output the distance function d(t)
(Figure 22). If you have some sense of what the question is asking—will
my journey to work tomorrow be much the same as my journey today if I
keep to much the same speed during my journey?—then your intuition is
doing well. Even then, there are important details to be filled in, namely
describing in each case just what ‘much the same’ means.
Thoughts on distance
Consider the important notions needed to define continuity in Chapter 3.
Loosely put, continuity requires that we can constrain the difference in
outputs by suitably constraining the difference in inputs. We will need a
more general notion of the difference between other types of input and
output: for example, what is the ‘difference’ between two points of the
plane?

Given two real numbers, x and y, we denote the difference between them as
|x–y|, and this is just another expression for the distance between x and y as
points on the real line (Figure 20(b)). Likewise, given two points (x1, y1)
and (x2, y2) in the plane, we might take the ‘difference’ between them to
mean the straight-line distance between them (Figure 28(a)). Pythagoras’
Theorem tells us this distance equals
28. Visualizing different metrics (a) Straight line distance, (b) Taxicab distance, (c) Distance
between functions.

so that the points (1,1) and (3,2) in Figure 28(a) are a distance
apart. But in some circumstances, you might decide that
the straight-line distance is not the best way to describe distance
realistically. For example, a taxi driver in Manhattan would be constrained
by New York’s grid of streets and so would need to take a journey as in
Figure 28(b), travelling along perpendicular streets and avenues. That
distance is given by the formula
For the taxi driver, the points (1,1) and (3,2) are a distance apart,
as the taxi may not go along the straight path through Manhattan’s
skyscrapers.

A more complicated example appears in Figure 28(c): what might we mean

by the distance between two functions? Again, we might choose different
ways to measure that distance; looking at the two functions f(x) and g(x)
graphed we might define the distance between them to be

the maximum distance between f(x) and g(x) as we range over all inputs
x,

as highlighted by the two dotted lines. But another reasonable notion of the
distance between the functions might be

the total area between the graphs of f (x) and g(x).

Again, what makes the better definition of distance between two functions
may depend on what the functions signify.

Whilst these examples might seem reasonable definitions for distance, we

are going to need a more detailed, firmer sense of the properties of distance
before any mathematical theory can be produced. What do we expect of
distance? In answering this, we find ourselves defining what it is to be a
metric.

Sets and metric spaces

Metric spaces were first introduced by Maurice Fréchet in 1906, though his
work was little appreciated at the time. A metric space is a set M together
with a distance function d called the metric. This function d has two points
x and y from M as its inputs and its outputs a number d(x, y) representing
the distance from x to y. Further the metric d must satisfy the following
properties for all x, y, and z in M
1. ;
2. and if then ;
3.
4. .

The first property says that distances can’t be negative and the second
implies that the distance between two distinct points is positive. The third
says that the distance going from one point to another is the same as the
distance coming back.

The fourth property is called the triangle inequality. In a triangle, the

combined length of two sides is always greater than the length of the third
side. Similarly property 4 says that the distance d(x,z) between two points x
and z is never more than the distance from x to z via a third point y, namely
. These combined distances might be the same if y is in
some sense ‘on the way’ from x to z but typically the diversion to y will
make for a journey of greater combined distance.

Figure 28’s examples are all metrics—straight-line distance and taxicab

distance are metrics on the plane; maximum-distance-apart and area-
between both define metrics for continuous functions on an interval
. Other metrics are common in mathematics, an important metric
in coding and information theory being Hamming distance, introduced by
Richard Hamming in 1950. Hamming distance is a metric for binary strings
of 0s and 1s, called codewords, used in coding and is useful when dealing
with errors when transmitting code. Given three codewords of the same
length (here length 8),

should we consider these codewords as being close to one another? Note w1

and w2 differ in only the first and last digits, whilst w1 and w3 differ in four
positions as do w2 and w3. It seems reasonable to say w1 and w2 are closer,
as being less distinct, and the Hamming distance between two codewords is
defined to be the number of positions where the two codewords differ, so
that w1 and w2 are distance 2 apart.

If codewords are communicated across distance, there may be a chance of

transmission error. We might hope that the chance of two errors occurring
while sending a single codeword is small enough to be considered
negligible, but single errors might occur. It may be that the string 11101101
is received, which agrees with none of the codewords and is frustratingly
Hamming distance 1 from both w1 and w2. The receiver would not know
whether w1 had been transmitted and the last digit mis-sent or w2 had been
sent with the first digit wrongly received. However, if we make sure to use
codewords that are distance 3 or more apart, and no more than a single error
takes place per codeword, then the receiver can correct the error by
replacing the mis-sent code with the nearest codeword.

Continuity between metric spaces

Continuity for a function with a single numerical input and output requires
that we can constrain the difference in outputs to any degree by constraining
the difference in the inputs. And recall that the difference between
two numbers x, y is also the distance between them on the real line. To
generalize the definition of continuity to functions between metric spaces,
we need to replace that previous notion of difference with that of distance,
as given by some appropriate metric.

More specifically, if we have a function f(x) taking inputs in a metric space

M with outputs in a second metric space N, then f(x) is continuous at input
if

for any positive e there is some positive d

such that the distance in N between the outputs f(x) and f(a) is less than
e
whenever the distance in M between the inputs x and a is less than d.
Note here that the input x need no longer be a single numerical input: x
could be a point in the plane, a subset of the plane, or a function itself,
amongst other possibilities, and likewise there is no restriction on the output
f(x) being a single number.

As we have seen there are different choices of metrics on the same set—for
example straight-line or taxicab on the plane—and two different metrics
might have quite different views of whether two points are close to one
another or not. So it is quite possible that the same function between two
sets might be continuous when using certain metrics, and discontinuous for
a different choice of metrics.

Before generalizing overly, let’s aim to understand what it means for a

function f(x, y) to be continuous when there is an input (x, y) from the plane
and the output f(x, y) is a single real number. We’ll use straight-line distance
in the plane (Figure 28(a)) for the inputs and the usual notion of distance (or
difference) for the outputs.

Here are three such functions.

with their graphs sketched in Figures 29(a)–(c).

29. Graphs of functions of two variables (a) Graph of , (b) Graph of
, (c) Graph of , (d) Eects of G and H, (e) Close and distant
functions.

Based on the graphs, you may think that f(x, y) and h(x, y) are continuous,
whilst g(x, y) is discontinuous—all of which is true. You may further
suspect that g(x, y) is discontinuous at the points (0, y) on the line ,
which is again correct. The jump in output g(x, y) across the line , as
we move between the two rules that define the function, is precisely why
g(x, y) is discontinuous there. h(x, y) is also defined by two separate rules,
one for inputs where , one for when . The graph
of h(x, y), when is the bowl-shaped part of Figure 29(c). As
we move towards the boundary of that rule’s application, the circle
, then gets ever closer to equalling 1
which is the rule for h(x, y) outside of the disc. So whilst h(x, y) is defined
by two rules, the rule for nicely hands over to the second rule
for without any jump in the output. This is why h(x, y) is
continuous.
As before there are nice algebraic results guaranteeing that if f (x, y) and
g(x, y) are continuous functions from the plane to the real numbers then so
are the functions

With the general approach of metric spaces, we can consider more

complicated examples, such as the set of continuous functions on the
interval . Here are three examples F, G, H of functions which
take inputs f(x) that themselves are functions.

F outputs real numbers, just evaluating the input function at 0, whereas G

and H output functions: G increases the input function by 1 so that its graph
moves up by 1 and H reflects the graph of a function in the y-axis (Figure
29(d)).

We cannot say whether F, G, H are continuous until we make clear which

metrics we’re using. But whether we use the maximum-distance-apart
metric or the area-between metric, the functions G and H are continuous as
the maximum distance between two graphs is unchanged by a vertical move
of 1, nor is the area between the graphs. The same is equally true when H
reflects the graphs in the y-axis.

Our choice of metric does matter for F, which evaluates an input f (x) at
. Continuity means being able to constrain the outputs by constraining
the inputs: must two functions be close at 0 if the functions f (x) and g(x) are
close? The answer is yes if we’re using the maximum-distance-apart metric;
the difference between the functions at the single input 0 can be no more
than the greatest distance between the functions when considering all
inputs. However, we cannot constrain the difference between the functions
at 0 by constraining the area between the functions (Figure 29(e)). A
function g(x) with a tall, but very thin spike around would produce a
large difference between f (0) and g(0) whilst f (x) and g(x) would be close
in terms of the area-between metric.

And we can now properly ask and answer the question that started this
chapter: is the distance a car has travelled, d(t), a continuous function of its
speed, s(t)? (See Figure 22.) Any journey to work taking time T or less can
be represented by a continuous function d(t) on the interval 0 ⩽ t ⩽ T. Is
the function that takes input s(t) to output d(t) continuous? If two cars have
speeds s1(t) and s2(t) that never differ by more than S then the two cars will
never be more than distance ST apart on their journeys. So using maximum-
distance-apart, we can constrain the distance between two journeys d1(t)
and d2(t) by constraining the distance between their speeds s1(t) and s2(t).
We have just shown d(t) is a continuous function of s(t) when using the
maximum-distance-apart metric. Using a little calculus, a similar argument
can be made for the area-between metric.

Equivalent metrics and continuity

Our main reason for introducing metric spaces was to provide a more
general setting in which to discuss continuity. But sometimes different
metrics on a set lead to precisely the same continuous functions. Two such
metrics are the straight-line distance and taxicab distance on the plane.

To help appreciate this, we introduce the idea of open balls which

generalize discs in the plane when we’re using straight-line distance. The
open ball B(a, r) of radius r about a point a in a metric space is the set of
points in the metric space that are less than distance r from a. When using
straight-line distance in the plane, this is the interior of the circle of radius r
with centre a (Figure 30(a)). The dashed circle in the figure indicates that
the circumference is not included in the open ball.
30. Open balls in the plane (a) Open balls in the plane, (b) Open balls within others.

However, if we were using the taxicab distance in the plane the ball B(a, r)
looks different. The taxicab distance between two points and
equals and so the points within
(taxicab) distance r of a0 are those satisfying

Some algebraic manipulation shows that these points form the interior of a
diamond (Figure 30(a)).
It is possible to fit such a diamond into any disc of the plane, and vice versa
it is possible to fit a disc into any given diamond (Figure 30(b)). This is
what it means for two metrics to be equivalent. More precisely, two
different metrics d1 and d2 are said to be equivalent if any ball B1(a, r)
contains some ball B2(a, s) and any ball B2(a, R) contains some ball B1(a,
S). The subscripts here refer to which metric is being used and the smaller
ball’s radii s and S will usually be different from the radii r and R of the
original balls.

The straight-line distance is always less than or equal to taxicab distance, as

the former is distance measured ‘as the crow flies’. But there are limits to
how much greater taxicab distance can be compared with straight-line
distance. The greatest taxicab distance from the centre of a circle of radius r
to a point of the circle equals . These are the points on the circle that
are north-east, north-west, south-west, and south-east of the centre. This
means that taxicab distance can be no more than times bigger than
straight-line distance, or expressed as an inequality,

for any points a and b, and where the subscripts SL and T signify the metric.
So BT(a, r) is contained in BSL(a, r) which is contained in , for
any point a and radius r (Figure 30(b)).

By contrast, the maximum-distance-apart and area-between metrics for

functions are not equivalent metrics. The maximum distance between two
graphs is a constraint on how great the area between the graphs can be, but
constraining the area between graphs does not constrain the maximum
distance between two graphs. We saw this earlier in Figure 29(e) where two
graphs differing only by a high, very thin spike have a large maximum
distance between them but only a small area between them.

The important point here is that two different but equivalent metrics lead to
the same functions being continuous. Now we can rewrite the definition of
continuity in terms of open balls: a function f(x), with inputs from M and
outputs in N, is continuous at input if

for any positive e there is some positive d such that f sends B(a, d) into
B(f(a), e).

This is purely a notational rewriting of the definition given on p. 69—all the

points within distance d of a, namely those in B(a, d), need to be sent to
points within distance e of f(a), namely sent into B(f(a), e). If a function is
continuous at when using a metric d1 on M then it would also be
continuous when using an equivalent metric d2. This is because the ball
B1(a, d) would contain a ball B2(a, D) using the second metric, and if B1(a,
d) is sent into B(f(a), e), then so is B2(a, D).

Open sets and continuity

An open ball is a basic example of a more general notion, that of an open
set. Recall that an open ball consists of all points strictly less than a given
distance away and so does not include the circumference of the ball.
Similarly, open sets can be thought of as those sets that don’t contain any of
their boundary points. The formal definition is that a set U, contained in a
metric space M, is open if around each point a in M there is an open ball
B(a,r) still contained in U (Figure 31(a)). Open balls are unsurprisingly
examples of open sets. If a point b lies in B(a, r), at a distance s from a,
then the ball B(b, r-s) is contained in B(a, r) (Figure 31(b)). As before the
dotted lines in Figure 31 denote that the boundary points aren’t included in
the sets.
31. Open sets in the plane (a) An open set in the plane, (b) Open balls are open, (c) Union and
intersection.

Two equivalent metrics determine the same collection of open sets. To

appreciate this, note that if a is in an open set U, we can place an open ball
B(a, r) around it inside U. If we have a second equivalent metric, then we
can place another open ball centred at a—using this second metric—
contained in B(a, r) and so inside U as well. This means U is also an open
set when using the second metric.

Equivalent metrics were mentioned earlier because a function that is

continuous when using one metric is continuous when using an equivalent
metric. And we have just seen that equivalent metrics lead to the same
collection of open sets. These two facts are not unconnected and indeed
knowledge of which sets are open is sufficient to determine which functions
are continuous. It is relatively straightforward to show that the following
definition of continuity in terms of open sets is equivalent to the ones given
on p. 69 and p. 75; one advantage of this new definition is that it is much
more generally applicable. Our new definition reads: a function f between
metrics spaces M and N is continuous if

whenever U is an open set in N, then the preimage f –1(U) is an open set

in M.

The preimage f –1(U) consists of all elements of M that f sends into U. For
example, if and U is the interval 0 < x < 2 then the pre-
image f–1(U) is the interval –1 < x < 1 as these are precisely those x that
satisfy the inequality .
That different metrics can lead to the same continuous functions and that
continuity can be rephrased in terms of open sets, making no mention of
metrics, suggest that the open sets, more so than metrics, are crucial to
continuity. However, if we are going to take this approach—starting with
open sets rather than metrics—we need to decide what properties open sets
need to have; so far we have only defined open sets in terms of metrics. The
two main properties of open sets are:

• for any collection of open sets, their union is also open;

• for any finite collection of open sets, their intersection is also open.

Given a collection of sets their union consists of those elements that are in
one or more of those sets; the intersection of those sets consists of those
elements that are in every one of those sets (see Figure 31(c)). The union of
any collection of open sets is open, but in general the intersection of an
infinite collection of open sets need not be open. Consider the interval –r <
x < r where r is a positive number. If we consider all such sets for positive r,
then the only element in each set is 0. So the intersection of these sets is
{0}, the set with sole element 0, an intersection which is not open.

Felix Hausdorff sought to capture and generalize the necessary properties of

open sets in his 1914 magnum opus Grundzüge der Mengenlehre or
Essentials of Set Theory. There Hausdorff introduced many key ideas of set
theory and general topology, extending and synthesizing the work of
Fréchet and others into a coherent whole. It was also here that he coined the
term metric space for the spaces Fréchet had introduced.

Given a set M, then any collection of sets in M which satisfies the

following:

• for any collection of sets in , then their union is also in ;

• for any finite collection sets in , then their intersection is also in ;
• M is in ;
• the empty set—the set with no elements—is in .
is called a topology on M and a set M, with a topology of sets , is known
as a topological space.

Note that M, by itself, is ‘just’ a set, not a metric space, with elements
having no sense of being close to one another; the collection of sets is an
effort to separate out M without necessarily going so far as introducing a
metric. Given a metric space then its open sets form a topology, but not all
topologies arise from metrics.

The following is an example of a topology on the set of whole numbers

which does not arise from a metric. Necessarily the entire set of whole
numbers and the empty set must be in the topology; the only other sets in
this topology are those of the form

where n is a whole number. The intersection of finitely many Ui is Umax(i),

where max(i) denotes the largest of the is, and so in the topology; the union
of any collection of Ui is Umin(i) if the minimum min(i) exists and is
otherwise the set of all whole numbers—in either case the union is in the
topology. So, we have just verified that this collection of sets satisfies the
rules of being a topology. In this space non-empty open sets always overlap,
which isn’t true of metric spaces, so this is one way to be sure this topology
doesn’t arise from a metric.

The smallest topology on M just includes the empty set and the set M.
This is known as the trivial topology—it does not separate out the elements
of M at all and the only continuous functions on M are the constant ones.
The largest topology is the discrete topology in which case every set is in ,
including single points; it introduces extreme separation, placing each
element of the set away from others and all functions on M are continuous.
In practice, important topologies are somewhere between these two
extremes. Many important topologies arise from metrics, but there are
important ones that do not, such as the Zariski topology, important in
algebraic geometry which studies sets defined by polynomial equations.
Working with topological spaces, rather than metric spaces, does more than
just generalize further the ideas of continuity. Many proofs in topology
appear cleaner, with the logic of the proof and the rules of being a
topological space meshing together much more naturally.

Convergence and continuity

Metric spaces are also a natural setting in which to define convergence, a
central idea of calculus and of mathematical analysis. A sequence from a
set is a list of elements in that set; for example, the following are three
sequences of real numbers.

The ellipsis ‘. . .’ at the end of each list denotes that the list goes on forever.
A sequence is commonly denoted as x1, x2, x3, … and the terms in these
sequences can each be described by giving a formula for the nth term xn:

Only the first of these sequences converges; the terms of the sequence are
steadily getting smaller and I hope it’s not surprising that this sequence
converges to 0 which is known as the sequence’s limit. The second
sequence does not converge to any limit; some of its terms though—like the
even terms (second, fourth, sixth, etc.) which are all 1—do converge but
overall the sequence does not. The final sequence does not converge either
and in fact no selection of its terms converges either.

The very word ‘converge’ suggests that the terms of the sequence get closer
and closer to some limit. Rigorously, a sequence xn, in a metric space M,
converges to a limit a in M if any ball B(a, r) contains a tail of the
sequence; this means that from some term onwards, all remaining terms of
the sequence are inside B(a, r).
Using this definition, we can see why the sequence converges to
0. The ball B(0, r) contains all terms xn of the sequence where n > 1/r,
which is a tail of the sequence. In Figure 32(a), a sequence spirals in to its
limit (0, 0). The pictured dashed disc B((0, 0), r) contains x10 and every
term afterwards; we would need to go farther down the sequence to find a
tail contained within a smaller open ball. If a sequence converges to a limit
in a metric space, then the sequence converges to the same limit if we use
any equivalent metric.

32. Visualizing convergence (a) A sequence converging in the plane, (b) Neither open nor
closed set.

Sequences also provide an alternative means of defining continuity. For a

continuous function f between metrics spaces M and N, if a sequence xn of
inputs converges to a limit a in M, then the sequence f(xn) of outputs
converges to f(a). For example, with the sequence which
converges to 0, and the continuous function then f(xn) is the

sequence which converges to .

Similarly for the continuous function , then g(xn) is the
sequence which converges to
. And the converse is also true: if, whenever a
sequence xn converges to a, then f(xn) converges to f(a), the function f(x) is
continuous at .
Having defined convergence, we can now define what it means for a set to
be closed. If C is a set in a metric space M, then it is closed if whenever a
sequence of points in C converges then the limit is also in C. Closed sets
might be thought of as those sets that contain all their boundary points. An
alternative definition of C being a closed set in M is that the complement of
C, that is all the points in M that aren’t in C, is an open set.

So, an open set is one which contains none of its boundary points, and a
closed set is one which contains all its boundary points. Clearly most sets
fall into neither of these categories and will contain some but not all of their
boundary points; it’s important to realize that being open and being closed
are not opposites of one another. In Figure 32(b) is a set which is neither
open nor closed: a disc with its upper circumference included but the lower
half omitted. The boundary point a is contained within the set so the set is
not open—or equally no open ball B(a, r) around a is contained within the
set. And the boundary point b isn’t contained within the set, so the set is not
closed—or equally there is a sequence of points in the set that converges to
b despite b not being in the set. And some sets can be both open and closed,
for example the whole plane is both an open and closed set of the plane—
there are no boundary points, so all and none of them are simultaneously in
the set.

Subspaces
Any set S in a metric space M inherits the structure of a metric space, as two
points of S are also points of M and we may assign them the same distance
as before when they were considered as points of M. For example, if M was
the set of cities in the United States, and S is the set of Californian cities,
then it’s natural to think of Los Angeles and San Francisco as being the
same distance apart whether they’re being considered as American cities or
Californian cities. The set S naturally becomes a metric space, in its own
right, called a subspace of M.

Subtleties arise when we consider the open sets of and continuous functions
on a subspace. It may help to remember the Flatland mindset mentioned in
Chapter 1; we need to imagine life as an inhabitant of the subspace S, as if
Californians somehow cannot see the rest of the USA.

For example, let M be the real line and let S be the union of the intervals 0
⩽ x ⩽ 2 and 3 < x ⩽ 4 (Figure 33(a)). What are the open balls B(1, 1) and
B(2, 1) as open balls in S? By definition, B(1, 1) is the set of points in S
within distance 1 of the point 1; this is the interval 0 < x < 2 which agrees
with what B(1, 1) is in the real line M. However B(2, 1) is the interval 1 < x
⩽ 2 as an open ball of S; this is the set of points in S that are within distance
1 of 2. This is different from the situation with the real line, and is yet more
surprising when we remember open balls are open sets. The point 2 appears
incongruous as points near it to the right are not included but, from the
Flatland mindset of the subspace S, those points are essentially invisible as
they are not in S. So, the interval 1 < x ⩽ 2 is indeed an open set in S.

33. Relating to a disconnected subspace (a) Open balls in a union of two intervals, (b) Graph of
f(x) on S.

Also, perhaps surprisingly, the function

is continuous on the subspace S (Figure 33(b)). Instinctively you notice a

jump between the outputs of –1 and 1 and may think f is not continuous, but
there is no point in S where a discontinuity can be identified.

Note also that the interval 0 ⩽ x ⩽ 2 is open as a set in S. This interval is

the open ball B(1, 2) in S and open balls are open sets; those points in S
which are within distance 2 of the point 1 form the interval 0 ⩽ x ⩽ 2.
Similarly, the interval 3 < x ⩽ 4 is open as a set in S as it is the open ball
B(4, 1) in S. As the complement of an open set is a closed set, then both
intervals are also closed sets in S. This again may seem counter-intuitive as
3 looks to be a boundary point that’s missing from the interval 3 < x ⩽ 4.
But from the Flatland mindset of S the point 3 is effectively invisible and so
not missing at all.

Compactness and connectedness

In Chapter 3 we met two important theorems relating to continuous
functions on an interval a ⩽ x ⩽ b, namely the boundedness theorem and
the intermediate value theorem. In general continuous functions need not be
bounded—for example x2 on the real line is unbounded—nor need they
attain intermediate values—for example the function f(x) in Figure 33(b)
does not take the value 0 despite attaining –1 and 1. There must be some
property or properties of the interval a ⩽ x ⩽ b of inputs that means
continuous functions on that domain are bounded or attain intermediate
values.

One of these properties can be captured in terms of the behaviour of

sequences in such an interval. A sequence like –1, 1, –1, 1, –1, … may not
converge, but some selections of the sequence do—for example the odd
terms (first, third, fifth . . .) are all –1 and so converge to –1 and the even
terms (second, fourth, sixth . . .) similarly converge to 1. But some
sequences like 1, 2, 3, 4, 5, … have no selection that converges; any
selection grows towards infinity. The Bolzano–Weierstrass theorem states
that any sequence in an interval a ⩽ x ⩽ b has a selection which converges
in that interval.

A metric space in which the Bolzano–Weierstrass theorem holds, so that all

sequences in the space have a selection that converges to a limit in the
space, is called compact. The theorem shows the interval a ⩽ x ⩽ b is
compact, the sequence 1, 2, 3, 4, 5 … in the real line shows that the real line
is not compact as no selection converges, and the sequence
shows the interval 0 < x ⩽ 1 is not compact as it has
no selection that converges in that interval; this last sequence’s limit of 0 is
outside the interval and any selection from this sequence also converges to
0.

For the real line, plane, and higher dimensional equivalents, the Heine–
Borel theorem states that the compact sets are precisely the closed and
bounded sets. We have already defined what a closed set is, and a set is
bounded if it is contained in some ball B(0, R). So we can see that the real
line is not compact as it is not bounded and the interval 0 < x ⩽ 1 is not
compact as it is not closed, missing the boundary point of 0.

Compactness was first introduced by Fréchet in 1906, and the compactness

of a ⩽ x ⩽ b is the reason that the boundedness theorem holds. More
generally it is true that real-output continuous functions on a compact
domain are bounded and attain their bounds. It is also true that compactness
is a topological invariant, so that two homeomorphic metric spaces are
either both compact or neither is compact. Compactness is a much more
important notion in topology than these few examples can make clear, with
many theorems of topology concerning compact spaces.

However, the intermediate value theorem on the interval a ⩽ x ⩽ b is not a

consequence of compactness, but rather of connectedness, a notion first
introduced by Hausdorff in 1914. Intuitively, connectedness is a simple
idea: being connected means being ‘in one piece’, but how can we capture
this definition? Surely the real line is connected and the set of whole
numbers disconnected, as ought to be the earlier space S which is the union
of the intervals 0 ⩽ x ⩽ 2 and 3 < x ⩽ 4.

Intuition tells us that S is disconnected and, more, that S is made up of just

two pieces, the two intervals. We noted earlier that the interval 0 ⩽ x ⩽ 2 is
both open and closed as a set in S, as is the second interval. In general, for a
metric space M, the whole set M and the empty set—the set with no points
of M—are both open and closed sets in M. A space M is defined to be
connected if these are the only open and closed sets in M. Or equally a
space is connected if it cannot be split as the union of two non-empty, open
sets with no elements in common.
The first interval 0 ⩽ x ⩽ 2 and second 3 < x ⩽ 4 which make up the set S
are called the connected components of S; these are the largest sets
contained in S that are themselves connected. For the set of whole numbers,
the connected components are the sets of individual numbers such as {1} or
{3}. For a space to have just one connected component is just another way
to say that a space is connected.

The intermediate value theorem holds on a connected space: let M be a

connected metric space, f be a real-valued continuous function on M and a,b
be points in M such that f(a) < 0 < f(b). Then there is a point c in M such
that .

This definition of connectedness may seem rather abstract, and there is a

more concrete version of connectedness, path-connectedness, which is
easier to appreciate. Given points a and b in a metric space M, then a path
between a and b is a continuous function p from the interval 0 ⩽ x ⩽ 1 to
M such that and (Figures 34(a), 34(b)). You might
consider p as being a journey from the point a as a starting point when
to the point b as finishing point when . The points p(x) where 0
< x < 1 are the points passed through getting from a to b.

34. Relating to path-connectedness (a) A path-connected set, (b) A convex set, (c) Topologist’s
sine curve.

Path-connected metric spaces are connected, though there are some weird
spaces that are connected without being path-connected (Figure 34(c)). For
many regions, paths between points can be straightforwardly defined and a
space quickly shown to be path-connected and so connected. For example,
in an open disc, the plane, a half-plane, any two points are connected by a
straight line. These are all examples of convex sets, which means that given
any two points in the set then the line segment between them is also
contained in the set (Figure 34(b)) which is not true of the Pacman-like
shape (Figure 34(a)).

The topologist’s sine curve in Figure 34(c) is an example of a space which

is connected but not path-connected. The space is the union of the y-axis
and the curve for x > 0. Loosely put, the space is connected
because the sine curve gets arbitrarily close to the y-axis but it is not path-
connected as the function sin(1/x) has no limit as x becomes small. Any two
points on the sine curve can be connected by a path, as can any two points
on the y-axis, but a point on the y-axis and a point on the sine curve cannot
be connected by a path in the space.

Again, connectedness is a topological invariant—if two metric spaces are

homeomorphic and one is connected then so is the other. Similarly, path-
connectedness is a topological invariant.

Topological invariants
In Chapter 1 we separated out which letters of the alphabet were
topologically the same—homeomorphic—or not. For two equivalent letters,
we described a way of deforming each into the other; for topologically
different ones, we needed to find a feature—a topological invariant, like the
T-junction in the E—that would remain a feature of the letter, even when
deformed. Compactness and connectedness are topological invariants and
so can be used to differentiate between spaces.

In Table 5 are eight spaces, none of which is homeomorphic to another.

Some are compact—closed and bounded—others not, so compactness can
be used somewhat to separate them. Using the Heine–Borel theorem, we
see the top row are compact, each being closed and bounded, and the
bottom row are not. So none of the top row is homeomorphic to one from
the bottom row, but compactness implies nothing about whether spaces on
the same row are topologically different. (A closed disc means a circular
disc including its circumference. A closed half-plane means all the points
on and to one side of a line in the plane; so the bounding line is included in
the closed half-plane.)

Table 5. Eight spaces which are not homeomorphic

All the spaces are connected, but with a little imagination we can still make
use of connectedness to discern topological differences. A point of a
connected space which, when removed, disconnects the space is called a
cut point. We also noted in Chapter 1 that the T-junction in an E is the
special cut point which, if removed, breaks E into three connected
components; removing a different point leaves just two components
remaining.

Looking at the top row of Table 5, neither the circle nor the closed disc has
any cut points, but the figure 8 and interval do have; this means the former
two are not homeomorphic to the latter two. However, we can disconnect
the circle by removing two points and such is not true of the closed disc.
Finally, the middle point of the figure 8 is the only cut point whilst any
point of the interval, except the ends, is a cut point.

In the bottom row, the line and interval have cut points, but not the half-
plane nor plane. But 1 is not a cut point of the interval whilst every point of
the line is a cut point. All that remains to do is show the closed half-plane
and plane aren’t homeomorphic.

For this we need a further topological invariant, discussed in more detail in

Chapter 5. The boundary points of a closed half-plane seem different from
other points. Neither the plane nor half-plane has any cut points, but
removing a boundary point from the half-plane doesn’t make a hole, the
way removing a point not on the boundary does. Informally put, a space
with no holes is called simply connected and this is a topological invariant.
As the removal of a point of the half-plane can leave a simply connected
remainder (without holes), and as this is not true of the plane, then the
closed half-plane and plane are not homeomorphic.

Taking the methods and theorems of continuity for single input, single
output functions and applying them to the more general settings of metric
and topological spaces proves to be a very powerful approach.
Consequently, properties such as compactness and connectedness are used
widely in mathematics and a classical theorem guaranteeing a real number
solution to an equation might in a modern setting show that there is a
continuous function which solves a differential equation. Seemingly these
are very different mathematical problems, but this abstract mode of thinking
helps mathematical ideas be applied in their fullest generality.
Chapter 5
Flavours of topology

From the mid-19th century, topological understanding progressed on

various fronts. The geometric topology of Chapter 2 concerned surfaces and
grew out of the work of Euler, Möbius, Riemann, and others. The general
topology of Chapters 3 and 4 was more analytical and foundational in
nature; Hausdorff was its most significant progenitor and its growth
mirrored other fundamental work being done in set theory. There are yet
more flavours of topology—a topologist might have texts on their
bookshelves entitled algebraic topology, differential topology, symplectic
topology, and other books on the subject with less explicitly topological
titles.

In Chapter 2 we met the Euler number and saw that it, together with
knowing whether a surface is one- or two-sided, identifies the shape of a
closed surface. Further, the Euler number imposes global constraints about
what is possible on a closed surface as we’ll see in the next two sections.

Differential topology
The continuous functions form an important class of functions, but we also
saw in Chapter 3 that continuous functions can still be quite nasty; for
example, the blancmange function does not have a defined gradient at any
point. The smooth functions also form an important class. A function might
be continuous but still fail to be smooth by changing in a jerky fashion;
smooth functions by contrast have a defined gradient everywhere. For
example, the speed function s(t) in Figure 22(b) is continuous but doesn’t
have a defined gradient at the times t1, … t6. Before and after those times,
the gradient is clear but there is no defined gradient of s(t) when the car
accelerates or brakes suddenly. The study of functions that change smoothly
on surfaces lies within the field of differential topology and we will see
that the Euler number of a closed surface impacts on what properties
smooth functions on a closed surface can (or must) have.

An important first result about smooth functions is that the gradient of the
function is zero at any maximum or minimum. Note, in Figure 24(a), how
the gradient of sinx is zero at its extreme values whereas, in Figure 27(a),
the (continuous but not smooth) function f1(x) has no defined gradient at its
maxima. This result is known as Fermat’s theorem and these maxima and
minima are known as critical values and the corresponding inputs as
critical points.

But a function of two inputs can have a greater variety of critical points
than a function with just one input. When we have two inputs and one
output, the notion of gradient is a little less clear. Sketched in Figure 35 are
three surfaces exhibiting different types of critical point.
35. Examples of a minimum, maximum, and saddle point (a) , (b)
, (c) ,

In Figure 35(a) we have a minimum at the bottom of a bowl-like graph,

above the point , . In whatever direction we move from the
bowl’s bottom, we move upwards. Likewise, in Figure 35(b), we have a
hill-like graph with a maximum at the top and however we move away from
that maximum we move downwards. The point x = 0, y = 0 in Figure 35(c)
is neither a maximum nor a minimum. The function is zero at that point and
if we move along the x-axis we get to a point where so
that the function has increased; but if we were to move along the y-axis we
get to a point where so that the function has
decreased. This critical point is an example of a saddle point and hopefully
the name is not surprising given the shape of the graph. As with a saddle on
a horse’s back, the arc of the horse’s back has a lowest point where the rider
sits, but the rider straddles the horse in a manner where the saddle is at the
highest point. There are yet more complicated examples of critical values
with two inputs; the function , for example, has a
graph with a saddle-like critical point that would suit a rider with two legs
and a tail—so this is called a monkey saddle—but in what follows we will
be interested in functions that only have maxima, minima, and saddle points
as critical points.

Consider now the height function z for the torus drawn in Figure 36(a). The
height function has a maximum value at the top of the torus A, and a
minimum value achieved at the bottom of the torus D. In fact, as the torus is
compact, the boundedness theorem guarantees that any continuous function
on it must have at least one maximum and at least one minimum.

36. Critical points of functions on a torus and sphere (a) Height function on a torus, (b) z2 – y2
on the unit sphere.
There are two further critical points: C, at the bottom of the hole and B, at
the hole’s top. If we move through C, passing through the hole, then our
height is at its greatest as we pass through C; but if we slide down one side
of the hole and up the other side, then the lowest point of our journey is at
C. This critical point C is a saddle point and similar paths can be made
through B which include it as the highest or lowest point on the journey, so
B is also a saddle point.

In Figure 36(b), we consider the function on the

sphere with equation . With a little work we can show
that this function’s critical points are

• maxima at (0, 0, 1) and (0, 0, –1), the sphere’s north and south poles;
• minima at (0, 1, 0) and (0, –1, 0), two points lying on the sphere’s
equator;
• saddle points at (1, 0, 0) and (–1, 0, 0), two more points lying on the
sphere’s equator.

At the north pole (0, 0, 1), z is as large as it can be on the sphere, so any
move away means that z2 – y2 decreases and (0, 0, 1) is therefore a
maximum of f. The same argument applies at the south pole. Similarly, at
(0, 1, 0), y is as large as it can be, so any move away means that z2 – y2
increases and so (0, 1, 0) is a minimum, as is (0, –1, 0). Finally, at (1, 0, 0),
x is as large as it can be. As we move from (1, 0, 0) then x will decrease and
either y or z or both might increase, meaning that f may decrease or
increase. This means that (1, 0, 0), and likewise (–1, 0, 0) are saddle points
of f.

At this point, any relationship with topology probably seems unclear. The
height function on the torus had one maximum, one minimum, and two
saddle points; z2 – y2 on the unit sphere had two maxima, two minima, and
two saddle points. However, if we continued considering smooth functions
on the torus, in each case we would find that
(as with the height function where ) and for any smooth
function on the sphere we’d find that

(as with z2 – y2 where ). That itself is perhaps surprising

enough, but when we remember that 0 and 2 are the Euler numbers of the
torus and sphere then we might suspect a deeper relation. Generally, then, it
is true for any smooth function on a closed surface that

In this way the Euler number is a global constraint as to what features

smooth functions may have on a closed surface. We might tweak our
function near any point to create a maximum, a minimum, or a saddle point,
but there will necessarily be consequences for the function elsewhere. The
Euler number limits overall possibilities in the unavoidable way of if-you-
push-down-here-it-pops-up-over-there DIY problems. As an example, we
can deduce that any smooth function on a torus must have at least two
saddle points: a smooth function on a torus must have at least one
maximum and at least one minimum because the torus is compact, and so

In fact, we can see that the sphere (being the only closed surface with an
Euler number of 2 or more) is the only closed surface on which a smooth
function may have no saddle points. An example of such a function is the
height function on a sphere.

The Scottish physicist James Clerk Maxwell was one of the first to
appreciate such a relation amongst critical points in an 1870 paper On Hills
and Dales, but he was working solely with functions on the plane—and so
arrived at the number 1 (the Euler number of the plane or a punctured
sphere). Poincaré would generalize the result to closed surfaces, using
techniques that would now be considered Morse theory, after Marston
Morse who, from 1925 onwards, would prove a series of deep results
connecting the topology of spaces and the analysis of functions on those
spaces.

The hairy ball theorem

Informally put, the hairy ball theorem says that you can’t comb a hairy
ball flat without creating a tuft or cow-lick where the hair refuses to lie flat.
We might brush and sweep the hair on the sphere in various styles but
something about the sphere’s underlying shape means we can’t ever get the
hair to lie flat everywhere. But it’s not hard to imagine how the hair on a
torus could all be swept flat in the same direction (Figure 37(b)). As before,
with smooth functions, we’ll see that it’s the sphere’s topology that makes
this impossible.

37. Vector fields on the sphere and torus (a) A hairy ball with no hair at poles, (b) A hairy-
everywhere torus.

The hair in this theorem is a metaphor for a tangent vector field which
might be easiest thought of as a fluid flowing on a surface; the hairy ball
theorem then states that for any fluid flow on a sphere, there will be one or
more points where the fluid is still and unmoving, the way in the eye of a
storm the wind is calm (Figure 37(a)). Such points, where the fluid is still,
are called singularities and there are various types of singularity that a
vector field can have. Three vector fields in the plane are shown in Figure
38, with the origin (0, 0) being the singular point for each.
38. Examples calculating indices of vector fields (a) , (b)
, (c) .

The velocity of the flow at the point (x, y) is given by a formula v(x, y).
Figure 38(a) has a source at (0, 0) with the fluid coming out of the origin;
Figures 38(b) and 38(c) are more complicated examples with fluid flowing
in and out of the origin along different lines.

We can associate with a singularity a whole number called its index.

Around each of the singularities in Figure 38 is drawn a circle, redrawn to
the right of each figure with the vector field showing on the circle as
arrows. If we walk anti-clockwise around the circle then the arrows initially
point in a certain direction, change as we move around the circle, but
ultimately must return to their initial direction as we complete the circle.
The index of the singularity is the number of times that the arrows
themselves have gone around anti-clockwise.

So, in Figure 38(a) if we start at 12 o’clock the arrow points north and as
we move anti-clockwise the arrows move around to west at 9 o’clock, south
at 6 o’clock, east at 3 o’clock, and ultimately back to north. The arrows
themselves have gone once anti-clockwise around the compass (north–
west–south–east–north) and so the index is 1 in this case. In Figure 38(b) at
12 o’clock the arrows point east; as we move clockwise around the circle
they go from east to south (at 9 o’clock) to west to north to east—this
journey of east–south–west–north–east is a single journey once around the
compass clockwise and so the index in this case is –1. Finally Figure 38(c):
at 12 o’clock the arrow is pointing west, by 10.30 it’s pointing south, by 9
o’clock it’s pointing east, by 6 o’clock it’s already back to west, and by 12
o’clock the arrows have gone around one further time anti-clockwise before
returning to west. The arrows have been around twice in an anti-clockwise
fashion and the index in this case is 2.

Poincaré’s theorem is a generalization of the hairy ball theorem and states

that
 The indices of the singularities of a vector field on a closed surface add
up to its Euler number.

(Indices is the plural of index.) If we look at the flow in Figure 37(a) then
we see that there is a source at the north pole which has index 1, and a sink
at the south pole which a check shows also has index 1 and then
adds up to the Euler number of the sphere. The hairy-everywhere torus in
Figure 37(b) has no singularities, so the sum of the indices is 0, the Euler
number of the torus.

The hairy ball theorem is then a consequence of Poincaré’s theorem—if

there were a flow on the sphere with no singularities at all, then the Euler
number of the sphere would be 0 and we know it is 2. In fact, we can see
that the only closed surfaces on which there may be flows without
singularities are the torus and Klein bottle, those being the only ones with
an Euler number of 0. Again, we see the Euler number is a global constraint
as to what flows are possible on a surface. We might at any point stir a fluid
in such a way as to make a singularity of our choosing, but there will
necessarily be consequences elsewhere on the surface, so that the total sum
of the indices still equals the Euler number. Poincaré proved his theorem in
1881, which was later generalized in 1926 by Heinz Hopf to higher-
dimensional equivalents of surfaces (manifolds). Consequently, the theorem
is often called the Poincaré–Hopf theorem.

In Chapter 2, we classified surfaces up to continuous equivalence

(homeomorphism)—we might instead have considered smooth surfaces up
to smooth equivalence (the technical word being diffeomorphism). At first
glance these are seemingly different classification problems—going back to
the alphabet I hope you can see that C and I can be smoothly deformed into
one another, but neither into a V because of its pointy base. Continuous
deformations can be jerky but smooth ones need to be fluid. However it
turns out in two dimensions that the classification problem has the same
solution: if we consider a smooth closed surface it must be diffeomorphic to
(a smooth version of) precisely one of or .
The classification theory for three-dimensional manifolds is yet more
complicated and its resolution brings us into the 21st century, but it remains
the case that the continuous and smooth versions of the classification
problem lead to the same solution. But the four-dimensional problem
provided further colour in the 1980s, when it was found that it is impossible
to ‘make smooth’ some topological four-dimensional manifolds and some
others are ‘smoothable’ in essentially different ways.

Structures
Shortly we will move on to a discussion of algebraic topology which
associates with a surface (and other spaces) algebraic structures that capture
something of the essence of the surface’s topology. In Chapter 4 we met
compactness and connectedness, and these are important topological
invariants but, ultimately, they are also binary. A space can be compact or
not—there are no shades of grey here. It would be useful to have an
invariant that retained something subtler of the topological character of a
space. Before that we will need to consider the algebraic structures such
topological invariants might take the form of.

Much of mathematics is concerned with the study of structures. The objects

of mathematics—whether numbers, sets, functions—are often in some
relation to one another or may be combined in certain ways. Much of the
technical language of mathematics describes the details governing such
relationships and combinations. A set (loosely speaking) is a collection of
objects without any further structure and we needed, for example, a metric
to introduce a notion of distance to a set to make it a metric space. The
important sets of mathematics typically come naturally with some further
structure. For example, with the natural numbers

{0, 1, 2, 3, 4 . . .},

we might identify 0 as the smallest of these numbers, but then we’re already
recognizing the set’s implicit order, and we might recognize the numbers
can be added together but not subtracted, at least not if we want the result to
remain in the set (e.g. is in the set but is not).
One of the most common algebraic structures within mathematics is a
group. A group is a set G together with an operation * which, for two
inputs x and y from G, combines them into an output which we denote as x
* y and which is importantly also in G. The set G might be a set of numbers,
functions, geometric transformations or even a set of sets, and the operation
* might be addition, multiplication, composition of functions or some set
operation. But, to be a group, further rules must also apply: for G and * to
form a group we need:

1. for all x, y, z in G;
2. there is an element e in G such that for all x in G;
3. for any x in G there is an element x–1 in G such that

The element e in rule 2 is called the identity and the element x–1 which
combines with x to give e is called the inverse of x. Rule 1 is known as
associativity and the purpose of this rule is that any product like a * b * c *
d needs no further clarification—we’ll always arrive at the same answer no
matter in what order we carry out the three multiplications involved in this
product.

The natural numbers and addition meet rules 1 and 2 (with ) as

for any natural numbers x, y, and z, but fail rule 3, as there is no natural
number x–1 we can add to to get a sum of . We would like x–1
to be –2 but this is not a natural number, being negative. However, there are
many important sets and operations that do meet the three rules—the
subject group theory is a significant part of modern algebra—but as our
focus here is topology, we only mention a handful:

• The set of whole numbers, denoted ℤ, with operation +. In this case

and .
• The set of non-zero real numbers, denoted ℝ*, with operation ≈. In this
case and .
• The set of rotations in the plane about the origin, with the operation °
denoting composition. So x ° y means doing rotation y and then doing
rotation x. Here e is rotation through zero degrees and if x is a rotation
through some angle anti-clockwise, then its inverse x–1 is the rotation
through the same angle clockwise.
• The set ℤ2 of pairs (x, y) of whole numbers with the operation + defined
by

If we could assign groups as topological invariants to spaces, this would be

useful as there is considerable variety amongst groups compared with just
being able to say that a metric space is compact, or not, connected, or not.
Further, the connection, between the groups we will introduce and the
spaces they help describe, has a naturality to it as we’ll see continuous
functions between two spaces yield algebraically nice functions between
their associated groups.

Algebraic topology
Our aim now is to associate with a space a group that captures something of
that space’s topological essence. The great French mathematician and
physicist Henri Poincaré had the idea to consider loops, paths in the space
that begin at a point and return to the same point. We might first try
appreciating Poincaré’s ideas on a torus. Our first problem is that there are
just too many loops; two different loops, such as l1 and l3 in Figure 39(a),
both go once around the hole in the torus in the same direction, but both
capture much the same about the shape of the torus. Secondly, Poincaré was
seeking to define a group and so he needed to find ways to combine loops
with some operation.

39. Loops on a torus (a) Loops on a torus, (b) As on a glued square, (c) Loops from a base
point.

Addressing this second point, we could combine loops by going around one
loop after going around the other; at least we could if the second loop began
where the first loop ended. So, first, we choose a fixed point of the torus, a
base point, and consider only loops that begin and end at the base point;
then, given two such loops l1 and l2, their product l2 * l1 is, reading from the
right, the path that starts at the base point, follows l1 and then l2, still
ultimately returning to the base point.

Would the set of loops based at a point, together with this operation *, make
a group? What, for example, would the identity loop e be, as described in
rule 2? This loop e would have to be such that for all loops l. This
means that the path l followed by e would have to be the same as just the
path l and so e would need to be the path that goes nowhere—e starts at the
base point and immediately finishes at the base point.

And what would the inverse l–1 of a loop l be, as described in rule 3? The
inverse l–1 of l must satisfy , knowing that e is the ‘don’t move’
path at the base point. This is problematic; even if we defined l–1 to be the
return journey of l, that is going around l in the opposite direction, e and l–
1* l would not be equal. To travel from London to Paris and back by the

same route is in no obvious way the same as just staying put in London.
Poincaré was able to resolve this issue with the same idea that he resolved
the first problem, that of there being too many loops. The two loops l1 and
l3 in Figure 39(a) seem to capture the same aspect of the topology of a
torus, but they’re different loops. However, either of them could be
continuously deformed into the other. The technical term for this is that the
two loops are homotopic. Any loop in the torus that is homotopic to l1 in
Figure 39(a) is a loop that goes once around the torus’s hole in the same
direction. But l2, a loop which goes once through the hole, is not homotopic
to either l1 or l3. In Figure 39(b) these loops are again depicted on a torus
represented as a glued square.

Why would it now be that following a path l, and returning along that same
path, is homotopic to making no journey at all? You might think of a
journey that did 99 per cent of the path l and returned from there, and then
one that did the same for 98 per cent, etc. Hopefully you now have the idea
that you can gradually do less and less of that return trip to Paris to the
point where you just stay in London—a kind of spectrum of increasingly
frustrating weekend breaks!

Poincaré introduced the fundamental group of a space in 1895 in his

seminal work Analysis Situs (the title being an old, now obsolete, name for
topology). The elements of the fundamental group are loops based at some
point in the space, but with the understanding that two loops that can be
continuously deformed into one another are to be considered the same. In
the case of the torus—or indeed of any path-connected space—the
fundamental group doesn’t depend on our choice of base point.

In Figure 39(c) a base point b is shown, and two loops l1 and l2, similar to
before but now based at b. A loop (based at b) that goes once around the
hole of the torus is homotopic to l1 and these loops are equal in the
fundamental group. A loop that goes once through the hole is homotopic to
l2. In fact it can be shown that any loop in the torus (based at b) is
homotopic to a journey around l1 some number of times followed by the
journey around l2 a certain number of times. What is being claimed here is
that knowing how many times a loop wraps around the torus’s hole and how
many times it wraps through the hole determines the essence (up to
homotopy) of a loop on the torus.

There is a subtle, implicit point here worth noting. The order in which the
loops wrap around or through the hole does not matter on the torus (but
does for other surfaces). This is because . This is easiest seen
in Figure 39(c). l1 is the loop running along the base (or top) of the square
and l2 is the loop running along the left (or right) side of the square. The
loop l2 * l1 is then the loop starting in the bottom left corner and going right
along the bottom and then up the right side, and l1 * l2 is the loop starting in
the bottom left corner and going up the left side and then right across the
top. We can see that l1 * l2 and l2 * l1 are homotopic—and so equal in the
fundamental group—by imagining the right-then-up path being
continuously dragged across the square until it becomes the up-then-right
path. The square’s diagonal, as in Figure 39(c), is a loop equal to either
product.

Because l1 * l2 = l2 * l1, a loop like l2 * l1 * l1 * l2 * l2 * l1 * l1 in the

fundamental group is the same as l1 * l1 * l1 * l1 * l2 * l2 * l2—the order that
the loops are travelled does not matter, only that four l1s and three l2s are
followed. A loop on the torus is essentially characterized by how many
times it wraps around the hole of the torus (as l1 does) and through the hole
of the torus (as l2 does). The full consequence of all this is that the
fundamental group of a torus is ℤ2, the group of pairs of whole numbers,
with (m, n) representing those loops going around the hole m times and
through the hole n times.

It may not now be that surprising to find that the fundamental group of a
circle is ℤ, the group of whole numbers under addition, as a loop in the
circle is essentially characterized by how many times it wraps around the
circle. Let’s remind ourselves of just what’s being claimed: a loop that goes
around the circle, starting and finishing at the same point, goes around the
circle a certain number of times (counted in an anti-clockwise sense).
Another, different loop may go around the loop the same number of times;
this second journey might have digressions here and there, back-and-forth,
but because the two journeys overall go around the circle the same number
of times then one journey can bit-by-bit be deformed into the second
journey; if the first journey gets behind/ahead of the second journey then it
can become increasingly hurried/delayed to rectify that. But a loop that
goes around the circle twice anti-clockwise could never be deformed into
one that goes once clockwise; these two journeys are essentially different—
that is, they’re not homotopic. These two journeys correspond to the
numbers 2 and –1 in the circle’s fundamental group. Any deformation of a
loop that goes twice around the circle will always be another loop that still
goes twice around the circle.

Finally, to say that the fundamental group of a circle is the group of whole
numbers under addition, means that loops around the circle combine in the
same way integers add. In the group ℤ of whole numbers, we know that the
numbers 2 and –1 add to give 1. In the fundamental group, 2 represents any
loop that goes twice anti-clockwise around the circle and –1 represents any
loop that goes once clockwise around the circle. If we combine such loops
using * then we get a loop that goes in total once anti-clockwise around the
circle and such a loop corresponds to 1 in the group of whole numbers. So
not only do the whole numbers correspond to loops around the circle, but
how whole numbers add corresponds to how loops combine using *.

In Chapter 1, we mentioned that the letter O is not simply connected,

meaning O ‘has a hole in it’. Unsurprisingly neither the circle nor torus are
simply connected either. But we can now formally say what it is for a space
to be simply connected, which means that the space is path-connected with
fundamental group {e}. This means that every loop is homotopic to the
constant loop e. So lines, planes, discs are simply connected; cylinders and
Möbius strips aren’t as a loop going once around either is not homotopic to
e. In general, a loop going around a hole in a space—like an elastic band
caught around a stick—cannot be transformed to a constant loop. The
elastic band analogy is a useful one: the plane with a point removed is not
simply connected—an elastic band can be hooked around the missing point
—but 3D space minus a point is simply connected as an elastic band cannot
be hooked by a single missing point in 3D. But 3D space missing a line isn’t
simply connected as the elastic band can become irremovably hooked on
the line.

A further important feature of fundamental groups is how a continuous

function between two spaces leads to a nice algebraic function between
their fundamental groups. Say f is a continuous function between two
spaces X and Y. If l is a loop in a space X based at a point b, then f(l) is a
loop in the space Y based at the point f(b). Take a moment to appreciate this:
f takes points of X to points of Y and l is a journey through such points of X
beginning and ending in b. So f(l) is a journey of points in Y beginning and
ending at f(b).

As an example, consider when f is the function which wraps the circle twice
on to itself as shown in Figure 40(a). A point making angle θ with the
horizontal is sent to a point making angle 2θ with the horizontal axis—so
the arc from a to p would map to the arc from a to f(p), and the upper
semicircle from a to b would map to the whole circle. And if we had a loop
l in the circle that goes once (anti-clockwise) around the circle then f (l)
goes twice around the circle; if l goes twice around the circle then f (l) goes
four times around the circle. Loops that go n times around the circle are
what n represents in the fundamental group of the circle, and these loops are
sent to loops that go 2n times around the circle, which are what 2n
represents in the fundamental group. The function f wraps the circle twice
onto itself and naturally gives a corresponding function f for loops, with a
loop l that wraps n times around the circle being sent to a loop f(l) that
wraps 2n times around the circle.
40. Functions on the disc (a) f wrapping the circle twice around, (b) A function f from D to D,
(c) Defining the function g.

Say l1 and l2 are loops that wrap around the circle n1 and n2 times, so that l2
* l1 is a loop that wraps around the circle times. So f(l1), f(l2), f(l2 *
l1) are loops that wrap around the circle 2n1, 2n2, and times;
similarly f(l2)*f(l1) wraps around the circle times. As

then both f(l2 * l1) and f(l2)*f(l1) wrap around the circle an equal number of
times, meaning they are homotopic—that is

(H)

as elements of the fundamental group.

More generally a continuous function f between two spaces X and Y

naturally gives a corresponding function f from the fundamental group of X
to the fundamental group of Y, as a loop l based at x is sent to a loop f (l)
based at f (x) and this function satisfies (H) in general. Any function
between groups that satisfies (H) is called a homomorphism. For groups,
homomorphisms are the natural functions to study, being the functions that
respect group operations.
We now use these ideas to prove Brouwer’s fixed point theorem, first
proved by L. E. J. Brouwer in 1910, which states:

Let D be a closed disc and f be a continuous function with inputs and

outputs in D.
Then f has a fixed point: there is some x in D such that .

The theorem applies to a closed disc D, a disc including its circumference.

(It’s not hard to construct a function from an open disc to itself with no
fixed points.) There are many continuous functions from D to D. Such a
function is rotation of D about its centre, and in this case the only fixed
point is the centre. In Figure 40(b), the disc has been shrunk and moved
somewhat to the left. Again, the theorem states that there is a fixed point
(which is unique again in this case). It’s not hard to visualize lots of
different ways that D might be transformed into D and, in all cases, there
must be at least one fixed point.

The theorem’s proof is not easy, perhaps involving the most conceptual
ideas in this text. First the proof is by contradiction which means that we
consider the possibility of there being a function f with no fixed points and
argue from that assumption to a position that is logically impossible. If a
function f has no fixed points, then for all x in D. Because of this
we can define the function g, from D to its circumference C as follows
(Figure 40(c)): as f(x) and x are different points, then we can draw a line
starting from f(x) and going through x which eventually meets the
circumference C at the point g(x). This way any point x in the disc D has
been sent to a point g(x) that lies on its circumference C. Note that if x is
itself on the circumference C, then .

So, there are now several functions we can consider:

• i, inclusion, which has inputs in C and outputs in D;

• g which has inputs in D and outputs in C;
• the composition g ° i, which has inputs in C and outputs in C.
Here inclusion means the function which takes a point of C, the
circumference, as an input, and returns the same point as the output, now
considered as a point of the disc D. The composition g ° i means the
function which performs inclusion i first and then performs g second.
Earlier we noted g fixes points on the circumference C and so, importantly,
the function g ° i fixes all points of C.

We arrive at our contradiction by considering the corresponding functions

on the fundamental groups of C and D. Recall that the fundamental group
of C, a circle, is the group of whole numbers, with the number n
representing those loops that wrap n times anti-clockwise around the circle;
the fundamental group of the disc D, which is simply connected, is {e}. The
corresponding functions for the fundamental groups satisfy the following:

• i has inputs in the whole numbers and an output of e, whatever the input,
so that this function is constant;
• g has a single input e and so some single output in the whole numbers;
• the composition g ° i, which has inputs and outputs in the whole
numbers.

Remember that the composition g ° i fixes all points of C and so any loop in
C is sent to the same loop by the corresponding function g ° i on loops. So,
in terms of the fundamental group of C, the function g ° i sends each whole
number n to itself. Instead we can consider the separate effects of doing the
corresponding function i first and then the corresponding function g. As
noted, i sends all the whole numbers to e, which g subsequently sends to
g(e). Note that this number g(e) is a single whole number not depending on
n.

Looked at one way, the function g ° i on loops sends all whole numbers n to
themselves, but considering the functions’ effects separately, every n is sent
to g(e); whatever this value g(e) is, it is a single value and so g ° i is a
constant function. This is the required contradiction—the function g ° i
cannot both send each n to n and also send each n to a single value g(e). So
a function f, without fixed points, cannot exist or else we’d be able to define
the function g and arrive at a contradiction.
The ideas of this proof are subtle, but the crux of it, and what makes such
methods powerful, is the following: fundamental groups (and other similar
algebraic invariants) are able to capture something of the topological
essence of a space; continuous functions between spaces lead to
algebraically nice functions (homomorphisms, that have the earlier property
H) between the spaces’ fundamental groups; if it can be shown no such
homomorphism exists, then no such continuous function existed.

There are further algebraic topological invariants capturing the essence of a

space in higher dimensions. The plane with a point missing is not simply
connected as a loop that goes around the missing point cannot be
‘unhooked’. 3D space missing a point is simply connected though—an
elastic band (a metaphorical circle) in 3D cannot be hooked by a single
missing point, but we can imagine a balloon (a metaphorical sphere) being
hooked onto the missing point. So there is something topological occurring
in punctured 3D space, but in a higher dimension than the fundamental
group, with its loops, can measure.

Some such information can be captured by Betti numbers, named by

Poincaré after Enrico Betti who first studied these. These Betti numbers are
topological invariants and an n-dimensional space has Betti numbers b0, b1,
b2 … bn with the ith Betti number bi capturing something of the essence of
the space’s topology in the ith dimension. , the torus with g holes, has
Betti numbers

where b1 captures there being 2g loops in the torus, one around each of the
g holes of the torus, and one in and through each hole. And the Betti
numbers of ℙ#k are

The Euler number of a space is defined in terms of its Betti numbers as the
alternating sum
so that the Euler number of equals and of ℙ#k
equals . That these Betti numbers are topological
invariants then means that the Euler number is a topological invariant.

Poincaré is often considered as the last universalist in mathematics,

someone making research contributions across mathematics, and he became
interested in topology via various routes. In 1889 he won a mathematical
competition funded by Oscar II, King of Sweden and Norway, for
contributions to understanding how a multiple body system (like the planets
in the solar system under gravity) evolves with time. So, for Poincaré,
topology was a means for qualitatively analysing such systems. In pure
mathematics, he was interested in the topology of spaces in their own right,
essentially studying manifolds though still lacking at the time a wholly
rigorous definition for such spaces.

It was also Poincaré who introduced combinatorial topology to help

calculate algebraic topological invariants. As I mentioned in Chapter 1,
there are many complicated examples of Jordan curves but, ultimately, all
of them are homeomorphic to a triangle; likewise the sphere’s topology is
no different from a cube, nor a torus’s from the polyhedron in Figure 11(a).
By using polygons, polyhedra, and higher-dimensional equivalents, quite
general spaces can be approximated by simpler polyhedra that can be
finitely described but still capture all of their topological essence. But,
again, Poincaré’s ideas were far-sighted but not entirely rigorously realized.

For around the next forty years, topologists would take forward and make
rigorous Poincaré’s vision, making algebraic topology a major theme of
mathematics and fully appreciating the range and power of his ideas, with
James Alexander—more of him in Chapter 6—perhaps being most
prominent in that role. Arguably the last piece of Poincaré’s legacy came
with the proof in 2003 of the Poincaré conjecture by Grigory Perelman.
The conjecture states that every simply connected, compact three-
dimensional manifold is homeomorphic to the three-dimensional sphere
(this does not mean a solid ball in 3D, but rather a 3D spherical shell that
sits naturally in 4D). His proof, based on a strategy developed by Richard
Hamilton, used ideas of differential geometry far beyond even Poincaré’s
imagination at the time. However, its solution is more evidence for how the
great problems of mathematics have led to progress in highly novel
directions.
Chapter 6
Unknot or knot to be?

Describing knots
A knot is a smooth, simple, closed curve in 3D space. Being simple and
closed means the curve does not intersect itself except that its end returns to
its start. Basically a knot is a loop in 3D space and by requiring smoothness
we exclude some nasty, so-called wild, knots from our study. As I noted in
Chapter 1, all knots are topologically the same as a circle; what makes a
circle knotted—or not—is how that circle has been placed into 3D space.

As knots—in and of themselves—are just circles, then we need a new

notion to say when two knots are the same, as situated in 3D space. That
notion needs to capture the idea that one knot, and the space around it, can
be continuously deformed into the other, and the space around it; this is
called an ambient isotopy.

So, two knots K1 and K2 are to be considered equivalent if we can start with
knot K1 and over a period of time continuously deform 3D space so that K1
becomes K2. If we set the time taken to deform the knots as a unit interval 0
⩽ t ⩽ 1, then by some time t in the middle of that interval we will have
deformed 3D space by some homeomorphism ht. An ambient isotopy is
then a continuous family of such ht, where at time we have yet to start
deforming so that and by the time the first knot has been
deformed into the second, that is .
The central problem of knot theory is then a classification theorem: when
are two knots equivalent—there is an ambient isotopy between them—or
how do we show that no such isotopy exists? This problem was first
identified by Maxwell in 1868, though his work went unpublished at the
time.

A more basic problem of knot theory is first a means of describing knots.

We can imagine laying any knot flat on a table top. Not literally flat—the
essence of the knot will be in how at certain crossings one part of the knot
goes over or under another bit of the knot (Figure 41); we can also be
careful to separate out those crossings so that no more than two parts of the
knot cross at the same place. The minimal number of crossings of a knot is
a knot invariant, so that equivalent knots have the same minimal number of
crossings. In Figure 41(a) appears a version of the unknot; by an unknot we
mean a loop that is not actually knotted, and so is isotopic to a circle. Figure
41(a) has two crossings but the same unknot can be represented as a circle
with zero crossings. The first significant effort at describing and classifying
knots by means of over-and-under crossings was by Peter Guthrie Tait,
between 1877 and 1885, who classified all knots with ten or fewer
crossings.

41. The unknot and trefoils (a) An unknot, (b) Left-handed Trefoil, (c) Right-handed Trefoil.

Knots only exist in 3D; any loop in 4D can be unknotted. Beginning from
the rightmost point of the unknot in Figure 41(a), and moving anti-
clockwise around the loop, we see that the two crossings both go over the
other part of the knot. More generally any loop that we could lay flat on the
table, at each crossing laying the loop over a previous part of the loop,
would lead to an unknot, as we could lift the loop off the table one crossing
at a time to make a circle. Imagine a genuine knot, with various under- and
over-crossings, but with us now permitted to move in 4D. Much as we
earlier used time as the fourth dimension to avoid the Klein bottle
intersecting itself, we could use this extra dimension to make any under-
crossing into an over-crossing; we could take the under-part of the knot
smoothly into the future, raise it above the over-part of the knot which only
exists in the present, and then smoothly return the knot from the future to
the present, now as an over-part. In this way all knots could be unknotted in
4D.

The trefoil is the simplest genuine knot and has three crossings. (You might
want to sketch for yourself loops with one or two crossings and convince
yourself these aren’t proper knots.) The trefoil though is complicated
enough to exhibit an important feature, that of chirality. The trefoil knots in
Figures 41(b) and 41(c) are mirror images of one another and aren’t isotopic
to one another. The trefoil is also an example of an alternating knot, a knot
where the crossings alternate between over- and under-crossings as we
travel around the knot. Alternating knots provide a somewhat simpler class
of knots which are well understood compared with knots generally.

Reidemeister moves
The unknot in Figure 41(a) can be quickly deformed into a circle by
untwisting it at the top, and doing similar to the twist at the bottom. Such a
move—a twist or untwist—is the first of three Reidemeister moves that
between them can be used to show any equivalent knots are indeed
equivalent. These moves are named after Kurt Reidemeister, who proved
this result in 1927, though the same result had been independently proved
by Alexander and Briggs in 1926, and in fact the moves were known much
earlier to Maxwell.

The first move then, R1, is a twist or untwist of a small part of the knot
(Figure 42). The second move R2 takes a small part of the knot and pokes it
under another part of the knot or undoes this. Finally, the third move R3 is
to slide a single crossing over a small part of the knot. Hopefully none of
these moves seems controversial and are evidently permitted manipulations
of the knot. The significant result is that, given any two equivalent knots,
repeated use of these three moves alone is sufficient to show that the two
knots are indeed equivalent.

42. The three Reidemeister moves.

Unfortunately, many theorems in mathematics demonstrate the existence of

a solution, without being constructive—that is, providing a means to find
that solution—or, even better, providing an efficient constructive means to
finding a solution. In 1961 Wolfgang Haken showed that the unknotting
problem is decidable, in the sense that there is a general algorithm that can
be implemented to decide whether a knot is the unknot. Others would later
extend this result to show the equivalence problem is decidable, that there is
an algorithm for checking whether two knots are equivalent. But it is still an
open problem as to whether there is an efficient algorithm for this problem.

Prime knots and adding knots

In Figure 43 appear the granny knot and the reef knot. Notice how the
granny knot is reminiscent of two merged trefoil knots—in fact, it is the
connected sum of two trefoil knots with the same handedness, whilst the
reef knot is the connected sum of a trefoil with its mirror image. As in
Chapter 2 with surfaces, we can create the connected sum K1#K2 of two
knots K1 and K2 by removing a small arc of each knot and gluing opposite
loose ends together.

43. The granny and reef knots (a) A granny knot, (b) A reef (or square) knot.

If K1 and K2 have n1 and n2 minimal crossings, then K1#K2 can clearly be

drawn with crossings. However, it is a currently unsolved problem
as to whether the minimal number of crossings for K1#K2 equals .
This has been shown to be true for alternating knots, and is suspected to be
true in general, but remains unresolved.

As our central problem is to classify knots then it makes sense to focus on

those knots that are not connected sums of simpler knots. Such knots are
called prime knots. Not counting mirror images, then the numbers of prime
knots with crossing number of 12 or less are given in Table 6.

Table 6. Number of prime knots sorted by minimal crossing number

Note how this number of different prime knots spirals enormously as the
minimal crossing number increases. It is an open and active area of research
seeking to prove asymptotic estimates for how quickly the number of prime
knots grows as the crossing number becomes large. Figure 44 depicts these
prime knots up to a minimal crossing number of seven, all of which are
alternating; the simplest non-alternating prime knots have a minimal
crossing number of eight. In 1998 asymptotic estimates were demonstrated
for prime, alternating knots; the same result also showed that alternating
knots become increasingly rare, as a fraction of all knots, as the minimal
crossing number increases.

44. Prime knots with seven or fewer crossings.

The knot group

Reidemeister moves provide a difficult-to-implement means for deciding
whether two knots are equivalent or not, and so topologists seek knot
invariants that are easy to calculate but, ideally, separate out inequivalent
knots. All knots are, in themselves, just circles—it’s how that circle has
been situated in 3D space that makes it a particular knot. So we might focus
instead on the complement of the knot, the remainder of 3D space that isn’t
the knot, to better understand knots. We might consider the fundamental
group of the complement, and this is known as the knot group.

Recall that the elements of a fundamental group are loops based at a point,
with loops understood to be the same if one can be deformed into the other.
So, starting from and finishing at a point outside a knot, these different
loops might weave in and out of a knot helping to capture something of the
essence of the knot. And this is indeed the case, but unfortunately knot
groups remain very complicated objects. The knot group of the unknot (a
circle) is just the group of whole numbers, with a loop being entirely
characterized by how many times it wraps in and through the unknot. But
even the knot group of the trefoil is a difficult group to describe.

Wilhelm Wirtinger, around 1905, found a way to describe such knot groups
in general. In Figure 45 is drawn an oriented trefoil with the three unbroken
arcs between the crossings labelled as a1, a2, a3 and the five regions the
trefoil splits the plane into denoted R1 … R5. Imagine the base point b of the
knot group being outside the trefoil, and loops beginning at the base point,
weaving in and out of the knot, and returning back to the base point. We
will write l1 for a loop that goes down through R3 and returns back out of
R5, or equally down through R1 and returns back out of R4. This is a loop
from the base point b that ‘hooks’ a1 going in on the left of a1 and coming
back out on the right of a1. We denote by l2 and l3 similar loops that hook a2
and a3 in a left-to-right manner. Their inverses are loops that
hook a1, a2 and a3 in the reverse right-to-left direction. Any loop beginning
and ending in the base point b can be written as a string using the symbols
such as
45. The knot group of a trefoil.

a recipe for how the loop weaves in and out of the knot, with careful
attention as to whether the loop went in-left and out-right or vice versa. But
already we see that this can’t be the whole description of the knot group: in
the middle of the above ‘recipe’ is the expression which cancel out one
another (remembering back to Chapter 5 and the diminishing London–Paris
return trip). To do a loop and then do it in reverse is essentially the same as
not moving. So, we should omit all occurrences of expressions like
, etc.

However, there are other ‘relations’ between these loops l1, l2, l3 and there
is one such relation for each crossing of the knot. Consider a loop from b
that hooks the knot at the crossing between R1, R3, R4, R5; this is a loop that
goes down through R1 and back out through R5. This can be achieved as
l3l1, which means doing l1 first and l3 second; so we go in through R1 and
return out of R4 and then back in through R4 and out of R5. Or we could
manage the same by doing l1l2 which means doing l2 first and l1 second;
this takes us in through R1 and returns out of R3 and then down through R3
and out of R5. Either of these loops hooks the knot at the crossing point and
so it’s the case that l3l1 = l1l2.

Now if there were n crossings in a knot, then we’d have strings involving l1
… ln and in the knot group. Wirtinger’s theorem showed that the
knot group consists of all such strings, with two such strings describing the
same loop if one string can be turned into the other by cancelling terms like
or using rules, like l3l1 = l1l2, with one such rule coming from each
crossing.

All this is impressive, given its generality, but is not very tractable given
our aim is to describe simple invariants to separate out inequivalent knots.
It’s also frustrating to find that inequivalent knots, such as the reef knot and
the granny knot, can have the same knot groups, as do left- and right-
handed trefoils. However, it was shown in the late 1980s that the knot group
determines a prime knot up to mirror images: two prime knots with the
same knot group are isotopic knots, or each knot is isotopic to the mirror
image of the other.

Alexander and Jones polynomials

In 1928 James Alexander introduced what is now known as the Alexander
polynomial and which, for a knot K, is usually denoted ΔK(x). Here x is the
variable of the polynomial. The Alexander polynomial is a knot invariant
and much simpler than the knot group, though it ultimately conveys less
information about a knot—knots with the same knot group have equal
Alexander polynomials.

The Alexander polynomial for the unknot is the constant polynomial 1 and
the Alexander polynomial of the trefoil T equals
So technically the Alexander polynomial is a polynomial in the variables x
and x–1. In his 1928 paper Alexander described an algorithm for calculating
his polynomial from an over- and under-crossings description of the knot
and that calculation, for the trefoil, is done in the Appendix. The algorithm
is a little technical but ultimately uses mathematics taught in schools and
colleges.

The Alexander polynomial also deals well with composite knots, having the
nice algebraic property

for two knots K and L, with K#L denoting their connected sum. So, the reef
and granny knots both have Alexander polynomial .

The Alexander polynomial cannot distinguish between mirror images, but

does distinguish between prime knots of up to eight crossings. Surprisingly
there are non-trivial knots, including one with just eleven crossings that has
a constant Alexander polynomial equal to 1, so the Alexander polynomial
cannot distinguish the unknot from all other knots.

It was some considerable time later, 1984, when a second polynomial

invariant, the Jones polynomial, was discovered by Vaughan Jones, for
which he would win the Fields Medal in 1990. I won’t define the Jones
polynomial algorithmically here, but rather describe properties which
characterize it uniquely.

The Jones polynomial of a knot L is denoted VL(x) and is actually a

polynomial in the variables and . The Jones polynomial of the
unknot is 1. The following skein relation then characterizes the Jones
polynomial. (The word ‘skein’ means a quantity of thread or yarn.)

We consider three knots L+, L0, and L– that differ only at one crossing. The
knots L+ and L– differ in which part of the knot makes the over-crossing and
the knot L0 has no crossing at all at this point (Figure 46).

46. Links involved in the skein relation.

The skein relation then states that

As mentioned before, if we made all the crossings of a knot over-crossings

(or all under-crossings) then we make the unknot. If in a knot an over-
crossing L+ makes that knot more ‘knotted’ than an under-crossing L– then
the skein relation describes the Jones polynomial of the more complicated
knot L+ in terms of the less complicated knot L– and a knot L0 with one
fewer crossing.

We will shortly use the skein relation to calculate the Jones polynomial of
the trefoil, but an important point has so far been ignored. Even if L+ and L–
are knots, L0 need not be. As we will see in the example of the trefoil,
eliminating a crossing might disconnect a knot and create what is called a
link, which is just a collection of knots. The skein relation, and Jones
polynomial generally, should be seen as relating to such links.

The Jones polynomial can sometimes differentiate between a knot K and its
mirror image K* as the identity holds; so, we need to
specify the trefoil being considered as right-handed. In our calculation we’ll
need to consider various knots and links (Figure 47).
47. Simple examples of links (a) Right-handed trefoil, (b) Twisted unknot, (c) Unlinked circles,
(d) Linked circles.

We will take L+ as the trefoil and focus on the bottom-right of the three
crossings. L- is then the unknot, perhaps easiest seen by unpoking the
bottom part of the knot with a Reidemeister move. And L0 has become
disconnected making two linked circles as in Figure 47(d). Remembering
that the skein relation states
where Vlinked(x) is the Jones polynomial of two linked circles. Using the
skein relation a second time, we take L+ as the linked circles, and consider
the rightmost crossing. Then L– is two unlinked circles (Figure 47(c)) and
L0 is the unknot. This time the skein relation states

Finally, we apply the skein relation to the twisted unknot (Figure 47(b)) to
find out Vunlinked(x). If we take L+ as the twisted unknot then L– is another
twisted version of the unknot and L0 is two unlinked circles. The skein
relation then states

which rearranges to give

Substituting this into earlier equations gives

and then with a little rearranging we find

So for the trefoil’s
mirror image, and the Jones polynomial can differentiate between the left-
and right-handed trefoils. The Jones polynomial distinguishes between
prime knots with up to nine crossings, but not beyond. Unlike the
Alexander polynomial it remains an unsolved problem as to whether there
exists a genuine knot with a Jones polynomial equal to 1.
Epilogue

In 1911 the French mathematician Jacques Hadamard wrote that

Analysis situs … constitutes a revenge of geometry on analysis.

(Recall that ‘analysis situs’ is an old name for topology.) Certainly

geometric topology and the visualization of Riemann, Klein, Möbius, and
Poincaré were in marked contrast to the analysis of Weierstrass. Riemann’s
and Poincaré’s work would have massive influence on the development of
mathematics. Topology would go on to become one of the central themes of
mathematics.

And from the earliest history of topology, connections with physics would
be apparent. In the 19th century Gauss and Maxwell would both recognize
such in electromagnetism—for example, in the study of the work done by a
magnetic pole moving in the presence of a wire carrying current, with
Gauss’s answer being in terms of a linking number for the pole’s path and
the wire. Both Gauss and Maxwell would bemoan the lack of progress with
the study of topology or ‘geometry of position’ as Maxwell referred to it at
the time.

The 20th century would develop a yet richer connection between topology
and physics. In 1965, Roger Penrose would use topological ideas to
demonstrate how the gravitational collapse of a massive star would lead to a
space-time singularity occurring, such as a black hole. The use of topology
meant that Penrose was able to impose qualitative assumptions about the
mass distribution, compared with earlier assumptions about the symmetric
distribution of matter that had been considered physically questionable.

The interaction between physics and topology would also not be one way. A
problem that was ostensibly in physics would become of interest to pure
mathematicians if it could be rephrased into mathematical language
involving mathematical objects. This was particularly the case with Yang–
Mills theory, a physical theory seeking to provide a unified description for
electromagnetism and the weak force. In 1983 Simon Donaldson would use
ideas from Yang–Mills theory to prove astonishing results about the
topology of four-dimensional manifolds.

Topology remains a large, active research area in mathematics.

Unsurprisingly its character has changed over the last century—there is
considerably less current interest in general topology, but whole new areas
have emerged, such as topological data analysis to help analyse big data
sets. The interfaces of topology with other areas, including physics, have
remained rich and numerous, and it can be hard telling where topology
stops and geometry or algebra or analysis or physics begin. Often that
richness comes from studying structures that have interconnected flavours
of algebra, geometry, and topology, but sometimes a result, seemingly of an
entirely algebraic nature say, can be proved by purely topological means. In
the words of Poincaré

Mathematics is the art of giving the same name to different things

and the rise of topology has certainly helped demonstrate the

interconnectedness of mathematics.
Appendix: Calculating an Alexander
polynomial

Calculating the Alexander polynomial of a knot is a little technical but

involves only mathematics that might be met at school or college, a
knowledge of matrices and determinants. The method below was given by
Alexander in 1928.

An oriented knot K, with n crossings c1, c2 … cn, divides the plane into
regions, , including the outside of the knot. Figure 48
shows a trefoil labelled in this manner.

48. Calculating the Alexander polynomial of a trefoil.

From such a diagram we create a matrix with n rows (corresponding to the
n crossings) and columns (corresponding to the regions). The
entries of this matrix are then filled according to the following rules:

• If the region is not adjacent to the crossing, the entry is 0.

• If the region is on the right before under-crossing, the entry is 1.
• If the region is on the left after under-crossing, the entry is x.
• If the region is on the right after under-crossing, the entry is −1.

This matrix for the trefoil in Figure 48 is

To explain the first row, R4 is not adjacent to c1 and whilst travelling

towards c1 as an under-crossing, the region R1 is on the left and R2 is on the
right and having passed through c1 the region R3 is on the left and R5 is on
the right.

We now remove any two columns corresponding to adjacent regions, say

R1 and R2 in this case, so that we have a square n by n matrix, and we take
the determinant of this matrix. For the trefoil this gives
The answer at this point depends somewhat on which columns we removed.
If we divide by a power of x so that the highest power of x is the same as
the highest power of x–1 (in our example we need to divide by x) and
multiply by ±1 so that the polynomial takes value 1 at x = 1, then we have
calculated the Alexander polynomial. In the case of the trefoil knot, that
equals .
Historical timeline
1639 Descartes’s work on angular defect, equivalent to Euler’s formula

1750 Euler discovers his formula

1817 Bolzano defines continuity, proves the Bolzano–Weierstrass and intermediate value theorems

1836 The word ‘topology’ is coined by Johann Listing

1851 Schläfli generalizes Euler’s formula to higher dimensional polyhedra

1857 Riemann’s paper Theory of Abelian Functions makes Riemann surfaces more widely known

1858 Möbius and Listing independently discover the Möbius strip

1861 Weierstrass lectures on the boundedness theorem

1861 Möbius gives a first sketch proof of the classification theorem for closed surfaces

1868 Maxwell first states the knot classification problem

1871 Betti publishes On Spaces of any Number of Dimensions

1874 Klein shows orientable closed surfaces are homeomorphic if and only if they have equal genus

1876 Klein gives his definition of one-sidedness

1881 Poincaré’s theorem, later generalized by Hopf in 1926

1882 Klein first describes the Klein bottle

1887 Jordan curve theorem appears in his Cours d’analyse

1895 Borel states a first version of the Heine–Borel theorem

1895 Poincaré defines the fundamental group in his Analysis Situs

1906 Fréchet defines metric spaces in his doctoral thesis

1907 Max Dehn and Poul Heegaard give the first rigorous proof of the classification theorem

1910 Brouwer proves his fixed point theorem

1914 Hausdorff defines topological spaces in the seminal Grundzüge der Mengenlehre

1919 Hausdorff dimension introduced, which can take fractional values

1925 Morse publishes his paper Relation between the Critical Points. . .

1927 Reidemeister moves introduced

1928 Alexander introduces his polynomial knot invariant

1930 Kuratowski’s theorem on planar graphs

1932 Čech defines higher homotopy groups, generalizing the fundamental group

1936 Whitney defines manifolds and proves his embedding theorem

1950 Hamming distance introduced in the paper Error Detecting and Error Correcting Codes

1952 Moise shows that every three-dimensional manifold is uniquely smoothable

1961 Haken shows that the unknotting problem is decidable

1982 Thurston wins Fields Medal for work on three-dimensional manifolds

1984 Jones polynomial introduced

1986 Donaldson and Freedman win Fields Medals for their work on four-dimensional manifolds

2003 Perelman proves the Poincaré conjecture

Further Reading

Colin Adams, Knot Book (2004) American Mathematical Society

Jeremy Gray, Henri Poincaré: A Scientific Biography (2012) Princeton
Stephen Huggett and David Jordan, A Topological Aperitif (2009) Springer
Matt Parker, Things to Make and Do in the Fourth Dimension (2014) Penguin
V. V. Prasolov, Intuitive Topology (1994) American Mathematical Society
David Richeson, Euler’s Gem: The Polyhedron Formula & the Birth of Topology (2012) Princeton

Online references
3Blue1Brown, Who Cares about Topology?<http://www.youtube.com/watch?
v=AmgkSdhK4K8&t=236s>
The Geometry Junkyard, 20 Proofs of Euler’s Formula
<http://www.ics.uci.edu/~eppstein/junkyard/euler/all.html>
Index

Note: For the benefit of digital users, indexed terms that span two pages (e.g., 52–53) may, on
occasion, appear on only one of those pages.

A
Alexander, James 114, 118, 123, 131
Alexander polynomial 123–124, 131
algebraic topology 90, 103–114
alternating knot 117, 120
ambient isotopy 115–116
Analysis Situs 105, 128

B
Betti numbers 112
blancmange function 62–63, 90
Bolzano, Bernard 53, 61
Bolzano–Weierstrass theorem 84
boundedness theorem 61, 85, 93
Brouwer’s fixed point theorem 109

C
Cauchy, Augustin-Louis 9
chirality 117, 123
classification theorem for knots 116
classification theorem for surfaces 38–40
closed sets 81
closed surfaces 29
combinatorial topology 113
compactness 63, 84–85
complex numbers 40–43
connected components 19, 86
connected sums 31–33, 119, 124
connectedness 63, 85–87
continuity 48, 57–62, 69–73, 76–77
convex sets 87
convergence 80
critical points 91–95
cut points 88

D
degree-genus formula 46
Descartes, René 10
differential topology 90–96
dimension 19, 21
domain of a function 51, 63
dual polyhedra 13

E
equivalent metrics 73–76
Euler number 25, 32–33, 37–40, 90, 95, 99, 112–113
Euler, Leonhard 5–6
Euler’s formula 5–10

F
Flatland 21–22, 82–83
football, see truncated icosahedron
fractals 21
Fréchet, Maurice 67, 78, 85
functions 50–51, 64
fundamental group 103–108
fundamental theorem of algebra 42

G
Gauss, Carl 42, 128
general topology 21, 78, 90
genus 33
graph theory 15–18
groups 101
H
hairy ball theorem 96–99
Hamming distance 68
Hausdorff, Felix 78, 85, 90
Heine–Borel theorem 85
historical timeline 135–136
homeomorphic 3–5
homomorphism 109
homotopic 104

I
index of a singularity 98
intermediate value theorem 60, 86
invariants 5, 85, 87–9, 100, 112–113

J
Jones polynomial 124
Jordan curve theorem 19
Jordan, Camille 19

K
Klein bottle 36–37
knot group 121–123
knots 22–23, 115

L
Lakatos, Imre 10
limit 80
links 125–6

M
manifolds 47, 99, 112–114
Maxwell, James Clerk 95, 116, 118, 128
metric spaces 67–76
Möbius strip 33–35
Morse theory 95

O
one-sided surfaces 33–39
open balls 73
open sets 76
orientable surfaces 35, 39

P
path-connectedness 86
physics 43, 128–129
Platonic solids 6, 10–13
Poincaré conjecture 114
Poincaré–Hopf theorem 99
points at infinity 43–45
prime knots 119–120
projective plane 38

R
Reidemeister moves 117–118
Riemann surfaces 45
Riemann, Bernhard 40, 46, 128

S
simply connected 5, 89, 107
singularity 96
skein relation 125
smooth functions 90–91
subdivisions 29–31
subspaces 82–84
surfaces 24–47

T
topological data analysis 129
topological spaces 78
topology (of open sets) 78
trefoil 22, 117
triangle inequality 67
truncated icosahedron 6, 14

U
Underground map 1–2
unknot 22, 116
unknotting problem 118

W
Weierstrass, Karl 53, 62, 128
Whitney, Hassler 47
Wirtinger presentation 121–123
ECONOMICS
A Very Short Introduction
Partha Dasgupta
Economics has the capacity to offer us deep insights into some of the most formidable
problems of life, and offer solutions to them too. Combining a global approach with
examples from everyday life, Partha Dasgupta describes the lives of two children who live
very different lives in different parts of the world: in the Mid-West USA and in Ethiopia. He
compares the obstacles facing them, and the processes that shape their lives, their families,
and their futures. He shows how economics uncovers these processes, finds explanations
for them, and how it forms policies and solutions.

‘An excellent introduction . . . presents mathematical and statistical findings in straightforward

prose.’
Financial Times

www.oup.com/vsi
INFORMATION
A Very Short Introduction
Luciano Floridi

Luciano Floridi, a philosopher of information, cuts across many subjects,

from a brief look at the mathematical roots of information - its definition
and measurement in ‘bits’- to its role in genetics (we are information), and
its social meaning and value. He ends by considering the ethics of
information, including issues of ownership, privacy, and accessibility;
copyright and open source. For those unfamiliar with its precise meaning
and wide applicability as a philosophical concept, ‘information’ may seem a
bland or mundane topic. Those who have studied some science or
philosophy or sociology will already be aware of its centrality and richness.
But for all readers, whether from the humanities or sciences, Floridi gives a
fascinating and inspirational introduction to this most fundamental of ideas.

‘Splendidly pellucid.’
Steven Poole, The Guardian

www.oup.com/vsi
INNOVATION
A Very Short Introduction
Mark Dodgson & David Gann
This Very Short Introduction looks at what innovation is and why it affects us so profoundly.
It examines how it occurs, who stimulates it, how it is pursued, and what its outcomes are,
both positive and negative. Innovation is hugely challenging and failure is common, yet it is
essential to our social and economic progress. Mark Dodgson and David Gann consider the
extent to which our understanding of innovation developed over the past century and how it
might be used to interpret the global economy we all face in the future.

‘Innovation has always been fundamental to leadership, be it in the public or private arena.
This insightful book teaches lessons from the successes of the past, and spotlights the
challenges and the opportunities for innovation as we move from the industrial age to the
knowledge economy.’
Sanford, Senior Vice President, IBM

www.oup.com/vsi
NOTHING
A Very Short Introduction
Frank Close
What is ‘nothing’? What remains when you take all the matter away? Can empty space - a
void - exist? This Very Short Introduction explores the science and history of the elusive
void: from Aristotle’s theories to black holes and quantum particles, and why the latest
discoveries about the vacuum tell us extraordinary things about the cosmos. Frank Close
tells the story of how scientists have explored the elusive void, and the rich discoveries that
they have made there. He takes the reader on a lively and accessible history through
ancient ideas and cultural superstitions to the frontiers of current research.

‘An accessible and entertaining read for layperson and scientist alike.’
Physics World

www.oup.com/vsi
NUCLEAR POWER
A Very Short Introduction
Maxwell Irvine
The term ‘nuclear power’ causes anxiety in many people and there is confusion concerning
the nature and extent of the associated risks. Here, Maxwell Irvine presents a concise
introduction to the development of nuclear physics leading up to the emergence of the
nuclear power industry. He discusses the nature of nuclear energy and deals with various
aspects of public concern, considering the risks of nuclear safety, the cost of its
development, and waste disposal. Dispelling some of the widespread confusion about
nuclear energy, Irvine considers the relevance of nuclear power, the potential of nuclear
fusion, and encourages informed debate about its potential.

www.oup.com/vsi
NUMBERS
A Very Short Introduction
Peter M. Higgins
Numbers are integral to our everyday lives and feature in everything we do. In this Very
Short Introduction Peter M. Higgins, the renowned mathematics writer unravels the world of
numbers; demonstrating its richness, and providing a comprehensive view of the idea of the
number. Higgins paints a picture of the number world, considering how the modern number
system matured over centuries. Explaining the various number types and showing how they
behave, he introduces key concepts such as integers, fractions, real numbers, and
imaginary numbers. By approaching the topic in a non-technical way and emphasising the
basic principles and interactions of numbers with mathematics and science, Higgins also
demonstrates the practical interactions and modern applications, such as encryption of
confidential data on the internet.

www.oup.com/vsi
TOCQUEVILLE
A Very Short Introduction
Harvey Mansfield
No one has ever described American democracy with more accurate insight or more
profoundly than Alexis de Tocqueville. After meeting with Americans on extensive travels in
the United States, and intense study of documents and authorities, he authored the
landmark Democracy in America, publishing its two volumes in 1835 and 1840. Ever since,
this book has been the best source for every serious attempt to understand America and
democracy itself. Yet Tocqueville himself remains a mystery behind the elegance of his
style. In this Very Short Introduction, Harvey Mansfield addresses his subject as a thinker,
clearly and incisively exploring Tocqueville’s writings-not only his masterpiece, but also his
secret Recollections, intended for posterity alone, and his unfinished work on his native
France, The Old Regime and the Revolution.

www.oup.com/vsi