ALEXEI SOSSINSKY

TRANSLATED BY CISH LE WEI S,.
KNOTS
MATHEMATICS WITH A TWIST
HARVARD UNIVERSITY PRESS
CAMBRIDGE, MASSACHUSETTS
LONDON, ENGLAND
2002
Copyright © 2002 by the President and Fellows of Harvard College
ALL RIGHTS RESERVED
Printed in the United States of America
Originally published as Nreuds: Genese d'une Theorie Mathematique
© Editions du Sew.!. 1999
Illustrations by Margaret C Nelson
Design by Marianne Perlak
Library of Congress Cataloging-in-Publication Data
Sossinsky. A.B.
(Na:uds. English]
Knots mathematics with a twist I Alexei Sossinsky ; translated by Giselle Weiss.
p. em.
Includes bibliographical references
ISBN 0-674-00944-4 (alk. paper)
1. Knot theory. 2. Low-dimensional topology. I. Title.
QA612 2 S6713 2002
514' 224-dc21 2002027295
CONTENTS
PREFACE vii
1. ATOMS AND KNOTS 1
Lord Kelvin • 1860
2. BRAIDED KNOTS 15
Alexander • 1923
3. PLANAR DIAGRAMS OF KNOTS 35
Reidemeister • 1928
4. THE ARITHMETIC OF KNOTS 46
Schubert· 1949
5. SURGERY AND INVARIANTS 58
Conway· 1973
6. JONES'S POLYNOMIAL AND SPIN MODELS 73
Kauffman • 1987
7. FINITE-ORDER INVARIANTS 90
Vassi/iev • 1990
8. KNOTS AND PHYSICS 105
Xxx? • 2004?
NOTES 121
WORKS CITED 127
PREFACE
Butterfly knot, clover hitch knot, Gordian knot, hangman's knot, vi-
pers' tangle-knots are familiar objects, symbols of complexity, oc-
casionally metaphors for evil. For reasons I do not entirely under-
stand, they were long ignored by mathematicians. A tentative effort
by Alexandre-Theophile Vandermonde at the end of the eighteenth
century was short-lived,! and a preliminary study by the young Karl
Friedrich Gauss was no more successful. Only in the twentieth century
did mathematicians apply themselves seriously to the study of knots.
But until the mid-1980s, knot theory was regarded as just one of the
branches of topology: important, of course, but not very interesting to
anyone outside a small circle of specialists (particularly Germans and
Americans).
Today, all that has changed. Knots-or more accurately, mathemati-
cal theories of knots-concern biologists, chemists, and physicists.
Knots are trendy. The French "nouveaux philosophes" (not so new
anymore) and post modernists even talk about knots on television,
with their typical nerve and incompetence. The expressions "quantum
group" and "knot polynomial" are used indiscriminately by people
with little scientific expertise. Why the interest? Is it a passing fancy or
the provocative beginning of a theory as important as relativity or
quantum physics?
vii
viii PREFACE
This book addresses this question, at least to some extent, but its
aim is certainly not to provide peremptory answers to global inquiries.
Rather, it presents specific information about a subject that is difficult
to grasp and that, moreover, crops up in many guises, often imbued
with mystery and sometimes with striking and unexpected beauty.
This book is intended for three groups of readers: those with a solid
scientific background, young people who like mathematics, and oth-
ers, more numerous, who feel they have no aptitude for math as a re-
sult of their experience in school but whose natural curiosity remains
intact. This last group of readers suffers from memories of daunting
and useless "algebraic expressions," tautological arguments concerning
abstractions of dubious interest, and lifeless definitions of geometric
entities. But mathematics was a vibrant field of inquiry before lacklus-
ter teaching reduced it to pseudoscientific namby-pamby. And the
story of its development, with its sudden brainstorms, dazzling ad-
vances, and dramatic failures, is as emotionally rich as the history of
painting or poetry.
The hitch is that understanding this history, when it is not reduced
to simple anecdotes, usually calls for mathematical sophistication. But
it so happens that the mathematical theory of knots-the subject of
this book-is an exception to the rule. It doesn't necessarily take a
graduate of an elite math department to understand it. More spe-
cifically, the reader will see that the only mathematics in this book are
simple calculations with polynomials and transformations of little di-
agrams like these:
PREFACE
;x
Readers will also have to draw on their intuition of space or, failing
that, fiddle with strings and make actual knots.
My desire to avoid overly abstract and technically difficult mathe-
matics led me to leave out completely the most classical tool of the
theory of knots (and the most efficient at the early stages), the so-
called fundamental group. The first successes with the theory-those
of the mathematicians of the German school (N. G. Van Kampen, H.
Seifert, M. Dehn), the Dane J. Nielsen, and the American J. W. H. Al-
exander-were based on the judicious use of this tool. Their work will
barely be mentioned here.
Given the diversity of the topics tackled in this book, I have not
tried to provide a systematic and unified exposition of the theory of
knots; on the contrary, various topics are scattered throughout the
chapters, which are almost entirely independent of each other. For
each topic, the starting point will be an original idea, as a rule simple,
profound, and unexpected, the work of a particular researcher. We will
then follow the path of his thinking and that of his followers, in an at-
tempt to understand the major implications of the topic for contem-
porary science, without going into technical details. Accordingly, the
chapters are ordered more or less chronologically. But I have striven to
minimize cross-references (even if it means repeating certain pas-
sages), so that the chapters can be read in whatever order the reader
chooses.
Before I review the topics taken up in each chapter, it is worth men-
tioning that prior to becoming the object of a theory, knots were asso-
ciated with a variety of useful activities. Of course, those activities are
not the subject of this book, but talking a little about their practical
charms will make it easier to glimpse the beauty of the theory.
x PREFACE
Since Antiquity, the development of knot making was motivated by
practical needs, especially those of sailors and builders. For each spe-
cific task, sailors invented an appropriate knot, and the best knots sur-
vived, passing from generation to generation (see Adams, 1994). To tie
a rope to a rigid pole (a mooring or a mast), one uses the clover hitch
knot (see Figure P.la), the rolling hitch knot (b), or the camel knot (c);
to tie two ropes together, the square knot (d) or the fisherman's knot
(f) (when they are the same size) or the sheet bend (e) (when one rope
is thicker than the other). And there are many other knots adapted to
these special tasks (see Figure P.2). Sailors use knots not only to moor
boats, rig sails, and hoist loads, but also to make objects as varied as
the regrettably famous "cat 0' nine tails" and straw mats woven in
Turk's head knots (Figure P.2b).
In the Age of Enlightenment (in England even earlier), oral trans-
mission of maritime knot making was supplanted by specialized
books about knots. One of the first authors in this genre was the Eng-
lishman John Smith, much better known for his romantic adventures
with the beautiful Indian princess Pocahontas. At the same time, the
terminology associated with knots became codified; it was even the
subject of a detailed article in Diderot's and d'Alembert's Enclopedie.
Sailors were not the only inventors of knots. The fisherman's hook
knot (Figure P.2f), the alpinist's chair knot (d), the engineer's constric-
tor knot (c), and the knitter's rice stitch (e) are only a few examples
among many. The classic reference for knot making is Ashley's famous
Book of Knots (1944). A few knots in particular derive from one of the
greatest technological inventions of the Middle Ages: the pulley (Fig-
ure P.3a), together with the compound pulley (b and c). This work-
saving device, a sort of Archimedes' lever with ropes, unites two major
PREFACE xi
Figure P.l . Some sailors' knots.
O ~
(b)
(d) (e) (f)
Figure P.2. Other knots.
xii PREFACE
(a)
Figure P.3. Pullies and a hoist.
inventions of Antiquity: the wheel and the rope. It is used to pull or to
lift all kinds of loads, usually also attached with the help of suitable
knots. Thanks to knots, the rope became the universal technological
tool of the age.
The technology for producing ropes (and cables) themselves-
braiding-became very important. Fibers (once made of plants such
as hemp, but synthetic in our times) had to be twisted into threads
that were then braided into thicker strands, called lines, which in turn
were braided in a specific way (generally involving three lines) to make
a rope (see Figure P.4). The procedure for making cables is more com-
PREFACE
Strand
(l ine)
Figure P.4. Anatomy of a rope.
XIII
Fibers
plex and involves four (or more) levels of cords, lines, and braided
ropes. For the mathematician, the technology of braiding is the model
for a basic idea in topology (as well as in mechanics)-the braid--
which we will discuss in detail in Chapter 2.
Utilitarian and technological considerations aside, knots also have
an aesthetic, mysterious, and magical aspect. As far as I know, it is pre-
cisely this feature of knots that is responsible for their first traces in
our civilization. I have in mind the remarkable representations of
knots on the megaliths and burial stones engraved by Neolithic peo-
ples, in particular the Celts, during the fourth century B.C. Actually,
these are chains of knots connected to one another (mathematicians
call them links), as shown schematically in Figure P.5. We do not know
the mystical and religious meaning of the links represented on
menhirs (upright monuments also known as standing stones), but the
geometric technology (based on regular figures) used to create these
bewitching designs has been decoded by mathematicians (see Mercat,
1996).
XIV PREFACE
Figure P.S. Links on a megalith.
Neolithic peoples were not alone in using links to decorate their ob-
jects of worship. Links are also found in the Middle Ages, in illumi-
nated manuscripts, in the architecture of certain Eastern civilizations
(friezes and other ornaments of the famous Alhambra palace in Spain
are examples), and in the decorative elements framing icons in ortho-
dox churches in northern Russia.
PREFACE xv
To end this overview of knots on a lighter note, think of the essen-
tial role they play in the magician's arsenal: knots that aren't, ropes
that come undone instead of strangling the sexy magician's assistant,
and so on. Some of these tricks (which amateur magicians can do) are
described, from a mathematical vantage point, elsewhere (Prasolov
and Sossinsky, 1997; Walker, 1985).
Let us move on to a summary of this book, to give a brief idea of what
is to come and to allow those who don't intend to read the book from
beginning to end to choose which chapters they wish to take in.
2
(Re-
member that the chapters are relatively independent.)
The first chapter has to do with the beginnings of the mathematical
theory of knots, which was not the work of mathematicians-what a
shame for them!-but that of physicists, more precisely, William
Thomson (alias Lord Kelvin). The starting point (dating from around
1860) was Thomson's idea of using knots as models for the atom,
models he dubbed "vortex atoms:' To study the theory of matter from
this point of view, he had to begin with knots. Fortunately for the self-
esteem of mathematicians, Kelvin's theory ran aground and was soon
forgotten, but not without leaving to posterity a series of problems
(the Tait conjectures), which physicists were unable to solve at the time
but mathematicians took care of a century later. The chapter not only
deals with this spectacular failure of a beautiful physical theory, it also
reviews various aspects of knot theory: Tail's tables of alternating
knots, the superb wild knots, and Antoine's necklace. This last object
provides us with an opportunity to talk about ... blind geometers.
The chapter ends with a brief discussion of the reasons for the failure
of Thomson's theory.
XVI PREFACE
The second chapter deals with the fundamental connection be-
tween knots and braids discovered by the American J. W. H. Alexander
a half-century after Kelvin's abortive start. The mathematical theory of
braids, which was formulated about the same time by the young Ger-
man researcher Emil Artin, is more algebraic (and consequently sim-
pler and more efficient) than knot theory. The connection in question
(a geometric construction of childlike simplicity: the so-called closure
of braids) enables one to obtain all knots from braids-Alexander's
result. And because Artin rapidly established the classification of
braids, it was natural to try to deduce the classification of knots from
it. Efforts in this direction were unsuccessful, but they gave nice re-
sults, among which are the algorithms and software recently devised
by French researchers.
In Chapter 3, I present a clever but simple geometric construction
by the German mathematician Kurt Reidemeister, which reduces the
study of knots in space to their planar projections (called knot dia-
grams). This gives us a chance to talk a little about catastrophe theory,
encoding of knots, and working with knots on the computer. We will
see that an algorithm invented by Reidemeister's compatriot Wolfgang
Haken to determine whether a given knot can be untied does indeed
exist, though it is very complex. That is because untying a knot often
means first making it more complicated (alas, also true in real life).
Finally, the functioning of an unknotting algorithm (which is fairly
simple but has the disadvantage of futility when it comes to trying to
unknot non-unknottable knots) will be explained: there, too, the
modern computer does a better job of unknotting than we poor Homo
sapiens.
Chapter 4 reviews the arithmetic of knots, whose principal theorem
(the existence and uniqueness of prime knots) was demonstrated in
PREFACE xvii
1949 by the German Horst Schubert. The curious resemblance be-
tween knots equipped with the composition operation (placing knots
end to end) and positive integers (with the ordinary product opera-
tion) excited all sorts of hopes: Could knots turn out to be no more
than a geometric coding of numbers? Could the classification of knots
be just a plain enumeration? In Chapter 4 I explain why such hopes
were unfulfilled.
Chapter 5 brings us to an invention that seems trite at first. It is due
to the Anglo-American John Conway, one of the most original mathe-
maticians of the twentieth century. As in Chapter 3, we will be dealing
with small geometric operations carried out on knot diagrams. Con-
trary to Reidemeister moves, Conway operations can change not only
the appearance but also the type of the knot; they can even transform
knots into links. They make it possible to define and to calculate, in an
elementary way, the so-called Alexander-Conway polynomiaP of a knot
(or link). These calculations provide a very easy and fairly efficient
way to show that two knots are not of the same type, and in particular
that some knots cannot be unknotted. But this method is probably not
what the reader of this chapter will find most interesting: a biological
digression explains how topoisomerases (recently discovered special-
ized enzymes) actually carry out Conway operations at the molecular
level.
Chapter 6 presents the most famous of the knot invariants, the
Jones polynomial, which gave new life to the theory fifteen years ago.
In particular, it allowed several researchers to establish the first serious
connections between this theory and physics. Oddly, it is the physical
interpretation
4
of the Jones polynomial that gives a very simple de-
scription of the Jones invariants, whose original definition was far
from elementary. This description is based on a tool-the Kauffman
xviii PREFACE
bracket-that is very simple but that plays no less fundamental a role
in modern theoretical physics. This chapter contains several digres-
sions. In one of them, readers will learn that the main ingredient in
the Kauffman bracket was already known in the Neolithic age by the
Celtic artists mentioned earlier.
Chapter 7 is devoted to the last great invention of knot theory,
Vassiliev invariants. Here, too, the original definition, which drew on
catastrophe theory and spectral sequences,5 was very sophisticated,
but an elementary description is proposed. Instead of complicated
mathematical formulas, readers will find abbreviated calculations in-
volving little diagrams, along with a digression on the sociological ap-
proach to mathematics.
The eighth and final chapter discusses connections between knot
theory and physics. Contrary to what I tried to do in the other chap-
ters, here I could only sketch out the most rudimentary explanations
of what is going on in this area. I had to use some new technical terms
from mathematical physics without being able to explain them prop-
erly. But I am convinced that even readers closer to the humanities
than to the sciences will succeed in getting through this chapter. Even
if they cannot grasp the precise meaning of the terms and equations,
they can focus on the gist of the discussion, on the role of coinci-
dences, and on the dramatic and emotional side of contemporary re-
search.
The brilliant beginnings of knot theory, over 130 years ago, were
marked by a ringing failure-as a physical theory of matter-but the
concepts were revived thanks to the repeated efforts of mathemati-
PREFACE XIX
cians, whose only motivation was intellectual curiosity. Progress re-
quired new, concrete ideas. And the ideas came, springing from the
imagination of the best researchers, often sparking exaggerated hopes.
But every failure made it easier to grasp the remaining problems, mak-
ing the final goal ever more attractive. Today we are in a situation sim-
ilar to that of 1860: some researchers think, as William Thomson did,
that knots playa key role in the basic theory of the structure of matter.
But that is not to say that we are back at the beginning: the spiral of
knowledge has made a full loop, and we find ourselves at a higher
level.
The theory of knots remains just as mysterious and vibrant as ever.
Its major problems are still unsolved: knots continue to elude efforts
to classify them effectively, and still no one knows whether they pos-
sess a complete system of invariants that would be easy to calculate.
Finally, the basic role knots are supposed to play in physics has not yet
been specified in a convincing way.
KNOTS
I
ATOMS AND KNOTS
(Lord Kelvin ·1860)
In 1860, the English physicist William Thomson (better known to-
day by the name of Lord Kelvin, but at the time not yet graced with a
noble title) was pondering the fundamental problems linked to the
structure of matter. His peers were divided into two enemy camps:
those who supported the so-called corpuscular theory, according to
which matter is constituted of atoms, rigid little bodies that occupy a
precise position in space, and those who felt matter to be a superposi-
tion of waves dispersed in space-time. Each of these theories provides
convincing explanations for certain phenomena but is inadequate for
others. Thomson was looking for a way to combine them.
And he found one. According to Thomson, matter is indeed consti-
tuted of atoms. These "vortex atoms," however, are not pointlike ob-
jects but little knots (see Thomson, 1867). Thus an atom is like a wave
that, instead of dispersing in all directions, propagates as a narrow
beam bending sharply back on itself-like a snake biting its tail. But
this snake could wriggle in a fairly complicated way before biting itself,
thus forming a knot (Figure 1.1). The type of knot would then deter-
mine the physicochemical properties of the atom. According to this
1
2 KNOTS
view, molecules are constituted of several intertwined vortex atoms,
that is, they are modeled on what mathematicians call links: a set of
curves in space that can knot up individually as well as with each
other.
This theory will no doubt seem rather fanciful to the reader accus-
tomed to Niels Bohr's planetary model of the atom taught in school.
But we are in 1860, the future Nobel laureate will not be born until 25
years later, and the scientific community is taking Thomson's revolu-
tionary idea seriously. The greatest physicist of the period, James Clerk
Maxwell, whose famous equations formed the basis of wave theory,
hesitated at first, then warmed to the idea. He insisted that Thomson's
theory explained the experimental data accumulated by researchers
better than any other.
To develop his theory, Thomson needed first of all to see which dif-
ferent types of knots are possible; in other words, he had to classify
knots. It would then have been possible to classify atoms by associat-
ing each type of knot with a specific atom. For example, the three
Figure 1.1. Model of an atom?
ATOMS AND KNOTS 3
~ ~ ~
Figure 1.2. Three knots: the trefoil, the figure eight,
and the unknot.
knots represented in Figure 1.2, the trefoil, the figure eight, and the
trivial knot or unknot, could be models of carbon, oxygen, and hydro-
gen, respectively.
So in the beginning the problem was mathematical (rather than
physicochemical): the problem of classifying knots. And it was a Scot-
tish physicist and mathematician, a friend of Thomson, Peter Guthrie
Tait, who set out to solve it.
Tait, Kirkman, and the First Tables of Knots
According to Tait, a knot, being a closed curve in space, could be
represented by a planar curve obtained by projecting it perpendicu-
larly on the horizontal plane. This projection could have crossings
(Figure 1.3), where the projection of one part of the curve crossed an-
other; the planar representation shows the position in space of the two
strands that cross each other by interrupting the line that represents
the lower strand at the crossing. I have already used this natural way of
drawing knots (Figure 1.2) and will continue to use it.
Posing the question of how knots should be classified requires
4 KNOTS
Figure 1.3. Planar projection of a knot.
specifying which knots belong to the same class. It requires, in other
words, precisely defining the equivalence of knots. But we will leave this
definition (the ambient isotopy of knots) for later, limiting ourselves
here to an intuitive description. Imagine that the curve defining the
knot is a fine thread, flexible and elastic, that can be twisted and
moved in a continuous way in space (cutting and gluing back is not al-
lowed). All possible positions will thus be those of the same knot.
Changing the position of the curve that defines a knot in space by
moving it in a continuous way (without ever cutting or retying it) al-
ways results in the same knot by definition, but its planar representa-
tion may become unrecognizable. In particular, the number of cross-
ings may change. Nevertheless, the natural approach to classifying
ATOMS AND KNOTS 5
Figure 1.4. Two representations of the same knot.
knots in space consists first of making a list of all the planar curves
with 1,2,3,4,5, ... crossings, then eliminating the duplications from
the list, that is, the curves that represent the same knot in space (Fig-
ure 1.4).
Of course, for the task to be doable within a human lifetime, the
maximum number of crossings of the knots considered has to be lim-
ited. Peter Tait stopped at 10. Tait had an initial stroke of luck: he
learned that an amateur mathematician, the Reverend Thomas Kirk-
man, had already classified planar curves with minimal crossings, and
all that remained was to eliminate the duplications systematically. But
that is not so easy to do. Indeed, for each crossing of a planar curve,
there are two ways to decide which strand in the crossing should be
uppermost. For a curve with 10 crossings, for example, a priori there
are 2
10
, or 1,024, possibilities for making a knot. Tait decided to list
only alternating knots. that is, those in which overpasses and under-
passes alternate along the curve (Figure 1.5). In this way, exactly two
6 KNOTS
Figure 1.5. An alternating knot (a)
and a nonalternating knot (b).
alternating knots corresponded to each planar curve, substantially fa-
cilitating Tait's task. Which is not to say that it became simple: he de-
voted the rest of his life to it.
Nonalternating knots (one with 10 or fewer crossings) were classi-
fied in 1899 by C. N. Little, after six years of work. Little managed to
avoid the systematic run-through of the 2
10
uncrossing possibilities
(for each knot) mentioned above. Unfortunately for Thomson, Kirk-
man, Little, and Tait, by the time Little and Tait finished their work, al-
most no one was interested in knot tables for reasons that will be ex-
plained at the end of this chapter.
Still, at the century's close, most of the work on classifying knots
(with 10 or fewer crossings) had been done, and tables of knots ap-
peared. Figure 1.6 shows an example, a table of (prime) knots with 7
or fewer crossings. The exact meaning of the expression "prime knot,"
which is analogous to "prime number" in the sense that it cannot be
factored, is explained in Chapter 4, which deals with the arithmetic of
knots. But before continuing this account of Kelvin's and Tait's work,
it is worth making a few points about classifying knots.
ATOMS AND KNOTS 7
% ~ % % ~ ~ ~ ~
0,
~ ,
Figure 1.6. Table of knots with seven or fewer crossings.
8 KNOTS
A Mathematical View of Knot Classification
Let us pose the problem in precise terms, rigorous enough to satisfy a
mathematician (readers disinclined to scientific rigor can skip this
part after glancing at the figures). First of all, the very concept of a
knot needs to be defined. We define a knot, or more precisely a repre-
sentation of a knot, as a closed polygonal curve in space (Figure 1.7a).
A knot is just a class of equivalent representations of knots, equiva-
lence being the relation of ambient isotopy, defined as follows: An ele-
mentary isotopy is achieved either by adding a triangle (as we add ABC
in Figure 1. 7b) to a segment (AB) of the polygonal curve, then replac-
ing this segment by the two other sides (AC and CB), or by doing the
opposite. Of course, the triangle must have no points in common with
the polygonal curve other than its sides. An ambient isotopy is just a
finite sequence of elementary isotopies (Figure 1.7c).
Clearly this definition corresponds to our sense of a knot as the
abstraction of a string whose ends are stuck together, and ambient
isotopy allows us to twist and move the knot in space as we do with a
real string (without tearing it). From the aesthetic point of view, it is
perhaps not satisfying to think of strings that have angles everywhere,
but it is the price to pay for defining a knot in a manner both elemen-
tary and rigorous. I
Representing a knot as a polygonal curve has a motive other than
the ability to attach triangles to it (which presupposes that the "curve"
is made of segments); in fact, it is also a necessary condition for avoid-
ing "local pathologies." It is required to avoid the so-called wild knots,
which are not topologically equivalent to polygonal curves (or to a
smooth curve). Wild knots are the result of a process of infinite knot-
ATOMS AND KNOTS
c
Figure 1.7. A knot drawn as a closed polygonal curve in space (a) and
two equivalent representations (b and c) of the same knot (isotopy).
Figure 1.8. Wild knots.
9
ting: the twists of the curve get smaller and smaller and converge to-
ward a limit point known as a wild point of the curve (Figure 1.8).
Rigorously defining a knot (as a polygonal curve or a smooth curve)
makes it possible to avoid these little horrors and simplifies the theory.
Before we continue our preliminary investigation of "tame" knots,
10 KNOTS
~ ~ % % % ~ % % % % ~ % %
here are a few remarks on their "wild" kin (with a few drawings in-
cluded).
Digression: Wild Knots, Spatial Intuition, and Blindness
The examples of wild knots shown up to now possess a single isolated
pathological point, toward which a succession of smaller and smaller
knots converge. Wild knots with several points of the same type can
easily be constructed. But one can go further: Figure 1.9 shows a wild
knot that has an infinite (even uncountable, for those who know the
expression) set of pathological points.
This set of wild points is in fact the famous Cantor continuum, the
set of points in the segment [0, 1] that remain after one successively
eliminates the central subinterval (1/3,2/3), then the (smaller) central
subintervals 0/9,2/9) and (7/9,8/9) of the two remaining segments,
o
.1 .l.
9 9
.1
3
.l.
3
L .a
9 9
-=========-- -==- -II:===-- c::::Jt •
Figure 1.9. A wild knot converging to Cantor's continuum.
ATOMS AND KNOTS 11
then the four (tiny) central subintervals (1/27,2/27), (7/27, 8127), (191
27,20/27), and (25/27,26/27) of the four remaining segments, and so
on to infinity.
A much more beguiling wild knot can be obtained by making a
curve pass through a set even more complicated than Cantor's contin-
uum, for example, Antoine's necklace. No, I do not mean a gift of the
Roman general
2
to Cleopatra, but a geometric construction devised by
the French mathematician Louis Antoine. Let us try to describe this
jewel of the mathematical imagination, shown in Figure 1.10.
Begin with a solid torus, T
1
, in the shape of a doughnut and place
inside it four thinner solid tori linked two by two to make a chain, T
2
,
of four rings. Inside each of these four rings of the chain T
2
, construct
a chain similar to the preceding one; the set formed by these four little
chains (constituting the 16 tiny tori) is denoted T
J
• Inside each tiny to-
rus knot, take ... The process continues indefinitely, and the set ob-
tained as the infinite intersection of the sets Tj will be Antoine's neck-
lace:
A=TlnT2n ... nT"n ...
Antoine's necklace has some remarkable properties that I shall not
dwell on: it will simply aid us to construct a wild knot invented by
the Russian mathematician G. Va. Zuev and represented in the same
figure. The knot in question is shown (partially) in the form of the
curve that insinuates itself inside the large solid torus, then inside the
smaller tori, then into the tiny tori, and so on, dividing each time it
penetrates a torus knot to tend toward Antoine's necklace. One can
show (but the rigorous demonstration is rather tricky) that the curve
KNOTS 12
~ ; ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
Figure 1.10. A wild knot converging
to Antoine's necklace.
ultimately obtained is indeed a simple dosed curve, and the set of its
wild points is precisely Antoine's necklace.
Inventing monsters such as Antoine's necklace or Zuev's wild knot
takes a considerable ability to visualize three-dimensional space. So it
might surprise the reader to learn that these two mathematicians are
ATOMS AND KNOTS 13
blind. But actually, it isn't all that surprising, since almost all blind
mathematicians are (or were) geometers. The spatial intuition that
sighted people have is based on the image of the world that is pro-
jected onto their retinas; thus it is a two- (and not three-) dimensional
image that is analyzed in the brain of a sighted person. A blind per-
son's spatial intuition, on the other hand, is primarily the result of tac-
tile and operational experience. It is also deeper-in the literal as well
as the metaphorical sense.
As we leave this digression, note that recent biomathematical stud-
ies (based on work with children and adults who were born blind but
gained their sight afterward) have shown that the deepest mathemati-
cal structures, such as topological structures, are innate, whereas finer
structures, such as linear structures, are acquired. Thus, at first, the
blind person who regains his sight does not distinguish a square from
a circle: he sees only their topological equivalence. In contrast, he im-
mediately sees that the torus is not a sphere. As for us, our tendency to
consider what we see the "absolute truth" often makes us conceive the
world in a very flat and superficial way.
The Failure of Thomson's Theory
While European physicists were debating the merits of Thomson's
theory and Tait was filling in his tables of knots, another researcher-
unknown and working in an immense, underdeveloped country-
was, like Thomson and Tait, reflecting on the structure of matter. He,
too, was trying to establish tables of atoms but, little inclined to geo-
metric considerations, he based his tables on arithmetical relation-
ships among the various properties of chemical elements.
This scientist made an unexpected discovery: there are very simple,
14 KNOTS
until then unnoticed, relationships among the chemical properties.
And he published what today is called the periodic table of elements. It
would take some time for this remarkable discovery to be recognized
in western Europe. Once that happened, my compatriot Mendeleev
buried the idea of atoms as knots. Thomson's theory had not done
much for chemistry, and it was quickly replaced by the arithmetical
theory of Mendeleev. And physicists, ashamed and embarrassed, for-
got about knots for nearly a century. It was the mathematicians who
would take up the subject once again.
BRAIDED KNOTS
(Alexander· 1923)
This chapter is devoted to a remarkable connection, discovered by
mathematicians, between two beautiful topological objects: braids and
knots. Mathematically speaking, what is a braid? Roughly, it is the for-
mal abstraction of what is meant by a braid in everyday language (a
braid in a young girl's hair, a plaited key chain, a braided dog's leash,
a classic twisted rope, and so on)-in other words, some strings tan-
gled in a certain way. More precisely, you can imagine a braid of n
strands as n threads attached "above" (to horizontally aligned nails)
and hanging "down;' crossing each other without ever going back up;
at the bottom, the same threads are also attached to nails, but not nec-
essarily in the same order (Figure 2.1).
The strands of a braid can be rearranged (without detaching the top
and bottom, and of course without tearing or reattaching them) to get
a braid that looks different but is equivalent to (or an isotope of) the
first braid (Figure 2.2). As with knots, we do not distinguish two isoto-
pic braids: we think of them as two representations of the same object
(from the formal mathematical point of view, this means that the ob-
15
Figure 2.1. Examples of braids.
Figure 2.2. Isotopes of a four-stranded braid.
BRAIDED KNOTS 17
ject in question is not a concrete braid but an equivalence class of
braids).
The basics of braid theory were developed by Emil Artin in the
1920s. The theory is a marvelous blend of geometry, algebra, and algo-
rithmic methods. It has a bunch of applications ranging from the tex-
tile industry to quantum mechanics by way of topics as varied as com-
plex analysis, the representation of functions of several variables as
the composition of functions of a lesser number of variables, and
combinatorics. But we will deal with this theory later, since our imme-
diate goal is to understand the connection between braids and knots.
Closure of Braids
A knot can be made from a braid by the operation of closure, which
means joining the upper ends of the strands to the lower ends (see
Figure 2.3a).
Will a knot always be obtained in this way? According to Figure
2.3b, not always: the closure of a braid may very well result in a link
of several components (that is, several curves, in contrast to a knot,
which by definition consists of only one curve). An attentive reader of
the previous chapter will recognize the trefoil knot in Figure 2.3a-
though perhaps not at once.
The following question arises immediately: Which knots can be ob-
tained in this way? The answer, found by J. W. H. Alexander in 1923,
explains the importance of braids in knot theory: they all can. Alexan-
der's theorem can be expressed as follows: Every knot can be repre-
sented as a closed braid. (Actually, Alexander showed that this assertion
18 KNOTS

is true for the more general case of links, of which knots are just an ex-
ample.)
Alexander probably hoped that his theorem would be a decisive step
forward in classifying knots. Indeed, as we will see later, braids are
much simpler objects than knots; the set of braids possesses a very
clear algebraic structure that enables one to classify them. Is it reason-
able, then, to try to use braids to classify knots? How did this idea de-
velop? We will see at the end of the chapter.
For now, let us return to Alexander's theorem: How can it be
proved? Given a knot, how does one find the braid whose closure
would be that knot? First of all, note that the desired braid can easily
be seen when the braid is rolled up in a coil, that is, when it always
turns in the same direction around a certain point (as the knot in Fig-
'--"
(a)
'--"
(b)
Figure 2.3. Closure of two braids.
BRAIDED KNOTS 19
Figure 2.4. Unrolling a coiled knot into a braid.
ure 2.4 turns around the center C). In this situation, all that is required
to find the braid is to cut the knot along a line extending outward
from the center and then to unroll it (Figure 2.4b).
But what if the knot is not coiled, for example, the knot shown in
Figure 2.Sa? (As readers of the first chapter will know, this knot is
called a figure eight knot.) In that case, just move the "fat" part (the
thicker line) of the knot (the one "going the wrong way") over the
point C on the other side of the curve. The resulting coiled knot (Fig-
ure 2.Sb) can then be unrolled into a braid as in the preceding exam-
ple (Figure 2.5c).
Actually, this elegant method (transforming any knot into a coiled
knot) is universal, and it allowed Alexander to prove his theorem. Its
weakness-and there is one-is its ineffectiveness from a practical
point of view; specifically, it is difficult to teach to a computer. An-
other method of braiding knots, more doable and easier to program,
20 KNOTS
(a) (b)
Figure 2.S. Coiling a figure eight knot and unrolling it into a braid.
was invented by the French mathematician Pierre Vogel. The reader
not inclined to algorithmic reasoning can blithely skip this description
and go on to the study (much simpler and more important) of the
group of braids.
Vogel's Braiding Algorithm
The braiding algorithm transforms any knot into a coiled knot. To de-
scribe it I must introduce some definitions related to the planar repre-
sentations of knots. Assume that a given knot is oriented; that is to say,
the direction of the curve (indicated by arrows) has been selected. The
planar representation of the knot defines a kind of geographic map in
BRAIDED KNOTS 21
Figure 2.6. Desingularized knots and Seifert circles.
the plane, the regions or countries being the areas bounded by parts of
the knot's curve. On this map, the border of each country consists of
several edges (oriented in accordance to the arrows) that join one
crossing of the knot to a neighboring crossing. Included among the
countries is the infinite region-the one located outside the curve.
Because the curve of the knot is oriented, the crossings are marked
with arrows; these make it possible to "desingularize" the knot N
unambiguously, in other words, to replace all the crossings of N by
their smoothings as shown in Figure 2.6a. Desingularization trans-
forms each knot into one or more oriented, closed curves (with-
out crossings), called Seifert circles, that represent the knot (Figures
2.6d, 2.6e). Two Seifert circles are nested if one of them is inside the
other and if the orientations of the two circles coincide. Note that
desingularizing a coiled knot always yields a nested system of Seifert
circles, and vice versa (Figure 2.6b).
On the other hand, when Seifert circles are not nested (as in Figure
2.7b), the change-of-infinityl operation nests the two circles (Figure
2.7c). In fact, this figure shows that while circles 1 and 2 are nested,
circle 3 does not encompass them; but inverting this last circle results
in the circle 3', which neatly encompasses circles 1 and 2. (In this case,
the change-of-infinity move resembles the operation carried out in
Figure 2.5.)
Let us now consider the planar map determined by the knot N. A
country in this map is said to be in turmoil if it has two edges that
belong to two different Seifert circles, labeled with arrows going in
the same direction around the region. For example, in the smoothing
of knot N in Figure 2.8, the country H is in turmoil, whereas regions
PI and P
2
are not: since the thick edges head in the same direction
around H and belong to two distinct Seifert circles, H is in turmoil; PI
is not, because its edges belong to a single Seifert circle; finally, P
2
is not in turmoil either, because its edges go around P
2
in opposite
directions.
An operation called perestroika can be applied to any country in
turmoil (Figure 2.9). Perestroika consists in replacing the two faulty
edges by two "tongues:' one of which passes over the other, forming
two new crossings. The result is to create a central country (not in tur-
moil) and several new countries, some of which (in this case, two)
may be swallowed up by bordering countries. I am sure that now the
reader understands the choice of the geopolitical term perestroika.
BRAIDED KNOTS 23
% % % % ~ ~ ~ ~ ~ ~ % ~ ~ % % ~ % ~ ~ % ~ %
(a) (b) (c)
Figure 2.7. Change of infinity.
N
Figure 2.8. Countries in and out of turmoil.
24 KNOTS
Figure 2.9. Perestroika of a country in turmoil.
Vogel's algorithm can now be presented in the form of a "program"
written in a sort of "pseudO-Pascal":
Do smoothing
While: There is a disjoint region
Do perestrOika
Do smoothing
End while
While: The Seifert circles are not nested
Do change infinity
End while
Stop
Most of this language was explained above, but the command Do
change i nfi n i ty requires clarification: it means taking one of the small-
BRAIDED KNOTS 25
est Seifert circles not nested with the others and sending a point inside
this circle to infinity.
First let us apply the Vogel algorithm to a very simple knot (the
unknot, in fact) to see how changes of infinity occur (Figure 2.10).
Following the first smoothing, there are no countries in turmoil, and
no nested Seifert circles. So we proceed to the command Do change
infinity, which must be carried out twice-(b) becomes (c) and (c) be-
comes (d)-to obtain a coiled knot (d), which can then be unrolled
into a braid as before (e).
Figure 2.11 shows how Vogel's algorithm coils a knot with five
crossings.
2
Following the initial smoothing, the loop (in the computa-
tional sense of the word) contains two perestroikas (Figures 2.11 band
2.11c); it is followed by a change-of-infinity move. The result (Figure
2.11d) is indeed a coiled knot, even if it does not look like one. To
make sure, I have redrawn it twice (Figures 2.l2b, 2.l2c). The reader
will have no difficulty recognizing the rolled-up knot from Figure 2.4a
(and so can admire the desired braid, which appeared earlier as Figure
2.4b).
00
C!5
o
o
(a) (b)
Figure 2.10. Vogel algorithm applied to the unknot.
U KNOn
~ ~ ~ ~ % ~ ~ ~ ~ % % % % % % ~ ~ % ~ ~
(a)
(b)
(c)
(d)
Figure 2.11. Vogel algorithm applied to knot 52.
BRAIDED KNOTS 27
(a) (b) (c)
Figure 2.12. Using the Vogel algorithm to unroll a knot.
Note that it is not at all obvious that the algorithm, which includes
two While loops that are a priori dangerous, will always terminate. Yet
it does, and very rapidly. Proving that the second loop always
nates is elementary. On the other hand, to prove that the same is true
for the first loop, Vogel had to use fairly sophisticated algebraic
ogy methods.
Of course, to transform our "program" into software for a real
puter, we have to know how to code knot representations in such a
way that the machine will be able to work with them. We will come
back to the coding of knots in Chapter 4.
The Braid Group
Let us return to the study of braids. First of all, we are going to define
an operation, the composition or product, on the set of braids with n
28 KNOTS
Figure 2.13. The product of two braids.
strands. This operation consists simply of placing the braids end to
end (by joining the upper part of the second braid to the lower part of
the first), as in Figure 2.13.
It turns out that the product of braids possesses several properties
that resemble the ordinary product of numbers. First, there is a braid
known as the unit braid (e), a braid that, like the number I, does not
change what it multiplies. This is the trivial braid, whose strands hang
vertically without crossing. Sure enough, appending a trivial braid to a
given braid amounts to extending its strands, which does not change
the class of the braid in any way.
Second, for each braid b there exists an inverse braid, b-
'
, whose
product with b gives the trivial braid: b . b-
'
= e (just as for each
number n, its product with the inverse number n-
I
= lin is equal to
one, n . n-
I
= 1). As can be seen in Figure 2.14, the inverse braid is the
BRAIDED KNOTS 29
braid obtained by taking the horizontal mirror image of the given
braid; indeed, each crossing cancels with its mirror image, such that all
the crossings gradually dissolve, two by two, beginning at the middle
of the product braid.
The third property common to braids and to numbers is the asso-
ciativity of the product operation: (a . b) . c = a . (b • c) is always
true. When a set is endowed with an operation that enjoys all three of
the properties just described, mathematicians call this set a group.
Thus, I have just shown that braids with n strands form a group. This
group will be denoted by Rn.
In contrast to numbers, however, the braid group Rn (for n > 2) is
not commutative: the product of two braids generally depends on the
order of the factors.
The product of braids makes it possible to replace the picture repre-
Figure 2.14. The product of a braid and its inverse.
30 KNOTS
~ ~ ~ ~ ~ ~ ~ ~ % % ~ % ~
2
•••
• • •

2 3 • n-1 n
1 1 I··· X -b._
1
Figure 2.1 S. Algebraic representation of a braid.
senting a braid by a word-the algebraic encoding of that braid. In-
deed, moving along a braid from top to bottom, we see that it is the
successive product of braids each with a single crossing (Figure 2.15);
we call these elementary braids and denote them by bl, b
I
, • •• , b
n
-
I
(for
braids with n strands).
So we have replaced braids-geometric objects-by words: their al-
gebraic codes. But recall that the geometric braids possess an equiva-
lence relation, namely, isotopy. What does that mean algebraically?
Artin had an answer to this question. He found a series of algebraic re-
lations between braid words that gave an adequate algebraic descrip-
tion of their isotopy. These relations are commutativity for distant
braids
i,j = 1,2, ... , n - 1
BRAIDED KNOTS 31
i i+l i+2 i+l i+2

. . .
·IT·· -.Q ..
Figure 2.16. Relations among the group of braids.
and Artin's relation (or the braid relation)
i = 1,2, ... , n - 2
Their geometric interpretation is shown in Figure 2.16. To an observer
with a little spatial imagination, it is immediately obvious that these
relations are valid for braids, that is, they do indeed correspond to
isotopies.
What is less obvious-and is one of Artin's key findings-is
that these two relations (if one adds the trivial relations,3 bjb
j
-
'
= e =
bj-'bj, also shown in Figure 2.16) suffice to replace the geometric ma-
nipulations related to isotopy by admissible algebraic calculations on
the braid words; each of the admissible calculations consists in replac-
ing a part of a word identical to one of the members of the relations
that appear in Figure 2.16 by the other member of that relation. Here
is an example of an admissible calculation in the group of braids with
four strands, B
4
:
32 KNOTS
(To make this formula easier to read, I have enclosed within parenthe-
ses the parts of the word that are replaced successively during the cal-
culation.)
More precisely, Artin's theorem affirms that
Two braids are isotopic if and only if the word representing one of
them can be transformed into the word representing the other by a
sequence of admissible calculations.
Artin's theorem is important because it reduces the geometric study of
braids to their algebraic study, which is generally more efficient. This
algebraic approach allowed Artin to classify braids. Put in other words,
it allowed him to find a comparison algorithm for them; for each pair
of braids, the algorithm says "no" if they are not isotopic and "yes" if
they are (as well as providing a set of admissible calculations that lead
from one to the other in the latter case).
Classifying Braids
I will not be describing any of the braid-comparison algorithms here,
neither Artin's (which he called by the lovely word combing), nor
the one recently discovered by the French mathematician Patrick
Dehornoy, which is much simpler and more efficient. But to convince
you of the efficiency of algebraic-algorithmic methods in geometry, I
chose (more or less at random) an example of a calculation carried
out by my little computer (which has in the recesses of its electronic
memory some software that drives Dehornoy's algorithm). This calcu-
BRAIDED KNOTS 33
lation, which takes place in the group B4 and uses the (more readable)
notation a, A, b, B, c, C for the basic braids bl> bl-I, ... , b
3
-I, respec-
tively, shows that a braid with four strands that looks rather compli-
cated is in fact the trivial braid.
ABBAAAAA[ Abbbbbbbbcba }AccBCaBBBBBBaaaaaaBB
= ABBAAAAA[ Aba }aaaaaaaBcbaBAccBCaBBBBBBaaaaaaBB
= ABBAAAA[Aba} BaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB
= ABBAAA[ Aba} BBBaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB
= ABBAA[Aba} BBBaaaaaaaBcbaBAcBCaBBBBBBaaaaaaBB
= ABBA[Aba}BBBBaaaaaaaBcbaBAcBCaBBBBBBaaaaaaBB
= ABB[Aba}BBBBBaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB
= [ABa} BBBBBBaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB
= [ABa} BBBBBBaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB
= b[ABBBBBBBBa}aaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB
= bbAAAAAA[Aba}aaaaaBcbaBAceBCaBBBBBBaaaaaaBB
= bbAAAAAAb [Aba }aaaaBcbaBAceBCaBBBBBBaaaaaaBB
= bbAAAAAAbb[Aba}aaaBcbaBAceBCaBBBBBBaaaaaaBB
= bbAAAAAAbbb[Aba}aaBcbaBAceBCaBBBBBBaaaaaaBB
= bbAAAAAAbbbb[ Aba }aBcbaBAceBCaBBBBBBaaaaaaBB
= bbAAAAAAbbbbb [Aba} BcbaBAceBCaBBBBBBaaaaaaBB
= bbAAAAAAbbbbbbAB[Bcb }aBcbaBAccBCaBBBBBBaaaaaaBB
= bbAAAAAAbbbbbbA[Bcb }CaBcbaBAceBCaBBBBBBaaaaaaBB
= bbAAAAAAbbbbbbAcbCC[ aA} ceBcbaBABBBBBBaaaaaaBB
= bbAAAAAAbbbbbbAcb [CCce} BCaBBBBBBaaaaaaBB
= bbAAAAAAbbbbbbA[cbBC}aBBBBBBaaaaaaBB
= bbAAAAAAbbbbbb[Aa}BBBBBBaaaaaaBB
34 KNOTS
= bbAAAAAA[bbbbbbBBBBBB JaaaaaaBB
= bb[AAAAAAaaaaaaJBB = [bbBBJ = e
For comparison, the reader can draw the given braid and try to un-
ravel it geometrically-although doing so may instill an inferiority
complex with regard to my notebook computer, which did the job in
less than a tenth of a second.
Can Braids Be Used to Classify Knots?
Alexander's theorem affirms that all knots are closed braids, and we
have just seen that braids can be classified. Can the classification of
knots be deduced from these two facts? Several mathematicians, and
not the least of them,4 have probably nourished this hope (I know
some who are still hoping). The attempt made for a rich history, brim-
ming with new developments, that began during the 1930s and per-
haps has not yet ended. But this chapter has gone on too long, and I
will end it by referring the amateur math lover of nice stories to an ar-
ticle by Dehornoy (1997).
3
PLANAR DIAGRAMS OF KNOTS
(Reidemeister· 1928)
During the 1920s, the German mathematician Kurt Reidemeister, fu-
ture author of the first book about the mathematics of knots, the fa-
mous Knottentheorie, began to study knots in depth. How could they
be classified? The problem of systematizing the possible positions of a
curve in space was to prove devilishly difficult.
The analytic approach (defining a knot using equations) did not
help, nor did the combinatorial approach (defining a knot as a closed
polygonal curve based on the coordinates of its successive vertices).
In both cases, the information did not enable one to see the knot or
to manipulate it. In practice, seeing a knot means drawing it, that
is, projecting it onto a suitably chosen plane to obtain what is called
a knot diagram. Manipulating the string that determines the posi-
tion of the knot causes its diagram to undergo continuous modi-
fications, which enables one to follow the evolution of its positions
in space. But can the process be inverted? Can the projection be con-
tinuously modified in such a way as to obtain all the possible posi-
tions of the string in space? That is the question Reidemeister asked
himself.
3S
36 KNOTS
Here is his answer: Just perform a finite number of operations on
the diagram, similar to those shown in Figure 3.1, while doing trivial
planar manipulations (that is, while continuously changing the dia-
gram of the knot in the plane without altering the number and rela-
tive disposition of the crossing points). The three operations shown in
the figure are now called Reidemeister moves; they are denoted by the
symbols QI' Q2, Q
3
.The moves that they allow are the following:
• QI: appearance (disappearance) of a little loop;
• Q2: appearance (disappearance) of twin crossings;
• Q
3
: passing a third strand over a crossing.
The following figure, which represents an unknotting process,
shows how Reidemeister moves are involved in representing the ma-
nipulation of a knot. The process begins with the uncrossing (disap-
pearance) of twin crossings at the top (shaded area of Figure 3.2a),
followed by the passage over a crossing (b), the disappearance of a pair
...
.
........ I ...........
. "":. :/"R.'\
f 1' .. " I '
t •• ",} ••••.•• •• )
••..•. Q, •.....
Q
...... 3 ••••••
/\.
.. ,)( ... .
, ,..,
\/
....................
Figure 3.1. Reidemeister moves.
PLANAR DIAGRAMS OF KNOTS 37
(e) (d)
Figure 3.2. Unknotting using Reidemeister moves.
of twin crossings (c), the disappearance of a small loop (d), and,
finally, the disappearance of a pair of twin crossings (e). The reader
will have noticed that between the actual Reidemeister moves, the dia-
gram of the knot undergoes trivial planar manipulations in prepara-
tion for these movements (these trivial manipulations change neither
the number nor the distribution of the crossings).
To understand where Reidemeister moves come from, and why they
38 KNOTS
suffice, we will have to spend a little time talking about knot pro-
jections.
Generic Projection and Catastrophic Projection
In introducing the diagrams of knots, I mentioned above that they are
projections in a "suitably chosen" plane. What does it mean to say
that? A mathematician would answer that the plane must be selected
so that the projection is generic, but this specification is utterly useless
if you do not know the term, which is in fact a basic mathematical
conceptI-intuitively clear but difficult to formulate in the general
case.
A generic projection is a projection without catastrophes, singulari-
ties, or degenerations that can be avoided, or gotten rid of by making
tiny changes to the projected object. Let us be specific about what all
these synonyms mean (catastrophe, singularity, degeneration) in the
case of a knot represented by a closed polygonal curve. By definition
(of a generic projection of a knot), we assume that
(1) two (or more) vertices cannot be projected to the same point;
(2) a vertex (or several vertices) cannot fallon the projection of an
edge or line segment (to which they do not belong);
(3) three (or more) points cannot be projected to the same point.
The existence of a generic projection for any knot is obvious and
can be demonstrated easily:2 the "forbidden catastrophes" (1), (2),
(3) must be removed by slightly displacing one of the knot's vertices.
Note that these three catastrophes differ from the crossing catastrophe,
which occurs when two points inside two distinct edges project onto a
PLANAR DIAGRAMS OF KNOTS 39
single point: this is inevitable, inasmuch as any little change made to
the knot moves the position of the crossing on the diagram somewhat,
but cannot totally eliminate it.
Catastrophes (2) and (3) are, in some sense, the catastrophes that
occur most often: among all the projections, they do represent excep-
tional events, but they are, so to speak, the most common among ex-
ceptional events. There are, of course, rarer catastrophes-also forbid-
den because they are special cases of catastrophes (1), (2), and (3). For
example, 17 points, including 5 vertices, can project to a single vertex;
7 edges (perpendicular to the plane of projection) can degenerate to a
single point; and so on.
Reidemeister moves correspond precisely to the most common for-
bidden catastrophes, as shown in Figure 3.3. Thus, in the upper part
of Figure 3.3a, we see a type (2) catastrophe in which the vertex A of
edge AB (which moves) projects momentarily (the middle frame) to a
point inside the projection of edge Be. This position corresponds to
the disappearance of a little loop on the projection (the knot dia-
gram); that is, it is the move QI. To the right of Figure 3.3a, this move
is symbolically represented in keeping with the style of Figure 3.1. The
shrewd reader, examining the next panels (b and c) of Figure 3.3, will
see how a type (2) catastrophe can give rise to move Q2, and a type (3)
catastrophe to move Q3.
The Sufficiency of Reidemeister Moves
Now that we know the origin of Reidemeister moves, we are in a posi-
tion to discuss their principal application (often called Reidemeister's
theorem or lemma):
(aj
C
! 1: c: ;: C /; i 0 ', •••••
I' I I I I I <. 'I ......
: i ! 1 ! iii ; 1 1 1 /' A ' ....
: T r: :., ": :: ::
:: :: :: :::: r: .. t
r: :;. t i : y T + :... ' ...... '
\W<1 \tl?1 @

(
b) ! ' ..... .
" • I I I I .' I • I I I ....... ..
1 iii i !: 1 1 ! !: ; /#
o 0 o. 0 0 Y . 0 0 • to. •
.: T:: ::::!::::: I
r : !::,.:! :! t :: :: \ ·
:! i:"::: : t : :: ! + .......... .,#

1'" I i
: to "'; : : If :, -)'t"':
11 i i:! 1 : i i ! i/ . '<' .. )t
:. : : :.:: '. :::: :
.: I. • I. I. I:' •••••• • ..' I ........
:: !: :: : :: :!: :::::: : ::::,' '.
T: :: :: : r: ::: ::: ... :: : ::::. ,
(c) :! :: :: : :: !:: :::!:! : :! i:', ..
l::: : it:: ::: l:"::: ! :: l-t ...... .,#

i :-i,hl .( .. /. ....
: .... : :: : ....... 1 .... : I ':: : .... : ......... : : ': .' "
:/; :\; i, i i : ,,'
il' .' I I • • .. #
. .. .......
Figure 3.3. Catastrophes and Reidemeister moves.
PLANAR DIAGRAMS OF KNOTS 41
If one knot can be transformed into another knot by continuous
manipulation in space, the same result can be obtained by a manip-
ulation whose projection consists uniquely of Reidemeister moves
and trivial manipulations of the diagram in the plane.
This means that it is possible to examine all the spatial manipula-
tions of knots by trivially manipulating their diagrams in the plane
and applying Reidemeister moves from time to time. In this way,
Reidemeister reduced the three-dimensional and rather abstract prob-
lem of knot equivalence to a two-dimensional problem, and a more
concrete one, to boot.
Before discussing what the Reidemeister theorem contributes to the
study of knots (in particular, to their classification), let me say a few
words about its proof. Unfortunately (or fortunately, depending on
your point of view), the proofs I know are not simple enough to be in-
cluded in this book. For readers more at home with mathematics, I
will simply say that all it takes is to analyze in detail a single basic tri-
angular movement (see the beginning of Chapter 1) and to perturb it
so that it is in a "general position"; in that case, only catastrophes (2)
and (3) can occur and, as we have seen, these correspond exactly to
Reidemeister moves.
Does the Reidemeister Theorem Classify Knots?
Let us put ourselves in Reidemester's shoes and imagine that we are
elated at having demonstrated his theorem. Caught up in the excite-
ment, we set out to classify knots-in other words, to devise an
algorithm that will determine whether two knots (as sketched) are
equivalent.
42 KNOTS
Take the first knot and compare it with the second. If the number of
crossings and their relative positions are the same, the knots are equiv-
alent, and we have succeeded. If not, (randomly) apply a Reidemeister
move to the first knot and again compare the result with the second. If
they are identical, we have won again. If not, we will have to remember
the changes to the first knot and apply to it another Reidemester
move, then compare again, and so on. If several moves applied to the
first knot don't work, we go back to the modified knot, apply yet an-
other move to it, compare, and so on. If the two knots are equivalent,
sooner or later a set of Reidemeister moves leading from the first knot
to the second will be specified.
The algorithm described above is easy to implement on a computer,
even a very little one. Thus, the humble "notebook" on which I am
writing this text has, among other things, software that can untie
knots (by comparing them with the unknot, as in the preceding para-
graph).3 Does that solve the problem of classifying knots? Of course
not, and for a good reason: applied to two nonequivalent knots, the al-
gorithm described above never terminates; it continues on and on
without ever stopping. And the user faces a dilemma: if the software
does not give an answer after, say, a full day's work, is it because the
knot is nontrivial or because the computer needs more time to find
the sequence of operations that will lead to an unraveling?
Must we kiss Reidemeister's idea good-bye? Not yet, because there is
another possibility that could salvage it. I am sure Reidemeister must
have had this thought at one time or another, and perhaps it evoked in
him that extraordinary feeling that researchers sometimes feel-the
feeling of being on the verge of finding out, of understanding. (I note
parenthetically that even for the best of us, this feeling is often fol-
lowed by despair when the idea turns out to be inadequate or illusory.)
PLANAR DIAGRAMS OF KNOTS 43
This other possibility is very simple: moves Q\) Q2, Q
3
may dimin-
ish the number of crossings (disappearance of a small loop or of twin
crossings), increase the number (appearance of a loop or twin cross-
ings), or, finally, not change the number (Q
3
simply passes a strand
over a crossing). To create an algorithm for unraveling (that is, simpli-
fying the knot), we just need the move Q
3
and allow the moves Q, and
Q
2
only when they decrease the number of crossings. If the process is
restricted this way, the number of crossings diminishes, and the algo-
rithm (thus perfected) will always terminate: either there are no more
crossings (and the knot was trivial), or none of the permissible moves
is applicable
4
(and the knot is nontrivial).
Unfortunately, this argument (however convincing) is erroneous.
In reality one cannot always unravel a knot by simplifying it (by di-
minishing the number of crossings) at each step of the unraveling:
sometimes, it is necessary at first to complicate things to make them
simpler. An example of a trivial, nonsimplifiable knot (that can be un-
raveled only by first increasing the number of crossings) appears in
Figure 3.4.
The hope of obtaining a simple and effective method of classifying
knots using Reidemeister's theorem was too optimistic. The world is
Figure 3.4. A trivial, nonsimplifiable knot.
44 KNOTS
Figure 3.5. Wolfgang Haken's "Gordian knot."
made that way: to unravel a knotted situation, often one should begin
by tangling it even further, only to unravel it better.
Since Reidemeister's theorem was discarded, devising trivial knots
that are hard to unravel has admittedly become an important exercise
in research on unknotting algorithms. A particularly barbaric example
of such a knot (very hard to unravel-try it!) is shown in Figure 3.5,
for which we can thank Wolfgang Haken. Moreover, it was Haken who
finally solved the problem of un knotting (Haken, 1961), but his algo-
PLANAR DIAGRAMS OF KNOTS 45
rithm (too complicated to be put on a computer) is based on a very
different class of ideas.
What Remains of Reidemeister's Theorem?
We should not take the collapse of naIve expectations to mean that the
application of Reidemeister's theorem is limited to an algorithm that
does not work, or that this theorem is simply yet another example of a
spectacular failure and dashed hopes.
The theorem occupied a central place in subsequent developments,
especially in the study of knot invariants by Vaughan Jones. Louis
Kauffman, and their followers (Chapter 6). In order that a knot dia-
gram's function proposed in the guise of a new invariant be indeed
invariant, the function must never change during the process of ma-
nipulating knots. According to Reidemeister's theorem, proving this
requires only verification that the function does not change when the
moves QI, Q2. or Q
3
are performed; well, these particular moves are
very simple, so verification is in general very easy.
But that is not all. The failure of the unknotting algorithm de-
scribed above is relative. Of course, from a theoretical point of view. it
is not a real unknotting algorithm (since it can "continue indefinitely"
without giving an answer). From a practical viewpoint, however, this
algorithm and its recent modifications may be used as a relatively ef-
ficient tool that often enables a (strong enough) computer to unravel
knots that cannot be undone "by hand"5 ... unless we use the algo-
rithm that Alexander of Macedonia applied with so much success to
the Gordian knot: cut it!
THE ARITHMETIC OF KNOTS
(Schubert· 1949)
Arithmetic-of knots? Absolutely. Why should the natural numbers-
1, 2, 3, 4, 5, ... -be the only objects that can be multiplied and
decomposed into prime factors? The same features describe other
mathematical entities, knots in particular. Moreover, knots possess
an arithmetic very similar to that of the natural numbers, with com-
mutative multiplication (called composition) and a theorem asserting
the uniqueness of decomposition into "prime knots:' The demonstra-
tion of this fundamental principle, foreshadowed by many researchers,
proved difficult (as in fact did the corresponding principle for num-
bers) and was not achieved until 1949 (by the German mathematician
Horst Schubert).
Thus, just as every whole number (say 84) factors out in a unique
way (84 = 2 X 2 X 3 X 7), every knot (for example, the one drawn at
the left in Figure 4.1) is the (unique) composition of prime knots, as
can be seen at the right in the same figure (it is the composition of two
trefoil knots and one Turk's head knot). The reader will have grasped
that the "composition" of knots consists more or less in setting them
end to end (as is done for braids, anyone who has read Chapter 2 will
say). To explain this operation, we are going to put knots· in boxes: in
46
THE ARITHMETIC OF KNOTS 47
-
~ c Y
Figure 4.1. Decomposing a knot into prime factors.
(a) (b)
Figure 4.2. Composing boxed knots.
this way, each knot appears as a knotted string inside a cube, with the
two ends stuck to two opposite faces of the cube (Figure 4.2a). (I will
leave to the reader already corrupted by the study of mathematics the
task of transforming this intuitive description into a rigorous mathe-
matical definition.) Making a closed-curve knot from a boxed knot is
easy (just join the two ends with a string outside the box), and vice
versa. Once all the knots have been boxed, defining their composition
is even easier: simply juxtapose the boxes and dissolve the double wall
that separates them
2
(Figure 4.2b).
48 KNOTS
Our immediate goal is to study the main properties of the composi-
tion of knots. The first is associativity, which tells us that:
(a#b)#c= a#(b#c)
where the symbol # denotes the composition of the knots. This equa-
tion means that composing first the two knots a and b and then com-
posing the knot obtained with the third knot c gives the same result as
composing first the two knots band c, then composing the knot ob-
tained with a. This assertion is obvious, for it means, roughly speak-
ing, juxtaposing the three knots (in both cases) and then eliminating
the two walls (in a different order, of course, but the result is the
same),
The following property is the existence of the trivial knot or unknot,
indicated by the number 1, which (like the number 1) does not change
the knot with which one composes it (just as the number 1 does not
change the number it multiplies):
a#l=a=l#a
In its "boxed version," the unknot can be represented as a horizontal
rectilinear thread in its cube. Juxtaposing such a box with that of any
other knot obviously does not change the type of the knot.
The following property, being trickier, merits a subheading of its
own.
Commutativity of the Composition of Knots
Like the multiplication of numbers, the composition of two knots is
commutative (the result does not depend on the order of the factors):
THE ARITHMETIC OF KNOTS 49
% % % % ~ % % % % % % % % ~ % ~ % ~ % % ~ ~ %
a#b= b#a
This relationship is not at all obvious, but I am sure that its demon-
stration, shown schematically in Figure 4.3, will please the reader.
What is going on in this figure? First, pulling on the ends of the
string that forms the first knot results in a small, tight knot (Figure
4.3b). Next, the little knot is slid along its own string, then along the
string that forms the second knot (c). Still sliding along the string, the
little knot traverses the big one and ends up to its right (d). Finally, the
second knot is sent into the first box, and the little knot is blown up to
original size. Voila-there you have it (f)!
It may be hard for the reader to understand how a knot can "slide
along a string." The simplest way is to take a good length of string (a
shoelace will do) and execute the maneuver. Certain organisms are
able to perform the process on themselves, which gives us the oppor-
tunity for another biological digression.
~ ~ .
~ - ® .
~ .
b) (c)
' A ~
Ila) :
II d)
® ~
-@-
Ie)
- ® ~ -
If) :
Figure 4.3. Composition doesn't depend on the order of the knots.
50 KNOTS
Digression: The Sliding Knot Fish
The strange fish in question is called a myxine (or, more commonly,
slime eel). The myxine inhabits the ocean bottom in temperate lati-
tudes. It has a supple backbone and secretes a very thick acidic saliva
that it uses to coat its body when a predator tries to grab it. It quickly
forms a knot with its tail and slides the knot along its body, thus
spreading the saliva (which it secretes simultaneously) along its length
(Figure 4.4a).
If you grab a myxine with your hand, it will slip through your
fingers-not only because it is coated with oily saliva, but also be-
(a)
Figure 4.4. The myxine's knot.
THE ARITHMETIC OF KNOTS 51
~ . 5 \ l ; % % % ~ % % ~ ~ ~ % %
cause of its knot, which it advances by pushing forcefully against your
fist while its head moves toward the rear and escapes your grasp (Fig-
ure 4.4b). The traction it gets from moving its knotted body enables
the myxine to carry out other vital tasks, especially feeding, which it
achieves by suction (the myxine is a necrophage that leaves behind
only the skin and bones of its prey).
Finally, once the danger has passed, the myxine gets rid of its layer
of saliva (otherwise it would smother in its slimy cocoon) by the
same sliding motion of the knot from the tail to the head. (For more
details about this unusual animal, see Jensen, 1966.)
Note that the slime eel's knot is a trefoil (that is, the simplest non-
trivial knot) and, in general, the left trefoil. To my knowledge, the
myxine does not know how to make any other knots, but one could
easily imagine a longer species of eel, with an even more flexible back-
bone, that would be capable of executing the same maneuver with
more complicated knots.
But let us leave aside these biological knots and come back to their
(admittedly more appetizing) mathematical models.
Can One Knot Cancel Out Another?
Having defined the composition of knots, one might ask whether in-
verse knots exist. That is, for a given knot, can one find another knot
that, composed with the first, gives the unknot? In more geometrical
language: If there is a knot at the end of a string, can a knot be made at
the other end in such a way that the two knots cancel each other out
when one pulls on the two ends of the string?
The analogous response to the question for natural numbers is no:
for every natural number n > 1, there is no natural number m such
52 KNOTS
that n . m = l. (Of course. one can take m = 1/ n, but then m would
be a fraction. not a natural number.)
We will see that it is exactly the same with knots: no trivial knot
possesses an inverse knot. This assertion is far from obvious. Indeed, it
appears false from the start: Why can't one make a "symmetrical" knot
at the other end of the string that would cancel out the first one?3
Why? Before reading the explanation below, try some experiments
with a string, beginning with the trefoil knot. Perhaps the ensuing fail-
ures will provide some clues for reflection.
Let us reason by contradiction: Let a and b be knots (nontrivial,
that is. a :;a!: 1, b :;a!: 1) such that a # b = 1. Consider the infinite compo-
sition:
c= a#b#a#b#a#b#a#b#a#b#a#b# ...
On the one hand, this composition is equal to the unknot, because we
can write:
C = (a # b) # (a # b) # (a # b) # ... = 1 # 1 # 1 # ... = 1
But by arranging the parentheses differently, we obtain:
C = a # (b # a) # (b # a) # ... = a # (a # b) # (a # b) # ...
= a # (1 # 1 # ... ) = a # 1 = a
Thus we deduce that a = 1, which contradicts the assumption a :;a!: 1
and "shows" that there are no inverse knots.
The quotation marks in the preceding sentence indicate that in fact
the "proof" is doubtful (that is the least one can say). In fact. coming
back to whole numbers, one can "prove" in the same way that 1 = 0:
THE ARITHMETIC OF KNOTS S3
~ ~ % % % % % ~ ~ % ~ % % %
just consider the infinite sum 1 - 1 + 1 - 1 + 1 - 1 + 1 - 1 + ...
and place parentheses in two ways, precisely as above. The error in rea-
soning is the same in both cases: one cannot manipulate infinite sums
or compositions (they must be defined beforehand) as one manipu-
lates finite sums or compositions.
In the case of knots, however, a slight modification of the argument
makes it totally rigorous. Just place copies of the knots a and b one by
one in an infinite series of boxes that become smaller and smaller and
converge to a point-which, by the way, rigorously defines the infinite
composition C (Figure 4.S)-and replace the questionable arithmetic
manipulations by correctly defined topological manipulations. I will
skip the technical details.
4
The reader will have to take my word that
this clever proof using infinite compositions is more than a brilliant
sophism (like that of Achilles and the tortoise); indeed, it is the basis
of a rigorous mathematical argument.
Finally, to end this section, let me add that the author of this clever
argument is not a specialist in knot theory but the German philoso-
pher and politician Wilhelm Leibniz, a great mathematician when he
put his mind to it (inventor, independently of Newton, of the differen-
tial and integral calculus). Leibniz discovered this argument in an en-
tirely different context, since knot theory did not exist at the time, in
order to prove (correctly) a theorem concerning a classical object in
the differential calculus-conditionally convergent series.
Prime Knots
We have just seen that there are no inverse knots, just as there are no
inverse natural numbers (which means, in other words, that the num-
M KNOTI



1=0
Figure 4.5. One knot cannot cancel out another.
THE ARITHMETIC OF KNOTS 55
ber 1 has no divisor other than itself). This property (of having no di-
visors other than itself and 1) is in fact the definition of prime num-
bers, in principle taught in the earliest grades at school. This simple
definition gives the following mysterious set:
2,3,5,7,11,13,17,19,23,29,31,37,41,43,47,53, ...
The set of prime numbers has continued to mystify mathematicians
since the beginning of their profession. But what is the situation for
knots? Are there prime knots, knots that cannot be represented as the
composition of two other nontrivial knots? The answer is yes: the tre-
foil, the figure eight knot, and the two alternating knots with five
crossings (Figure 4.6) are prime, whereas the square knot (called a
double knot by boy scouts) and the granny knot are composite knots.
How can we establish that prime knots exist? How does one prove,
for example, that the trefoil is indeed a prime knot? The notion that
comes immediately to mind is to use the minimal number of knot
crossings: if the trefoil (which has three crossings) is the composition
Figure 4.6. Prime knots and composite knots.
56 KNOTS
~ % % ~ % ~ ~ ~ % % % % ~ % % % % ~ % %
of two other nontrivial knots, those would have at least three crossings
each (because knots with two crossings or fewer are trivial). Two times
3 makes 6, 6 is greater than 3, and so there is a contradiction: QED.
Unfortunately, this argument is insufficient because we do not know
whether the minimal number of crossings of a composite knot is
equal to the sum of the minimal number of crossings of the two fac-
tors. This assertion is actually true, but its proof is too difficult to in-
clude in this book.
Thus, every knot decomposes into prime knots. Now, the decompo-
sition of each natural number into prime factors is unique. How does
that work for knots?
Unique Factorization into Prime Knots
Here, too, the parallelism with natural numbers is complete: every
knot decomposes uniquely into prime knots. Obtaining the proof of
this marvelous theorem was the ambition of many researchers. It was
Horst Schubert who did it at the end of the 1940s, though his proof, at
once profound and very technical, is beyond the scope of this book.
Schubert's theorem, which is associated with other properties com-
mon to knots and to natural numbers, brings us to the obvious idea of
numbering knots in such a way as to conserve the decomposition into
prime factors. Such a numbering would associate a prime number to
each prime knot, a composite number to each composite knot, such
that the prime factors of the number are the prime factors of the knot.
Alas! such numbering may indeed exist in principle, but there is no
natural algorithm that can produce it.
The reason is that, in contrast to numbers, two knots cannot be
THE ARITHMETIC OF KNOTS 57
added; they can only be multiplied (composed). Each positive whole
number can be obtained by adding an appropriate quantity of ones
("trivial numbers for multiplication"); thus 5 = 1 + 1 + 1 + 1 + l.
On the other hand, it is not possible to obtain all knots by "adding
copies of the trivial knot;' since this operation of addition does not
exist.
5
Another reason there is no natural numbering of knots is the lack
of order in the set of knots. Natural numbers possess a natural total
order (1,2,3,4,5); this order does not exist (or has not been discov-
ered!) for knots. Of course, it is usual to order knots by the number of
minimal crossings of their diagrams, but this order is not linear: which
of the two knots with five crossings (see the table of knots in Figure
1.6) is the "smallest"?
So the arithmetic of knots has not helped us to classify them. But
there is scant reason to talk of failure here: Schubert's theorem does
not need any applications; it is mathematical art for art's sake, and of
the most exalted kind.
SURGERY AND INVARIANTS
(Conway· 1973)
In 1973, the English mathematician John Conway discovered the fun-
damental role played in knot theory by two very simple "surgical op-
erations;' two ways of modifying a knot near one of the crossings of its
strands.
The first, which we will call the flip, consists in transforming the
chosen crossing (on the planar representation of the knot) into the
opposite crossing, that is, running the upper strand under the lower
strand (Figure 5.1); with a string, the flip can be performed by cut-
ting the upper strand, then reattaching it under the other strand.
Of course, the flip can change the type of the knot; for example,
"flipping" one of the crossings of a trefoil knot produces a trivial knot
(the trefoil unknots-try to draw it, and you'll see).
Conway's second little surgical operation, called smoothing, consists
in eliminating the crossing by interchanging the strands (Figure S.2a);
with a string, it is done by cutting the two strands at the point of
crossing and reattaching them "the wrong way" (Figure S.2b). Note
that when the strands are not oriented, there are two ways of reattach-
ing the four ends two by two (b or c), but the orientation of the knot
58
SURGERY AND INVARIANTS 59
x x
Figure 5.1. The flip (the upper strand becomes the lower strand).
x
)(
<
>
(a)
~ /' ) ( ~ \11, .;
-, - - ~
I ' ~ ~ g i A
(b) (c)
Figure S.2. Smoothing (the strands are reattached the wrong way).
60 KNOTS
~ % % % % ~ ~ % ~ ~ %
leaves us with no choice as to which pairs of ends should be reattached
(this is dictated by the arrows, as in Figure S.2a).
The flip and the smoothing operation were known and often used
by topologists well before Conway; in particular, J. W. H. Alexander
used them to calculate the polynomials that bear his name (to which
we will return later). Conway's contribution was to show that these
two operations can serve as the basis for a fundamental definition of a
knot invariant (the Conway polynomial), which we will take up a little
further in this chapter.
Actually, the importance of Conway's operations extends far be-
yond the scope of knot theory. Flipping and smoothing play an essen-
tial role in life itself, as they have been routinely used by nature ever
since biological creatures began to reproduce. The description of this
role-which is quite extraordinary-merits at least a small digression.
Digression: Knotted Molecules, DNA,
and Topoisomerases
Watson and Crick's seminal discovery of DNA, the molecule that car-
ries the genetic code, presented biochemists with a series of topologi-
cal puzzles, among other problems. We know that this long, twisted
double helix is capable of duplicating itself, then separating into two
identical molecules that-unlike the two constituents of the parent
molecule-are not bound together but are free to move. How is that
possible topologically?
Intricate studies have shown that enzymes called topoisomerases
specialize in this task. More specifically, topoisomerases perform the
three basic operations shown in Figure S.3. The reader will immedi-
SURGERY AND INVARIANTS 61
Figure 5.3. Topoisomerase operations on DNA.
ately recognize operations (a) and (b): they are the flip and Conway's
smoothing! The third operation (Figure 5.3c), which is called the
twist, is also known in topology; it refers to the mathematical theory
of ribbons, which is currently very useful in theoretical physics.
Let us look more closely at how these strange enzymes affect long
molecules, especially DNA molecules. First of all, note that the rear-
rangement of the strands occurs at the molecular level and is not visi-
ble: even the most powerful electron microscopes provide only indi-
rect information.
Remember, too, that the DNA molecule appears in the form of a
long double helix, each strand of which is constituted of subunits, the
bases A, T, C, and G, whose arrangement on the strand codes the ge-
netic properties of the organism (a little like the way the arrangement
of the numbers 0, 1, 2, ... , 9 on a line of text gives the decimal code of
a number). Figure 5.4 is a schematic representation of a fragment of a
DNA molecule.
It is well known that the double strands of DNA ordinarily have free
Bases:
A - adenine
T - thymine
C - cytosine
G - guanine
Figure 5.4. The structure of a DNA molecule.
ends, but it is not always so; there are also molecules with closed dou-
ble strands (the twice-coiled snake biting its tail) and single-stranded
molecules with both closed and open ends. These molecules take part
in three classical genetic processes-replication, transcription, and re-
combination; moreover, the double-stranded molecules form super-
coils or tangles (this transforms these drawn-out objects into compact
ones). The topoisomerases playa crucial role in this whole process
by carrying out the tasks of cutting, rearranging, and rejoining. First,
the enzymes nick one of the strands, pass the second strand through
the opening, then reattach the cut so that the strands change place
(Conway's flip). Conversely, by means of two cuts and two repairs,
SURGERY AND INVARIANTS 63
the enzymes are able to reattach the two strands "the wrong way"
(Conway's smoothing).
The precise mechanism of the cutting, rearranging, and rejoining
operations is still not well understood. But it is known that there are
different types of topoisomerases (they are not the same for single-
and double-stranded DNA). And we have some idea, from the work of
James Wang, how supercoiling (and the reverse operation) of a closed
double-stranded DNA molecule works.
DNA supercoiling is analogous to what often happens to a tele-
phone whose handset is connected with a long spiral cord. When the
user twists the cord while returning the handset to the base, the con-
necting wire gets more and more tangled, eventually becoming a sort
of compact ball. For the phone user, this result is rather annoying,
since the coiling shortens the distance between the phone and the
handset. In the case of DNA, supercoiling also transforms the long
spiral into a compact ball, but here the result is useful, because trans-
forming the long molecule (several decimeters) into a tiny ball makes
it easy for the molecule to enter the nucleus of a cell, whose dimen-
sions are measured in angstroms. I
In its normal state (not supercoiled), the DNA spiral makes a com-
plete rotation with each series of 10.5 successive bases. In making
twists (take another look at Figure 5.3c), the corresponding topoisom-
erase transforms the simple closed curve of DNA in the manner indi-
cated in Figure 5.5. Note that, from the topological point of view, one
of the results of the twist is to change the winding number of the two
strands of DNA (this invariant, which we owe to Gauss, measures the
number of times one of the strands wraps around the other). There
64 KNOTS
figure 5.5. A supercoiled, double-stranded DNA molecule.
are many other topological phenomena that fascinate biologists, but
our goal is not to give a detailed description of the findings. For a
more thorough introduction, the reader is referred to Wang (1994).
Invariants in Knot Theory
Let us return to the mathematical theory of knots, to speak-at last-
of invariants. How do they appear, and what role do they play in the
theory?
Roughly, knot invariants serve above all to respond, when needed,
in the negative to the most obvious question regarding knots, which
we have called the comparison problem: Given two planar representa-
tions of knots, can we say whether they represent the same or different
knots? For example, drawings (a) and (e) in Figure 5.6 represent the
same knot-the trefoil. Indeed, this same figure shows how represen-
tation (a) can be converted to representation (e). On the other hand,
SURGERY AND INVARIANTS 65

Figure 5.6. Six representations of the same knot?
any attempt to transform drawing (f) into a representation of a tre-
foil is doomed (try it!). But how do we prove it? The fact that we have
not succeeded in turning one sketch into another does not prove any-
thing; a more clever person, or a luckier one, might be able to do it
easily.
Now suppose that we have at our disposal a knot invariant, that is,
a way of associating, with each planar representation of a knot, a cer-
tain algebraic object (a number, a polynomial) so that this object
never varies when the knot is manipulated, as in the first five sketches
(a-e) in Figure 5.6. When two planar representations are given (for ex-
ample, f and e in Figure 5.6), one can calculate their invariants. Dif-
fering values for the invariant prove that the two representations do
not define the same knot (if they did, they would have the same in-
variant!).
66 KNOTS
~ % % % % ~ % % % ~ ~
For example, the calculation of Conway's polynomial (to be de-
scribed in more detail later) of representations (a) and (f) in Figure
5.6 gives :il + 1 and -,x4 + x-I, respectively; therefore, the diagrams
in question clearly represent two different knots.
Before moving on to the study of Conway's invariant, let us try to
find a numerical invariant for knots ourselves. The first idea that
comes to mind is to assign to each knot diagram the number of its
crossings. Unfortunately, this number is not an invariant: as one ma-
nipulates a knot in space, new crossings may appear on its planar pro-
jection, and others may disappear (for example, see Figure 5.6). Those
who have read Chapter 3 may remember that the first and second
Reidemeister moves change the number of crossings by adding ± 1
and ± 2, respectively.
But it is easy, using this idea, to try to find a genuine invariant of the
knots: just consider the minimal c(N) number of crossings of all the
projections of knot N. This number (a nonnegative integer) is by
definition an invariant (it does not depend on the specific projection
given, since the definition involves all projections). Unfortunately, it is
useless for comparing knots, because it would work only if we could
calculate c(N) from a given projection N. Since we do not presently
have any algorithm for doing this calculation, we will move on to a
more sophisticated but calculable invariant: Conway's.
Conway's Polynomial
To every planar representation N of an oriented knot, Conway associ-
ates a polynomial in x, noted V (N), which satisfies the three following
rules:
SURGERY AND INVARIANTS 67
[nvariance Two representations of the same knot have the same
polynomial:
N - N => V (N) = V (N)
(I)
Normalization The polynomial of the unknot is equal to 1 (it is re-
garded as a "zero-order polynomial"):
V(O) = 1 (II)
Conway's skein relation Where the three planar representations N+,
N_, and No are identical outside the neighborhood of one crossing
and have forms as indicated in the diagrams as follows:
then
(III)
(In other words, No and N_ are derived from N+ by a smoothing and a
flip, respectively.)
For example, when N + defines the trefoil knot, the Conway notation
specifically gives:
The attentive reader will have noticed that in this case the diagram No
(on the right side of the equal sign) no longer describes a knot: it con-
sists of two dosed curves instead of one, and it is the diagram of a link
(a family of curves in space that can knot separately as well as link to-
gether). But never mind that-Conway's polynomial is actually de-
fined for all links (of which knots are just a particular case).
From now on, we will write Conway's skein relation (and other sim-
ilar relations) in the following symbolic form:
which means that there are three identical links outside the circular
areas (bounded by the dotted lines) that each contain one crossing.
The second and third links are obtained by flipping and smoothing
the first link in this neighborhood.
Examples of the Calculation of the Conway Polynomial
One of the advantages of the Conway invariant is the ease with which
it can be calculated. Here are some examples.
Consider the link consisting of two separate circles. Then we have
V(OO) = O. In fact:
-
- V(O) j,g 1 - 1 = 0
Now assume that the links consist of two linked circles, known as the
Hop! link, H = (jJ). Following Conway's notation,
and since V(OO) = 0 and V(O) j,g 1, we deduce that V(H) = x.
SURGERY AND INVARIANTS 69
Finally, let us calculate Conway's polynomial for the trefoil T. For
that, we return to equation 5.1; by virtue of rule I, the second term on
the lefthand side is equal to V(O), and by virtue of rule II, to 1; the
term in the righthand side, according to the preceding calculation, is
equal to x . x = il. Thus one obtains V(T) = il + 1.
Expressed in this way, the calculation of Conway's polynomial for a
knot (or link) appears to be a rather original mix of geometrical oper-
ations (namely, the flip and the smoothing) and classical algebraic op-
erations (sums and products of polynomials). The reader with a taste
for this type of thing will undoubtedly enjoy calculating V(P), where P
is the knot represented in Figure 5.6£.
Discussion of Results
What can we deduce from these calculations? Many things. In particu-
lar, we now have formal proof that:
(1) The two curves of a Hopf link cannot be separated:
(2) A trefoil knot cannot be unknotted:
(3) The knot represented in Figure 5.6fis not a trefoil:
Of course, the reader who has not yet been ruined by mathematics will
say that there is no point in having a formal proof of something so ob-
vious as statements (1), (2), or (3). But one can argue that what we
have here is a general method that also works in more complicated sit-
uations, where intuition proves of no avail.
For example, for the planar representations of Figure 5.7, my in-
tuition of space (fairly well developed) does not necessarily tell me
anything about the knot(s?) they represent. Yet my little notebook
computer, which has software for calculating Conway's polynomial,
trumpeted after a few seconds' reflection that V(A) = 1 and V(B) = :xl
+ 1, which proves that the knots represented by A and B are different.
Thus we have in Conway's polynomial a powerful invariant that al-
lows us to distinguish knots. Does it always work? In other words, does
the equality of the polynomials of two knot representations imply that
they are two representations of the same knot? Does one always have
V(K,) = V(K
2
) ::;. K, = K2? Unfortunately, the answer is no: a calcula-
Figure 5.7. Two representations of the same knot?
SURGERY AND INVARIANTS 71
tion similar to this one shows that the Conway polynomial for the fig-
ure eight knot (Figure 1.2) is equal to X2 + 1: it is the same as that for
the trefoil. The Conway polynomial does not distinguish the trefoil
from the figure eight knot; it is not refined enough for that.
But-the sceptical reader will counter-what tells us that the trefoil
and the figure eight knot are not, in fact, the same knot? Good ques-
tion. We will only be able to answer it conclusively when we have
an invariant more sensitive than Conway's. One example is Jones's
famous two-variable polynomial (the topic of the next chapter) or
the Homfly polynomial, which can also be obtained using Conway's
method and with which we will end this chapter.
The Homfly Polynomial
"Homfly" is not the name of the inventor of this polynomial: it is an
acronym for the six(!) researchers who discovered the same polyno-
mial at the same time and published their results simultaneously (in
1985) in the same journal. They are H = Hoste,O = Ocneanu, M =
Millet, F = Freyd, L = Lickorish, and Y = Yetter.
2
The simplest way to define the Homfly polynomial P(x, y) (with
two variables x and y) is to use Conway's axioms I, II, and III with Pin
the place of V and with the following modification to the skein rela-
tion (axiom III):
xp(X)-yp(X)=p()(). (III')
The reader who has grasped how to perform the little calculations of
Conway polynomials of knots will perhaps enjoy redoing these calcu-
lations with the new skein relation III' for the same knots and links. In
72 KNOTS
particular, he will then see that the Homfly polynomials for the trefoil
and the figure eight knot are not the same.
The Homfly polynomial is thus more sensitive than Conway's. But
is it a complete invariant? Can it distinguish all nonisotopic knots?
Unfortunately, the answer is no: Figure 5.8 shows two different knots
that have the same Homfly polynomial.
That is why the search for a complete invariant continues in the fol·
lowing chapters.
Figure 5.8. Two knots that have the same Homfly polynomial.
6
JONES'S POLYNOMIAL AND SPIN MODELS
(Kauffman· 1987)
The discovery by Vaughan Jones of the polynomial that bears his
name reinvigorated the study of knot invariants. Of that there can be
no doubt. But the significance of the famous polynomial extends far
beyond the scope of knot theory: it is popular because of its connec-
tions to other branches of mathematics (the algebra of operators,
braid theory) and especially to physics (statistical models and quan-
tum groups).
It thus makes sense to devote this chapter-a key one-to Jones's
theory. Unfortunately, as conceived by its author at the outset, this
theory is far from elementary (see Stewart, 1989), and it exceeds, by a
long shot, the assumed mathematical sophistication of the wide read-
ership I hope to reach. But it so happens that another approach to
Jones's polynomial, that of Louis Kauffman of the University of Chi-
cago, has the double advantage of being very easy and of clearly show-
ing the relationship of the polynomial with statistical physics. It is on
this branch of physics that I shall base an explanation of the theory,
and so I will begin with a few of its fundamental notions.
73
74 KNOTS
Statistical Models
For a good thirty years (and especially since the publication in 1982 of
Roger Baxter's book on this subject), statistical models and in particu-
lar the famous Ising model have captured the attention of both math-
ematicians and physicists. What is it all about? It has to do with theo-
retical models of regular atomic structures that can adopt a variety of
states, each state being determined by the distribution of spins on the
atoms (a very simple example is shown in Figure 6.1). At any given in-
stant, each atom (represented in the figure by a fat dot) is character-
ized by its interactions with its neighboring atoms (an interaction is
represented by a line joining the two atom-dots) and its "internal
state"-what physicists call its "spin:' a parameter
l
that has a finite
number of values (in the model here, two). The two spins in this
model are up and down and are represented by arrows pointing up
and down, respectively.
For the model to be wholly deterministic, we must specify its parti-
tion function. This is an expression of the form:
Z(P) = 2: exp ~ ; 2: e[s(a j),s(a)) (6.1)
,ES (.,./)EA
where the external sum is taken over the set S of all states, and the in-
ternal sum is taken over all edges (interactions), while e[s(aj), s(aj)) is
the energy of the interaction between the atoms aj and aj (which ac-
tually depends only on their spins), T is the temperature, and k a coef-
ficient called Boltzmann's constant (whose value depends on the choice
of units).
JONES'S POLYNOMIAL AND SPIN MODELS 7S
~ ~ ~ ~ ~
Figure 6.1. The spin model.
Using the partition function Z, we can calculate a given model's to-
tal energy and the probability that it is in a given state and, espe-
cially, study its phase transitions-for example, in the case of the Potts
model of freezing water, its change from the liquid state (water) to the
solid one (ice) and vice versa.
2
I do not intend to delve further into the study of statistical models.
The little bit I have just said about it will be enough for the reader to
understand where Louis Kauffman got his funny idea of associating a
statistical model with each knot.
Kauffman's Model
Take any (unoriented) knot, such as the one shown in Figure 6.2. Look
closely at one of the knot's crossings; locally, each intersection divides
76 KNOTS
the plane into two complementary angles, one of which is type A (or
up type) and the other type B (or down type). The type-A angle is the
one we see to our right when we start moving along the upper strand,
and then (after we cross over the lower strand) to our left. (The direc-
tion chosen for crossing the intersection can be either of the two pos-
sibilities: the resulting type of angle does not depend on this choice-
check that!) In Figure 6.2, A angles are shaded, and B angles are left
blank.
Thus, for a given knot, we can choose at each intersection what
might be called a Kauffman spin. In other words, we can associate
with each crossing the word up or the word down. We say that such a
choice (at all crossings) is a state of our knot. A knot with n intersec-
tions thus has 2" possible states. To represent the knot in a specific
state, we could have written up or down next to each intersection, but I
prefer to draw a little stick inside the chosen angle (look again at Fig-
ure 6.2a, as well as Figure 6.2b).
This notation also has the advantage of dearly indicating the choice
between one of two ways of smoothing a crossing (of an unoriented
knot) by exchanging strands "following the little stick" (Figure 6.3b).
We will need to make this choice right away. Let us denote by S(K) the
set of all the states of the knot K. To completely define the Kauffman
model associated with the knot K, we need only define the corre-
sponding partition function. It will be denoted by (K), called the
Kauffman bracket, and defined by the equation:
(K)
'" a(,)-p(,) ( 2 -z)y(,l-I
= L.Ja -a -a
(6.2)
where the sum is calculated for all the possible 2" states s E S(K) of
the knot K, where a(s) and f3(s) denote the number of type-A and
JONES'S POLYNOMIAL AND SPIN MODELS 77
~ % % ~ ~ ~ %
(a)
Figure 6.2. State of a knot, illustrating angles of type A and B.
A
((=3
{J=ly=2
A
Figure 6.3. Smoothing a figure eight state.
78 KNOTS
type-B crossings, respectively, whereas y(s) denotes the number of
closed curves obtained when all the knot crossings are smoothed fol-
lowing the little s-state sticks.
One might ask where Kauffman got this unusual formula (which
is quite unlike its prototype in equation 6.1). Without going into
details, I would simply say that he found it by backward reasoning
(and trial and error), that is, by starting from the result he wanted to
obtain.
Be that as it may, the application of this formula is very simple
(though very laborious if the knot has many crossings). Figure 6.3
shows a possible state for the figure eight knot diagram (a) and the re-
sult of the corresponding smoothing (b); for this example, a(s) = 3,
/3(s) = 1, and y(s) = 2 (two closed curves appear after all the crossings
are smoothed). Consequently,
Obtaining the value of Kauffman's bracket for the figure eight knot di-
agram requires drawing all 16 possible states of the diagram (16 = 24)
and summing the 16 results for the equation above (which describes
just one of the 16 states). In this way, we get a polynomial
3
(in a),
which is the value of the Kauffman bracket for the diagram of the
given knot.
Note in addition that equation 6.2 is also valid for links of more
than one component.
Before continuing with our study of the Kauffman bracket, let us
pause a moment to compare the result we get with a classical model,
such as the Potts model. Let us begin by comparing Figures 6.1 and
6.2. Aren't they awfully similar? Of course, the graphical similarity re-
JONES'S POLYNOMIAL AND SPIN MODELS 79
~ ~ % % ~ ~ %
sults from a judicious choice of the knot shown in Figure 6.2, but
generally one can say that the state of the knot and the state of a pla-
nar regular atomic structure are more or less the same. On the other
hand, equations 6.1 and 6.2, which give the partition functions of the
models, are totally different, and Kauffman's equation (6.2) has no
physical interpretation. Kauffman's model is thus not a "true" statisti-
cal model-which does not in any way diminish its usefulness for
knots. Moreover, we will see later on that true statistical models (in
particular Potts's model) can in fact be used to construct other knot
invariants.
Properties of the Kauffman Bracket
Our first goal is to indicate certain properties of this bracket to see
how to deduce an invariant of knots from the bracket.
The three principal rules are the following:
(KU O) = (-tal - a-
2
)(K)
(O) = 1
(I)
(II)
(III)
Let us begin at the end. The third and simplest rule tells us that the
Kauffman bracket for the diagram of the unknot 0 is equal to 1 (that
is, the zero-order polynomial with the constant term 1). Rule II, in
which K is any knot (or link), indicates how the value of (K) changes
when one adds to it a trivial knot unlinked with K: this value is multi-
plied by a coefficient equal to (-tal - a-
2
).
80 KNOTS
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ' C \ 1 ; ~ %
The first rule-which, despite its simplicity, is the fundamental rule
underlying Kauffman's theory and this chapter-describes the con-
nection between the brackets of the three links (or knots) symbolized
by the three icons:
which differ only by a single small detail. More precisely, the icons de-
note three arbitrary links that are identical except for the segments of
the strands indicated inside the dotted circles.
Those who have read the chapter devoted to the Conway surgical
operations will doubtless have noticed the analogy that exists between
Kauffman's rule I and skein relations. Let us simply recall what the
icons for the skein relations look like:
(X) {X:; :::5"C
,.,.,." ....... '" ... ,'
How are they different from those in rule I? First, the knots considered
by Kauffman are not oriented (there are no arrows). Consequently,
there is only one type of crossing (two according to Conway), but
two ways of smoothing (Conway's arrows impose a single smoothing,
the same for the two different crossings). Having said that, rule I of
Kauffman's theory is, like Conway's skein relations, nothing more than
a very simple equation related to a local surgical operation.
Rules I-III make child's play of calculating a Kauffman's bracket
from the diagram of a knot (or link): just apply rule I (taking care to
note the intermediate equations obtained) until all the crossings are
gone, then calculate, with the help of rules I and II, the bracket of the
link of N disjoint circles-which is obviously equal to (-a
2
- a-Z)N-I.
JONES'S POLYNOMIAL AND SPIN MODELS 81
% ~ % ~ % % ~ ~ ~ ~ % % % % ~ % % % % % ~
Then use that value to find the given knot step by step, employing the
intermediate equations.
For example, for the unknot (using rule III) and the trivial two-
component link (using rule 11), we get:
(0) = 1
(00) = (-a
2
- a-
2
)
Following rule I and the preceding result,
(00) = a(O) + a-I(OO)
= a . 1 + (a-
I
)( -a
2
- a-
2
) = -a-
3
In a similar way,
(00) = -a-
3
For the Hopf link, using the preceding calculations, we get:
For trefoils, still using the preceding results, we obtain these formulas:
Of course, the Kauffman bracket can only be used in knot theory if it
is invariant, that is, if two diagrams of the same knot always have the
82 KNOTS
same Kauffman bracket. This basic question is treated in detail in the
following section, which is intended more specifically for those who
have read the chapter on Reidemeister moves. The reader who has not
(or who does not like mathematical proofs) will miss little by going
directly to the subsequent paragraphs, where I finally introduce the
Jones polynomial.
I nvariance of the Kauffman Bracket
Thanks to Reidemeister's theorem, proving the invariance of the
b r a c k ~ t requires only showing that its value does not change when the
knot (or the link) undergoes Reidemeister moves. The reader will per-
haps remember that there are three of those moves; take a look at Fig-
ure 3.1 in Chapter 3.
Let us begin with the second move, O
2
, Using rule I several times
and rule II once, we get:
Comparing the first and the last member of this series of equalities, we
see that the invariance with respect to O
2
has been established. The at-
tentive reader will have noticed the miraculous disappearance of the
JONES'S POLYNOMIAL AND SPIN MODELS 83
~ ~ ~ ~ ~
different powers of a to give 1 for the coefficient at the desired icon
) C, and 0 for the undesirable icon:-g::. Of course, this is no coin-
cidence: the selection (a priori bizarre) of the coefficients in equation
6.2 for Kauffman's bracket is motivated precisely by this calculation.
Inspired by our little victory, let us move on to verifying the invari-
ance relative to Q3, the most complicated of the Reidemeister moves.
Still using the basic rule I, we can write:
(6.3)
Obviously,
( ~ ) = ( ® )
since these two diagrams are isotopic in the plane. Now, twice apply-
ing the invariance relative to Q
2
(which we have just demonstrated),
we obtain:
(®) = (.g::) = (§)
Comparing the righthand sides of the two equalities in equation 6.3
shows that they are equal term by term. That is therefore also true for
the lefthand sides. But it is precisely this equality that expresses the
invariance of the bracket with respect to Q
3
!
84 KNOTS

A Little Personal Digression
God knows I do not like exclamation points. I generally prefer Anglo-
Saxon understatement to the exalted declarations of the Slavic soul.
Yet I had to restrain myself from putting two exclamation points in-
stead of just one at the end of the previous section. Why? Lovers of
mathematics will understand. For everyone else: the emotion a mathe-
matician experiences when he encounters (or discovers) something
similar is close to what the art lover feels when he first looks at Michel-
angelo's Creation in the Sistine Chapel. Or better yet (in the case of a
discovery), the euphoria that the conductor must experience when all
the musicians and the choir, in the same breath that he instills and
controls, repeat the "Ode to Joy" at the end of the fourth movement of
Beethoven's Ninth.
Invariance of the Bracket (Continued)
To prove the invariance of Kauffman's bracket, it remains only to ver-
ify its invariance relative to the first Reidemeister move, QI, the sim-
plest of the three. Using rules I and II, we obtain:
(
:.)(') = a('o ....• ) + a-I( ..:ji·.) = A( St.··.···)
..
where A = a(-a
2
- a-
2
) + a-I = -a
3

Disaster! This blasted coefficient a
3
has refused to disappear (for
the other little loop, we get the coefficient a-
3
). Failed again! The
Kauffman bracket is not invariant relative to Q
I
and thus is not an in-
variant of knot isotopy. For example:
JONES'S POLYNOMIAL AND SPIN MODELS 85
~ ~ % % % % % % % % ~ % ~ ~ ~
But the two diagrams both represent the trivial knot, and thus we
should get:
()(») =()(») = (0) = 1
Should we plunge into despair?
Another Little Personal Digression
That's exactly what I did fifteen years ago when I was working on these
same questions. The Jones polynomial and skein relations had made
their appearance, and like many other mathematicians, I was playing
with variants of these relations in the hope of finding invariants more
sensitive than Jones's. Later, among my scribbles, I retrieved some for-
mulas very close to Kauffman's rule I, and I remembered having been
stymied by this same move 0
1
(unable to get rid of a stubborn coef-
ficient that just would not disappear) and dropping it.
But Kauffman persevered. Budding young researchers will grasp the
moral of the story, even if perseverance doesn't always pay off-I am
not at all sure that by continuing I would have managed to discover
the fantastic little trick that allowed Louis Kauffman to succeed.
Kauffman's Trick and Jones's Polynomial
The starting point is obvious: since the coefficient a:!:3 refuses to go
away, we will have to add a supplementary factor to our bracket whose
purpose would be to rid us of this tiresome a:!:3. But how?
86 KNOTS
~ ~ % % % % % % % ~ %
Let us calion a classical tool of knot theory, the writhe, which is de-
fined in the following way: for every oriented knot K, the writhe w(K)
is the integer equal to the difference between the number of positive
crossings and the number of negative crossings.
It is easy to see that the writhe is an invariant of the Reidemeister
moves O
2
and 0
3
, In contrast, move 0
1
changes the writhe: it adds
1 or -1 to it according to whether the eliminated loop is negative
("i"' ) or positive (:"i":)·
Still following Kauffman's lead, let us now define the Jones poly-
nomial
4
of an oriented knot (or link) K by writing:
X(K) = (_a)-3w(K)( I KI)
(6.4)
where the non oriented diagram I K I is obtained from the oriented
diagram Kby forgetting its orientation (erasing the arrows) and where
( . ) is the same Kauffman bracket that has caused us so much joy and
sorrow.
Kauffman's trick is this factor (-a)-31<\K), which does a superb job
of killing off the annoying coefficient a±3 resulting from move 0
1
,
(I leave to the "mathematized" reader the pleasure of unraveling the
details of this sophisticated murder, worthy of Agatha Christie.)
It is now obvious that the Jones polynomial is an isotopy invariant
of knots (and links). Indeed, we have seen that the bracket (.) as well
as the factor (-a)-31<\K) are invariant with respect to Reidemeister
moves O
2
and 0
3
, and everything goes smoothly with 0
1
as well (lazy
readers will have to take my word for it); the isotopy invariance of
X( . ) therefore follows from Reidemeister's theorem.
Before describing what Jones's polynomial contributes to knot the-
ory, I am going to take advantage of the fact that the basic rule (I) of
JONES'S POLYNOMIAL AND SPIN MODELS 87
~ % ~ ~ ~ % % % % % ~ % %
Kauffman's bracket is still fresh in our minds to linger a little on the
history of this equation.
A New Digression-On Menhirs
No mathematician would deny Louis Kauffman the honor of having
invented rule I, an insignificant-looking little formula whose funda-
mental character was immediately obvious. Yet, only a few years ago,
Kauffman learned that he was not the first to come up with the for-
mula: a specialist of ancient Celtic culture explained to him that sculp-
tors who worked on menhirs six thousand years ago used exactly the
same rule to alter the structures of connected ribbons (thus of knots
and links) that decorated these burial stones. Readers will find motifs
from rule I (can I still call it Kauffman's rule?) in the links of the rib-
bons carved on the menhir shown in the Preface (Figure P.5).
Rules for the Jones Polynomial
I have just demonstrated the first fundamental rule of Jones's polyno-
mial:
(1) Two diagrams of the same knot (link) have the same Jones poly-
nomial.
The second fundamental rule-whose proof derives from a fairly easy
calculation based on the "Celtic rule" (I) of Kauffman's bracket and
equation 6.4-is the skein relation for the Jones polynomial:
88 KNOTS
The two other rules are obtained directly from rules II and III for the
bracket:
(3) X(KUO) = (-a - a-
2
)X(K).
(4) X(O) = 1.
These rules are sufficient to calculate the Jones polynomial for specific
knots and links. (In fact, one can show that rules (1)-(4) entirely de-
termine the Jones polynomial.)
Let us do the calculation for the trefoil (to simplify the notations, I
have written a-
4
= q).
We have obtained the trivial knot and the Hopf link. Let us do the cal-
culation for the latter:
Following rules (3) and (4),
X( (0)) = q-2(qIl2 + q-1/2) - q-l(qI/2 - q-1/2) =
= _q-1/2 _ q-'/2
Thus, for the trefoil:
X(&) = q-2 + q-l(q-1/2 + q-S/2)(qI/2 _ q-1/2)
= q-I + q-3 _ q-4
Readers who have developed a taste for these calculations can check
that the same result is obtained if we use equation 6.4 and the preced-
ing calculation for the Kauffman bracket of the trefoil.
JONES'S POLYNOMIAL AND SPIN MODELS 89

Similar calculations show that the knots in the little table of knots
presented in the first chapter are all different. Do not think that prov-
ing this fact has no purpose other than to satisfy our mathematical
pedantry. When Jones's polynomial appeared, its calculation for the
knots with 13 crossings or fewer gave different values for all the knots,
except for two specific knots with 11 crossings. This result was suspect,
and a closer comparison of the diagrams of these knots (which look
quite different) showed that they were in fact isotopic diagrams (dia-
grams of the same knot): the table was wrong.
This application (which caught the attention of specialists in knot
theory) led Vaughan Jones to hope that his polynomial would be a
complete invariant, at least for prime knots. Unfortunately, that wasn't
to be: there are nonisotopic prime knots that have the same Jones
polynomial (see, for example, Figure 5.8).
Nonetheless, for knot theory, the role of the Jones polynomial re-
mains very important: it is a sensitive invariant-more sensitive, for
example, than Alexander's polynomial. It distinguishes, among others,
the right trefoil from the left trefoil, which Alexander's polynomial
cannot do. Moreover, Jones himself and his followers found new ver-
sions of his polynomial that were even finer.
Having said that, I must point out that no one has succeeded in
finding a complete invariant along these lines. Another effort, based
on totally different ideas, is described in the next chapter.
FINITE-ORDER INVARIANTS
(Vassiliev· 1990)
Victor Vassiliev should never have worked on knots. A student of
Vladimir Arnold, and consequently a specialist in the theory of singu-
larities (better known in the West under the media-friendly term ca-
tastrophe theory), he was unable to apply the techniques of this theory
directly to knots-objects with a regular local structure, smooth and
continuous, without the least hint of a catastrophe.
Perhaps a wise humanist whispered in his ear, "Since the singularity
doesn't exist, it must be invented:' Be that as it may, Vassiliev did in-
vent it.
The concept is of disarming simplicity. Together with proper knots,
Vassiliev explains, one must consider singular knots; these differ from
true knots in that they possess double points, where one part of the
knot cuts another part transversally (::.)-.<:: ).
In the planar representation of a knot, the appearance of double
points barely differs from that of crossings, ((Z)) or ( :: .. X.::). You
might say that moving a true knot in space produces a "catastrophe"
90
FINITE-ORDER INVARIANTS 91
when one part of the knot crosses another; at that instant, the knot be-
comes singular, then immediately reverts to an ordinary knot, but one
that may be different from the initial knot. As an example, Figure 7.1
shows how the trefoil knot is changed (as it undergoes a catastrophe)
into a singular knot with a single double point and then becomes a
trivial knot.
Vassiliev jumbles together in one set (denoted by <ji) ordinary knots
with singular knots, which may possess any (finite) number of dou-
ble points. Ordinary knots thus form a subset of <ji denoted by Io,
whereas the others form what is called the discriminant I. This is bro-
ken up into strata II, I
2
, I
3
, ••• , constituted of singular knots with 1,
2, 3, ... double points, respectively. It is in the vicinity of these strata
that we are going to pursue our study of the invariants of knots.
Unfortunately, the stratified set <ji :::> Io U II U I2 U ... is of
infinite dimension, so it is difficult to visualize. Nonetheless, I shall de-
scribe it very geometrically (though not very rigorously), shamelessly
employing crude drawings, where the space (of infinite dimension!)
<ji will be represented by a square: the one at the center of Figure 7.2.
The points of <ji thus represent knots (singular or ordinary); around
the square, we see more "realistic" representations of some of these
knot-points, which show the process of deformation of a knot (the
"figure eight knot") in Euclidean space 1R3. Inside the square, we see
the path taken in the space <ji (a "symbolic" representation of the same
deformation) by the mobile point H..,. G..,. F..,. D..,. C..,. B..,. A ..,. 0
corresponding to the knot undergoing deformation.
At the first catastrophic moment (when the double point 1 forms
on the figure eight knot H), the mobile point cuts through the stratum
II (the stratum of the singular knots having exactly one double point)
Figure 7.1. The trefoil becomes singular, then unknots.
dJ +-- ~ ..... ~ @o
G ~ ~ - - H ~ ~ ~ __ ~ ~
i
~
(t
A
t
i
L---L--_-'--_-'--_----JF
d)D ~ ( ~ c ~ ~ B
2
Figure 7.2. Deformation of a knot in space 1R3 and in '?l'.
FINITE-ORDER INVARIANTS 93
at point G. The knot then becomes trivial (unknotted, and denoted F)
and continues to deform until the second catastrophic instant, when a
new double point, point 2, forms and disappears immediately, and the
un knot changes into a trefoil. This event corresponds (in the symbolic
representation inside the square) to a new crossing (F ... D ... C)
through stratum l:l' but at another location (D). Finally, a new cross-
ing through another part of the same stratum occurs (C ... B ... A),
and the result is knot A, which is in fact the trivial knot (0).
Every Vassiliev invariant assigns to each knot (and in particular to
singular knots) a certain numerical value. We will begin by giving a
simple example of a specific Vassiliev invariant, which we will call Vo.
To define it, we will assume it to be equal to zero for the trivial knot
[vo(O) = 0] and postulate that each time the moving point M (which
represents our knot) cuts across the stratum l:l in the positive direc-
tion (that of the arrows
l
issuing from l:l), the value of vo(M) increases
by 1. Thus one can easily calculate the value vo(H) of the selected
Vassiliev invariant for the figure eight knot. For that, begin at point 0
(corresponding to the unknot 0), assuming vo(O) = 0, and follow the
curve shown in Figure 7.2-0 ... A ... B ... C ... D ... F ... G ... H-by
cutting across the stratum l:l three times, once in the positive direc-
tion and twice in the negative direction, to obtain:
vo(H) = 0 + 1 - 1 - 1 = -1
Right away, a question arises. Is the invariant in question correctly de-
fined? Doesn't its value depend on the choice of the path that connects
points 0 and H? Will one get the same result if one takes, for example,
the path shown in a dotted line in the figure? Happily, the answer is
94 KNOTS
yes, both in this particular case [ Vo( H) = 0 - 1 = -1) and in the gen-
eral case. But this fundamental fact is not at all obvious, and it took all
of Vassiliev's cleverness (and very sophisticated techniques from alge-
braic topology) to prove it.
What do these calculations tell us? First, that the figure eight knot
cannot be unraveled, since its invariant is different from that of the
unknot (-I:¢: 0). On the other hand, we have seen in passing that the
trefoil is not trivial either [since vo( C) = 1 :¢: 0), and also that the fig-
ure eight knot is not equivalent to the trefoil [vo(H) = -1 :¢: 1 =
vo( C)).2
So this Vassiliev invariant does a good job of its basic task: it suc-
ceeds in distinguishing certain knots. It is, however, true that it is not a
complete invariant: it cannot tell all knots apart; for example, simple
calculations show that the value of Vo for right and left trefoils is the
same: the invariant cannot tell the difference between a trefoil and its
mirror image. But this is not the only Vassiliev invariant. There are
infinitely many of them! In particular, it is not all that difficult to find
another Vassiliev invariant that can distinguish two trefoils; for that,
we must move a little more deeply into the strata; in this case, de-
scending to the neighborhood of stratum l:2.
But before proceeding from examples to general theory, I want to
take a little rest from mathematical reasoning by making a short di-
gression about the method used here.
Digression: Mathematical Sociology
Vassiliev's approach to knots could be called sociological. Instead of
considering knots individually (as Vaughan Jones does, for example),
he considers the space of all knots (singular or not) in which knots are
FINITE-ORDER INVARIANTS 95
only points and therefore have lost their intrinsic properties. More-
over, Vassiliev does not go looking for just one invariant-he wants to
find all of them, to define the entire space of invariants. In the same
way that classical sociology makes an abstraction of the personality of
the people it studies, focusing only on their position in the social, eco-
nomic, political, or other stratification, here the mathematical sociolo-
gist focuses only on the position of the point with respect to the strati-
fication in the space gjP: gjP ::) l:o U l:, U l:2 U ...
This sociological approach in mathematics is not Vassiliev's inven-
tion. In the theory of singularities, it is due to Rene Thorn, and it
remains the weapon of choice of Vladimir Arnold and his school.
Much earlier it was used by David Hilbert to create functional analysis
(functions lose their own personality and become points in certain
linear spaces), and by Samuel Eilenberg, Saunders MacLane, Alexan-
der Grothendieck, and others, in a more striking manner, to lay the
basis of category theory. This theory was ironically called "abstract
nonsense" by more classically inclined mathematicians, perhaps to ex-
orcise it: in the beginning, it seemed ready to devour mathematics
whole. (Fortunately, it is dear today that nothing like that actually
happened.)
But let us come back to Vassiliev and his singular knots. In this spe-
cific situation, the sociological approach-which we will soon delve
into in detail-seems especially fruitful. All the information that is
needed to define the invariants of knots can be found by exploring the
strata l:" l:2, ... Like Vassiliev, we are going to try to find all of these
invariants, by always going deeper and deeper, that is, by studying the
strata l:" with greater and greater indices n. Since that requires a cer-
tain ease with mathematical reasoning, readers who do not have it can
skip directly to the conclusion of the chapter.
96 KNOTS
A Brief Description of the General Theory
To recap: a singular knot K is any smooth curve
3
in space jR3 whose
only singularities are double points (in finite number), points where a
part of the curve transversally cuts another part. Note that singular
knots, like ordinary knots (see Figure 7.3), are oriented (marked with
arrows).
For singular knots, as for ordinary knots, there is a natural relation
of equivalence called ambient isotopy: two knots (possibly singular), K,
and K
2
, are isotopic if there is a homeomorphism of jR3 (preserving the
orientation) that sends K, to K
2
, respecting the arrows (and the cyclic
order of the branches with double points). (Then the terms knot or
singular knot may stand both for a specific object and a class of iso-
topic equivalence-the reader can decide depending on the context.)
We will denote by l:o the set of (ordinary) knots and by l:n the set of
singular knots with n double points.
®@
o . . - ~
tJ
~
(a) (b)
Figure 7.3. Ordinary knots (a) and singular knots (b).
FINITE-ORDER INVARIANTS 97

Slightly displacing one of the branches of a singular knot near a
double point smoothes the singularity with two different crossings:
:x········· :X··· .. ··· .. X·· .. )··
\. <+-: .. >-+' . ./.)
'...... ' ... " ......
Recall that the smoothing to the left is called positive, the other nega-
tive.
4
We say that a function 11: ?f -+ IR is a Vassiliev invariant (in the broad
sense) if, for each double point in a singular knot, it satisfies the fol-
lowing relation:
(7.1)
which means that the function v is applied to three knots identical ev-
erywhere but inside a little ball, where the knots appear exactly as
shown in the three little dotted circles; the parts of the knot outside
the ball are not shown explicitly, but they are the same for the three
knots. The function v must be defined on the equivalence classes (ele-
ments of ?f), and thus v(K) = v(K') if K and K' belong to the same
class.
From definition 7.1 we immediately deduce the so-called one-term
relation:
and the four-term relation:
v(;"'o\) = 0
··b··
(7.2)
v(%) - v(®) + - v(®) = 0 (7.3)
98 KNOTS
Actually, deducing equation 7.2 simply requires applying definition
7.1 once
and noting that the two little loops obtained by solving the dou-
ble point can be eliminated isotopically, giving two identical knots
for which the difference between the invariants will indeed be zero.
Providing relation 7.3 requires smoothing the four off-center double
points, which gives eight terms (each with a single double point) that
neatly cancel out two by two.
We say that a function 11: 'JF -+ IR is a Vassiliev invariant of order less
than or equal to n if it satisfies relation 7.1 and vanishes on all knots
with n + 1 double points or more.
s
The set Vn of all Vassiliev invariants of order less than or equal to
n possesses an obvious vector space structure and has the inclusions
V
O
CV
t
CV
2
CV
3
•••
Lemma The value of the Vassiliev invariant of order less than or
equal to n of a singular knot with exactly n double points does
not vary when one (or several) crossings are changed to opposite
crossings.
The idea behind the proof is very simple: according to equation 7.1,
changing a crossing into an opposite crossing causes the value of the
invariant v to make a jump equal to v( [.X:;); but here this jump is
zero, since the argument of v in this case is a singular knot with n + 1
double points.
An obvious consequence of the lemma is that zero-order invariants
FINITE-ORDER INVARIANTS 99
are all constants (in other words, Vo = IR, the set of real numbers) and
thus uninteresting. Indeed, we know that all knots can be un knotted
by changing a certain number of crossings; and since these operations
do not change the value of any zero-order invariant (according to
the lemma), their value is equal to the value of the invariant of the
unknot.
It can be shown almost as easily that there are no nonzero first-or-
der invariants (in other words, Vo = VI), but, fortunately, the theory
becomes nontrivial from the second order onward. To illustrate, we
will distinguish among the elements of V
2
a specific invariant, denoted
vo, by taking it equal to 0 on the unknot and equal to 1 on the singular
knot with the following two crossings: £. Using the lemma, one
can show that Vo is well defined. The calculation, which uses definition
7.1 three times and the equality vo(O) = 0 three times, is shown in
Figure 7.4.
In fact, what we have is the same invariant Vo whose value for an-
other knot. the figure eight knot, was calculated (without the validity
of the calculation being rigorously established) at the beginning of
this chapter. But this time our calculation is quite rigorous. Readers
Figure 7.4. Calculating a second-order invariant of the trefoil.
100 KNOTS
who like these things can redo the calculation for the figure eight knot,
as well as for other knots.
Gauss Diagrams and Kontsevich's Theorem
We are now going to divest ourselves of the geometry underlying
our study of knot invariants and consider them in terms of a purely
combinatory theory.
The lemma of the previous section tells us that the value of an
nth-order invariant of a knot with n double points is unaffected by
changes in the crossings. Thus, its value does not depend on the phe-
nomenon of knotting; it depends only on the order (a combinatorial
concept!) in which the double points appear when following the curve
of the knot. We propose to code this order in the following way. Con-
sider knot K: SI ... 1R3 with n double points. Proceeding around the cir-
cle SI, we will label all the points sent to double points by the mapping
of K, then join all the pairs of labeled points sent to the same double
point by chords (Figure 7.5). The resulting configuration is called the
Gauss diagram or the chord diagram of order n of the singular knot K.
Figure 7.6 shows all the Gauss diagrams of orders n = 1, 2, 3.
Note that all the nonsingular knots have the same diagram (the cir-
cle without any cords). A good exercise for the reader who is hooked
is to draw eight singular knots for which the eight diagrams of Figure
7.6 are the corresponding Gauss diagrams (there are of course many
knots that correspond to the same diagram).
Now we will rewrite the one-term (equation 7.2) and the four-term
(equation 7.3) relations in the language of Gauss diagrams. In this no-
tation, a Gauss diagram actually stands for the value of an nth-order
invariant (always the same one in the given formula) of one of the sin-
FINITE-ORDER INVARIANTS 101
3 4
2
3
2 3
Figure 7.5. Gauss diagram of a singular knot.
Figure 7.6. Gauss diagrams of order n :s 3.
gular knots with n double points that corresponds to the diagram
(which specific knot is chosen is immaterial because of the lemma).
When there are several diagrams, we will not draw all their chords, but
we will understand that the undrawn chords are identical for all the
diagrams. In this way, we get:
0=0 flj -QJ + fb -f6 = 0 (7.4)
How are we to understand this notation? The first formula means that
the value of each nth-order invariant for a singular knot with n double
points that contains a little loop-with a double point (see equation
7.2)-is zero; thus, in this formula, I have omitted the terms 11( ••• ),
102 KNOTS
and I have not drawn the other n - 1 chords of the diagram; more-
over, it is understood that none of these chords can terminate on the
little fat arc in the diagram. Similarly, the second formula describes the
alternate sum of the values of the same nth-order invariant for four
diagrams with n chords, but I have only drawn two in each diagram;
the other n - 2 chords are exactly the same in all four. Moreover, it is
understood that these supplementary chords cannot terminate on the
little fat arc.
For example, for n = 3, the four-term relation specifically gives:
(7.5)
and since the third diagram vanishes-by virtue of the one-term rela-
tion (see the first equality in equation 7.4)-we get:
(7.6)
This relation can be envisaged as an equality in the vector space IR of
Gauss diagrams with three chords. More generally, one can consider
the vector space 211
n
= 1 R ( ~ n ) of all finite linear combinations of Gauss
diagrams 211 E ~ n ; one can then write for 211
n
all the relations that fol-
low from the one- and four-term relations and take the quotient of211
n
by these relations. We obtain a vector space, which we denote by sIn.
For example, for n = 3, one has dim 2113 = 5 (Figure 7.6), but the
one-term formula cancels the last three "basis vectors" of 2113 shown in
Figure 7.6; equation 7.6 expresses one of the two nonzero vectors re-
maining in terms of the other, so that dimsI
3
= 1. (The reader can
check that dimsI
4
= 3.)
FINITE-ORDER INVARIANTS 103
~ % % ~ % % % % % % % % % % % % %
The main result of this combinatorial theory is that the space .<A
4
completely describes nth-order Vassiliev invariants.
Kontsevich's theorem The vector space VJV" - 1 of nth-order
Vassiliev invariants is isomorphic to the space .<An of Gauss dia-
grams with n chords modulo the one-term and four-term relations.
The proof of this theorem, which is even more remarkable than the
theorem itself, is unfortunately too long and difficult to be presented
here (Bar-Natan, 1995). But from it we see that the study of the space
of nth-order Vassiliev invariants (and the determination of their di-
mensions) can be reduced to a purely combinatorial calculation. True,
this calculation is far from easy. But with the help of a supercomputer,
Dror Bar-Natan at Harvard succeeded in finding the dimensions of
the spaces .<An $I! V,,/ Vn 1 for n = 0, 1,2, ... , 9. The values of these di-
mensions are 1,0,1, 1, 3, 4, 9, 14,27, and 44, respectively.
The usefulness of this combinatorial theory (a more detailed study
can be found in CDL, 1994) is not limited to calculating the dimen-
sions of Vassiliev spaces; the theory can also be used to find the values
of specific invariants of specific knots. For example, the invariant v} E
V}, defined by the formulas V.l( 0) = 0 and V.l(@) = 1, can be
used to show that the right trefoil is not equivalent to its mirror image,
the left trefoil. I'll leave this calculation to the deft reader.
Conclusion: Why Vassiliev Invariants?
Given Jones's polynomial invariants and those of his followers, is there
really any need to invent more? Of course there is: all the polynomial
invariants known to date are not complete, which means that two
104 KNOTS
nonequivalent knots can have the same polynomial invariant. In con-
trast, for Vassiliev invariants, the following assertion holds:
Conjecture Finite-order invariants classify knots; that is, for each
pair of nonequivalent knots KI and K2 there is a natural number n
E ~ and an invariant v E Vn such that v(K
1
} ¢ v(K
2
}.
For the moment, this conjecture has neither been proved nor dis-
proved.
Another justification for Vassiliev invariants is their universality: all
the other invariants can be deduced from them. Thus, Joan Birman
and Xiao-Song Lin, of Columbia University, demonstrated that the
coefficients of Jones's and Kauffman's polynomials can be expressed in
terms of Vassiliev invariants. Similarly, but at a more elementary level,
I would suggest that readers of Chapter 5 prove, for example, that the
coefficient at X- in Conway's polynomial V(N} for any knot Nis a sec-
ond-order Vassiliev invariant.
Today there are many other examples showing that Vassiliev's
method makes it possible not only to obtain previously known in-
variants of knots, but also to define invariants-classical and new-
for many other objects (and not only knots). But this aspect of the
theory is beyond the scope of this book.
Finally-and this side of Vassiliev's approach seems the most inter-
esting to me, for it is still developing-there are obvious and natural
links (perhaps more than with the Jones and Kauffman polynomials)
with physics. But I will talk more about that in the next and conclud-
ing chapter.
KNOTS AND PHYSICS
(Xxx? . 2004?)
This last chapter differs radically from the preceding ones. Their aim
was to present the history of some of the (generally simple) basic ideas
of knot theory and to describe the varied approaches to the central
problem of the theory-that of classifying knots, most often tackled
by means of different invariants. All the cases had to do with popular-
izing research results that have taken on a definitive shape. This final
chapter, on the other hand, deals with research still ongoing, even re-
search that is only barely sketched out.
Of course, one cannot make serious predictions about future scien-
tific discoveries. But sometimes researchers working in a particular
area have a premonition of an event. In everyday language, we say of
such a situation (and usually after the fact), "The idea was in the air:'
The classical example-perhaps the most striking-is that of the inde-
pendent discovery of non-Euclidean geometry by Janos Bolyai and
Nicolai Lobachevski, which was anticipated by many others, and the
unbelievable failure of Carl Gauss, who understood everything but did
not dare.'
Is there "something in the air" today with regard to knots? It seems
105
106 KNOTS
to me that there is. I am not going to chance naming the area of math-
ematical physics where the event will occur, nor to name the future
Lobachevski, nor predict (at least in any serious way) the date of the
discovery: that is why the title of this chapter refers to Xxx with a
question mark as the future discoverer and the fictitious date of 2004
(the end of the world, according to certain "specialists").
I will return briefly to predictions at the end of the chapter. But
first, I wish to explain the sources of the remarkable symbiosis that al-
ready exists between knots and physics.
Coincidences
Connections linking knots, braids, statistical models, and quantum
physics are based on a strange coincidence among five relationships
that stem from totally distinct branches of knowledge:
• Artin's relation in braid groups (which I talked about in
Chapter 2);
• one of the fundamental relations of the Hecke operator algebra;
• Reidemeister's third move (the focus of Chapter 3);
• the classical Yang-Baxter equation (one of the principal laws gov-
erning the evolution of what physicists call statistical models,
which I talked about in Chapter 6);
• the Yang-Baxter quantum equation (which governs the behavior
of elementary particles in certain situations).
These coincidences (visible to the naked eye without having to un-
derstand the relationships listed below in detail) are shown in part in
Figure 8.1. At the left of the figure, we see the Yang- Baxter equation
KNOTS AND PHYSICS 107
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
b; i b;+1 i b;
· .
~ f < 1
1----; ~
II
II
tJ>
i ~
· .
· .
· .
b;
Figure 8.1. Three aspects of a single relationship.
RjR
j
+ IR
j
= R
j
+ IRjR
j
+ I; at the center Artin's braid group, in alge-
braic form {bibj + I bj = bj + I bib; + d and in graphical representation; and
at the right, a drawing showing Reidemeister's third move. The two
equations are in fact similar (just replace b by R, or vice versa), as are
the two drawings (look closely!).
It was while exploiting these coincidences that the New Zealander
Vaughan Jones; the Russians Vladimir Turaev, Nickolai Reshetikhin,
Oleg Viro, and Vladimir Drinfeld; the Englishman W. B. Raymond
Lickorish; the American Edward Witten; the Frenchman Pierre Vogel;
and others discovered certain (profound? fortuitous?) connections be-
tween knot theory and several branches of physics.
A strange kind of statistical model devised by Louis Kauffman en-
108 KNOTS
abled him to describe the knot invariant actually discovered earlier
by Jones-the famous Jones polynomial. Jones's original definition
(not elaborated here) was based on braids and Hecke algebras (and
thus on the coincidence between the Hecke and Artin relations). In
Kauffman's approach (a version of which is presented in Chapter 6), it
is the third Reidemeister move that plays the key role. Jones described
a version of Potts's model (a statistical model very different from
Kauffman's) based on the Yang-Baxter equation, which allowed him to
rediscover his own polynomial in another way. Using certain solutions
to the Yang-Baxter equation, Turaev discovered a whole series of poly-
nomial knot invariants.
Should I go on? Wouldn't it be more reassuring to offer an explana-
tion more specific and logical than "coincidence" for all these interdis-
ciplinary links? Unfortunately, if a specific explanation exists, I do not
know it. Yet there is indeed a more general explanation in the context
of connections between mathematics and reality.
Digression: Coincidences and Mathematical Structure
All the sciences, natural or social, have an object: they purport to de-
scribe a certain part of reality, of real life. What is the object of mathe-
matics?
The response is paradoxical: everything and nothing. "Nothing:'
because in mathematics one studies only abstractions, such as num-
bers, differential equations, polynomials, geometrical figures. Mathe-
maticians have no specific object of study in material reality.
2
"Every-
thing" because one can apply them to anything, any object that has the
same structure as the abstraction in question. I am not going to try to
KNOTS AND PHYSICS 109
explain the meaning of the expression that appears above in italics, in
the hope
3
that the reader will understand, for example (looking at Fig-
ure 8.1), that the Yang-Baxter equation has "the same structure" as the
third Reidemeister move.
A (perhaps unexpected) consequence of this state of affairs is the
importance of coincidences: if the structures of two objects happen
"accidentally" to "coincide" (even if these objects have completely dif-
ferent origins), they are described by "the same mathematics;' by the
same theory. Thus, if the trace of an operator belonging to a Hecke al-
gebra has the same properties as those of a knot invariant, why not
produce a knot invariant by using this trace (which is what Jones
did)? And if quantum particles, like knots, satisfy an equation that co-
incides with the Yang- Baxter equation, why not invent a theory of
quantum particles using knot invariants (as did Sir Michael Atiyah,
about whom I shall say more later)?
We have returned to concrete physical considerations related to
knot theory; the general digression is therefore ended.
Statistical Models and Knot Polynomials
I talked about statistical models at the beginning of Chapter 6, in par-
ticular Ising's and Potts's models. Recall that they have to do with reg-
ular structures (for example, crystals) made up of atoms (with spins,
say) that have simple local interactions (symbolized in the figures by
the line segments joining the interacting atoms). Such a system X
must have a partition function Z(X) (which is the sum over all possi-
ble states of X of certain expressions that depend on the energies oflo-
cal interactions); this function permits one to calculate the principal
110 KNOTS
global parameters of the system (temperature, total energy) and to
study phase transitions (such as passage from the liquid state to the
solid state).
In Chapter 6, we saw how a kind of partition function allowed us to
define and to calculate the Jones polynomial for knots. Actually, this
function does not correspond to any real statistical model-rather, it
is the fruit of the fertile imagination of Louis Kauffman. But it is most
surprising that there exists a real statistical model, as noticed by Jones
himself, with a genuine partition function, for directly constructing
his polynomial. My immediate aim now is to describe this construc-
tion, without going into too much detail.
Given a planar diagram of a knot (or a link), begin by drawing its
dual graph (or dual statistical model of the knot), as shown in Figure
8.2; you do that by alternately painting in black and white the parts
of the plane delimited by the knot projection (taking care that the
outside part be white), take the black regions for the vertices of the
graph (or the atoms of the model) and join two vertices by a line or an
edge (the interaction) if the black regions possess a common crossing.
Moreover, declare the edges (interactions) to be positive or negative
according to a convention that the reader can identify by looking at
Figure 8.2.
Next, define the state of the system as an arbitrary function that as-
signs a spin to each atom, the spin taking only two values, which phys-
icists call up and down. When the model is in a well-determined state s
E S (S denotes the set of all possible states), the (local) interaction en-
ergy E[s( VI), s( V2)] of the two atom-vertices joined by the edge [VI' V2]
is assumed to be equal to ± 1 if they have the same spin and to a:!: I if
the spins are opposite; choose the plus sign or the minus sign depend-
KNOTS AND PHYSICS 111
Figure 8.2. Dual graph of a knot.
ing on whether the edge (interaction) is positive or negative; here, a is
the name of the variable of the polynomial (in a and a-I) that you
want to obtain. (It is this specific choice of the interaction energy of
the atoms that is unique to Potts's model, the model of the phase tran-
sitions between water and ice.)
That done, we can define the partition function of the model by the
formula:
Z(K) = ( ~ )In E[s(vj ),s(v j)]
sES (v,.v, lEA
where A is the set of all the edges.
Deriving Jones's polynomial from this partition function requires
applying to it a variation of "Kauffman's trick" (described in detail in
Chapter 6).4
Thus we see that Potts's model for freezing water brings us rather
112 KNOTS
easily to the most famous invariant of knots. In analyzing this con-
struction in surveys or popular articles, mathematicians have a ten-
dency to enthuse over «the application of knot theory to statistical
physics." Curious analysis! Knot theory has nothing to contribute to
physics here-on the contrary, it is statistical physics that produces a
construction that can be applied to mathematics. (To spare the self-es-
teem of mathematicians, recall that Jones's original construction-
purely mathematical-preceded the "physical" construction just de-
scribed.)
Of course, what is important here is not the rivalry between physi-
cists and mathematicians, but this unexpected coincidence between
two areas of knowledge, which are, a priori, very far apart. Let us move
on to another coincidence, one that really deals with an application of
knot theory to physics.
Kauffman's Bracket and Quantum Fields
I described Kauffman's bracket in Chapter 6, where it was used to de-
fine Jones's polynomial, the most famous invariant of knots. We are
going to see that it can be used for something else entirely.
Recall that this bracket associates to every planar knot diagram K a
polynomial (K) in a and a-I, defined by a precise formula inspired by
the partition functions of statistical models. It has already been noted
that this equation (which we have no need of here) has no physical in-
terpretation, at least in the framework of a realistic statistical model. It
is in another branch of physics-topological quantum field theory-
that it plays a role.
KNOTS AND PHYSICS 113
This theory, generally designated by the acronym TQFT, seeks to
formalize the quantum version of classical field theory (gravitational
force fields, electromagnetic fields, and so on) in the most general con-
text, the topological one. In this context, the physical magnitudes that
one studies-the observable5-Cannot depend at all on the system of
coordinates under consideration; they must conserve the same values
for every topological transformation of the coordinates. They are thus
topological invariants, like knot invariants.
It was Witten's idea to use a generalization of the Jones polynomial
(often called
5
the /ones- Witten invariant); he deserves the credit for
finding it (and it won him the prestigious Fields medal) and for using
it to construct a TQFT. This TQFT was simply a model whose dimen-
sion is 2 + I, where 2 is the dimension of "space" and 1 that of "time;'
these three coordinates being mixed, as relativity demands. The model
is thus a three-dimensional one that may contain knots, which in this
context physicists call Wilson lines.
Later on, Michael Atiyah (also a Fields medal winner, but for earlier
work) rethought Witten's model from a mathematical viewpoint and
generalized it to create an axiomatic theory of TQFTs. Specifying this
theory, Vogel and his coauthors constructed a whole series of exam-
ples of TQFTs, in which Kauffman's bracket actually plays the key role.
There is no question here of explaining this theory and these exam-
ples-the mathematics required are too sophisticated. I will restrict
myself to the context that features the bracket.
In this context, instead of a plane, think of a surface with a bound-
ary, on which the diagram of a knot (or a link) is drawn. The strands
of the knot may have ends at the boundary of the surface; a typical ex-
114 KNOTS
ample is shown in Figure 8.3. Each of these diagrams is associated
with a polynomial in a and a-I that satisfies two very simple rules (al-
ready seen in Chapter 6):
(KU O) = (-a
2
- a-
Z
) (K)
Readers who recall this chapter will immediately recognize two funda-
mental properties of Kauffman's bracket. Note, to be done with coin-
cidences, that a special case of this construction (when the surface is a
disk) gives the so-called Temperley-Lieb algebra, an algebra of opera-
tors that satisfies the rules of Artin, Yang, Baxter, Reidemeister, Heeke,
and so on.
I do not want to judge the interest of these TQFT models from the
point of view of physical reality. Physicists take them very seriously,
Figure 8.3. Diagram oflinks on a bounded surface.
KNOTS AND PHYSICS 115
but perhaps not as seriously as the (mathematical) idea of the quan-
tum group, the study of which (in the context of its connection to
knots) we will move on to next.
Quantum Groups as Machines for Making Invariants
Quantum groups appeared twenty years ago and today are an object
of intense study by both mathematicians and physicists. Their formal
definition, however, has little appeal: a quantum group is a set of ab-
stract elements that must satisfy a whole list of formal algebraic axi-
oms whose real meaning is not very obvious.
Rather than trying to explain quantum groups in detail, I will focus
on their physical significance. First off, note that, despite their name,
quantum groups are not groups at all; they are algebras, and even
"bialgebras." That means that two operations are given for any set Q: a
multiplication and a comultiplication. A multiplication, of course, as-
sociates to each pair of elements a well-defined element from Q-
their product. The comultiplication does the opposite: it associates a
pair of elements
6
from this set with a single element from Q-its
coproduct. From the point of view of physics, these two operations
correspond respectively to the fusion of two particles into a single one,
and to the splitting of a single particle into two. I have tried to repre-
sent this correspondence graphically in Figure 8.4.
The operations (of multiplication and comultiplication) must sat-
isfy some very obvious axioms (such as associativity) that endow Q
with what mathematicians call a bialgebra structure.
7
These axioms
are not all that restrictive, and there are so many quantum groups
that one is led to consider a narrower class, for example, the class of
116 KNOTS
~ ,.0 0 .... ). .. /
~ .... ~ ~
Figure 8.4. Product and coproduct of two particles.
quasi-triangular quantum groups defined by Drinfeld (another Fields
medalist!). The axiom of quasi-triangularity implies that the Yang-
Baxter equation holds for this class, and this, the reader will of course
have guessed, provides the link between quasi-triangular quantum
groups and knots. More precisely, the representations of these quan-
tum groups allow us to define a lot of invariants, both new and famil-
iar, one after the other. Quantum groups, as it were, are the truly scien-
tific way to mass-produce knot invariants.
Vassiliev Invariants and Physics
As we saw in the preceding chapter, Vassiliev invariants are obtained
by applying a very general construction, ideologically close to catas-
trophe theory, to knots. Can one give a fundamental physical meaning
to the flip (the principal catastrophe, in the course of which the lower
strand of a knot breaks through the upper strand, ending up on top)?
No, it seems, at least not in any obvious way. The underlying physics is
not evident in Vassiliev's work; it is buried in the algebraic and combi-
natorial structures characterizing invariants.
The thing is that the set V of invariants (which is in fact a vector
KNOTS AND PHYSICS 117
% ~ % ~ ~ % ~ % % ~ ~ % % % % % ~ %
space) is not only endowed with a multiplication (obtained by multi-
plying the values of invariants, which are ordinary numbers), but also
a comultiplication: Ll: V -+ V ® V. The latter is defined by means of
the connected sum # of two knots via the following obvious formula:
It is easy to see that these two operations make a bialgebra out of V.
Thus, right from the start, this "very physical" structure (fusing and
splitting of particles) obviates the need to go looking "outside" for an-
other algebraic object (such as the quasi-triangular quantum group
for Jones-Witten-type invariants) to make "physical" invariants. This
bialgebraic structure is inherent in Vassiliev invariants.
But there is more. First, at the analytical level, the Vassiliev invari-
ants of a knot can be expressed via the admirable Kontsevich integral.
In a certain sense, it is a generalization of the Gauss integral in electro-
magnetism, and thus should have a physical interpretation. What in-
terpretation? Nobody knows.
Next, at the combinatorial level, the interpretation of Vassiliev in-
variants by chord diagrams (which I talked about in the previous
chapter), another contribution from Maxim Kontsevich (yet another
Fields medalist!), also lends algebra a physical orientation, several
even. In particular, the algebra of Chinese characters (which until re-
cently was called "the algebra of Feynman diagrams") is dose to physi-
cal theory, as its (former) name indicates. But there, too, we are still at
the stage of hope and speculation.
A final important point, also still not understood, is the four-term
relation that I spoke about in Chapter 7. Dror Bar-Natan has exploited
the fact that this formula is none other than a form of the classical
118 KNOTS
Jacobi identity to construct Vassiliev invariants using representations
of Lie algebras. Will this coincidence between fundamental mathe-
matical relations have a "physical" development?
Conclusion: Nothing Is Finished
At the beginning of this book, we saw how William Thomson's idea
of using the knot to make a model of the atom almost a century
and a half ago was the start of the theory of knots. Very recently,
knot invariants, in particular Kauffman's bracket, became the basis for
physics-oriented theories, such as topological quantum field theory.
Where are we now?
Thomson's idea was ephemeral. From the viewpoint of actual phys-
ical reality, the significance of TQFTs (in the style of Witten, Atiyah,
Vogal, Crane, Yetter) remains unclear, to say the least. Will interest in
the connections between physics and knots be short-lived?
For specialists in knot theory, there is still a lot to do. For example,
there is still no unknotting algorithm simple and efficient enough to
be taught to a computer, and many other important questions have
been shelved. For researchers in mathematical physics who look at
knots from the side, many areas remain unexplored, particularly con-
cerning Vassiliev invariants.
Finally, do not forget that aside from classical knots (three-
dimensional curves in space), there are some less-studied "general-
ized knots;' such as (two-dimensional) spheres in four-dimensional
space (or surfaces, more generally speaking). According to Einstein,
we live in four-dimensional space-time. According to specialists in
string theory, the propagation of a particle can be modeled by sur-
KNOTS AND PHYSICS 119
~ % ~ % ~ % % % ~ % % % % % % ~ % %
faces. Is a quantum theory of gravitation hiding in there somewhere?
Do Vassiliev invariants (which also must exist in this context) have a
real physical interpretation?
Research always begins with a question, and hope. To conclude, I
hope that the reader (and myself as well!) will again experience, in the
context of knot theory, the incomparable joy of understanding in-
spired by a great discovery.
NOTES
Preface
1. For those who know, he is indeed the Vandermonde of the determinant.
2. Another strategy, simpler and perhaps no less efficient, is to thumb the
book and choose which chapter to read according to the most interesting
pictures.
3. The first definition of a polynomial invariant for knots, due to Alexander,
was based on mathematical ideas that were very sophisticated for the time:
homology theory and covering spaces.
4. In this case the physical interpretation uses the ideas of statistical physics.
5. Here I mean precise mathematical terms and not descriptions of TV horror
films. Modern mathematical terminology, like that of theoretical physics,
tends to favor everyday words over serious, scientific-sounding words.
1. Atoms and Knots
1. Differential geometry defines knots more neatly as "smooth closed curves;'
but this involves using the calculus.
2. Antoine is French for Antony.
2. Braided Knots
1. This terminology derives from what geometers call an inversion (in our
case, a symmetry with respect to a small circle whose center is situated in
121
122 NOTES
one of the regions bounded by the given Seifert circle). This inversion
sends the center "to infinity" (and transforms the region into an infinite re-
gion).
2. This knot is numbered 52 in the tables of knots (Figure 1.6).
3. So called because they show up in all groups, not only in the group of
braids Bn.
4. Among whom are certainly A. A. Markov, Joan Birman, and perhaps also
William Thurston. And Emil Artin himself.
3. Planar Diagrams of Knots
1. Especially in "singularity theory" (often also called "catastrophe theory"),
the basics of which were set down by Hassler Whitney between the First
and Second World Wars and subsequently developed by Rene Thom, Vladi-
mir Arnold, and their followers.
2. To the mathematically knowledgeable reader I should note that for a knot
represented by a differentiable curve, proof of the analogous assertion is
much more difficult and requires special techniques.
3. The reader may wonder how a computer can "see" knots. In fact, there
are several effective ways of "coding" knots. For example, the one I use
in my unknotting software gives the following description of the left trefoil:
1 + -2 - -3 + -1 - -2 + -3 - -1. My computer understands it. Can
the clever reader guess the principle behind the coding?
4. Perhaps I need to say here that for a projection of a given knot, the number
of different applications of (.h is finite. Moreover, be careful that the soft-
ware doesn't make the mistake of following a specific application of "h by
the inverse of the same application. Otherwise, the software may trigger a
useless and infinite loop (in the computational sense of the term).
5. Actually, recent work by three American mathematicians (Joel Hass, Jeffrey
Lagarias, Nicholas Pippinger) implies that the number of Reidemeister
moves needed for unknotting is bounded, which means that in principle
this algorithm either unknots a knot or declares that it cannot be unraveled
NOTES 123
~ ~ ~ % % % % ~ % ~ %
(since it hasn't succeeded in unraveling within the prescribed bound for the
number of moves). Unfortunately, the bound is huge, and hence this argu-
ment is important only from the theoretical viewpoint. For practical pur-
poses, however, a fast and usually effective unknotting software has recently
been devised by the Russian I. Dynnikov.
4. The Arithmetic of Knots
1. Knots, remember, were defined as closed curves in three-dimensional
space.
2. The pedantic reader will say that the box obtained is no longer cubic, and
thus that the composite knot is not a true boxed knot. That is correct. By
way of reward, let us leave to him or her the task of modifying the defini-
tion so that it will not be open to criticism; in particular, she/he should
redefine the equivalence (isotopy) of knots and say that the knot, or the
type of knot, is an equivalence class.
3. This idea will certainly occur to the reader who has assimilated the chapter
on braids, for which this construction works perfectly well.
4. For the reader more familiar with mathematics, note that adopting the ar-
gument represented in Figure 4.5 specifically requires using another defini-
tion of knot equivalence, one that is based on the notion of
homeomorphism.
5. It would be safer to say that we do not know whether an appropriate opera-
tion of addition exists for knots, since we know only that those who have
searched for it have not managed to find one. We can, however, say with
confidence that, if it exists, the geometric addition of knots is not simple-
if it were, someone would have discovered it.
S. Surgery and Invariants
1. An angstrom (A) is one ten-billionth of a meter.
2. This acronym is a flagrant injustice to two Polish mathematicians, Jozef
124 NOTES
Przytycki and Pawel Traczyk, who made the same discovery at the same
time but published later, not to mention many Russians who didn't publish
because they considered the polynomial to be just a variant of the Jones
polynomial. Another acronym, Lymphotu, was proposed later by Dror Bar-
Natan. He gave credit to the Poles and to others (U = unknown discover-
ers), but the acronym didn't stick.
6. Jones's Polynomial and Spin Models
1. Or "intrinsic angular moment."
2. At issue here is a very theoretical model: "two-dimensional water." Of
course, there is a more realistic three-dimensional model. We use "flat wa-
ter" not only to simplify the drawing but also because it is the model most
studied by physicists. But especially-as we will see later-we use it because
the two-dimensional model relates to knots.
3. Generally, this polynomial can contain negative powers of a. Mathemati-
cians call it a Laurent polynomial.
4. Strictly speaking, X(·} is not the Jones polynomial; to obtain it requires
changing the variable of the polynomial by writing q = a
4
, but this is essen-
tially only a change in notation.
7. Finite-Order Invariants
1. One chooses the direction of the arrows, called coorientation, in such a way
that the follOwing catastrophes (cutting across II)
are positive and negative, respectively.
2. From a mathematical point of view, however, it must be emphasized that
NOTES 125
we have not shown anything at all, and not only because we have not
proved that Vo is well defined. The problem is that our argumentation is
based on the configuration of strata as it appears in Figure 7.2, but we
know nothing of their actual configuration. Mathematized readers desiring
a rigorous account of the preceding calculations will find one a little fur-
ther on in this chapter. The others will have to take my word for it: the rig-
orous version of this calculation is really very simple, at least for anyone
used to reasoning mathematically.
3. That is, every smooth mapping K of the circle 51 into Euclidean space 1R3.
4. The positive (or negative, respectively) resolution is well defined: it is the
one for which the traveller moving along the upper branch (following the
arrow) sees the arrow of the lower branch pointing to the left (or right, re-
spectively).
5. Vassiliev invariants are frequently also called finite-order invariants and
sometimes Gusarov-Vassiliev invariants, because they were discovered inde-
pendently by M. N. Gusarov from St. Petersburg (but he published his re-
sults only much later).
8. Knots and Physics
1. Gauss discovered the first principles of hyperbolic non-Euclidean geome-
try well in advance of Lobachevski and Bolyai but lacked the courage
to publish this scandalous theory. (Lobachevski, who did publish it,
quickly became the laughing stock of his contemporaries. And lack of rec-
ognition drove Bolyai to drink.) When Lobachevski's publication came
out, Gauss had already created the differential geometry of surfaces, on
which modelling hyperbolic geometry is mere child's play for a profes-
sional of Gauss's level. How could the brilliant Gauss, with all the tools at
hand, have missed the discovery, as though he had suddenly been struck
blind?
2. Barring the conviction-from a strictly Platonic point of view-that ab-
126 NOTES
stractions, which belong to the world of ideas, are more real than the mate-
rial world.
3. A famous attempt was made by Nicolas Bourbaki; I have nothing to say
about it.
4. Actually, it is easy to show that the polynomial Z(K) is invariant relative to
the second and third Reidemeister moves, whereas the first move gives a
superfluous factor that can be got rid of owing to another factor dependent
on the writhe, exactly as shown in Chapter 6.
5. Unfairly. Witten's construction lacked mathematical rigor. It was N.
Reshetikhin and V. Turaev who succeeded (thanks to another approach) in
providing a correct mathematical definition for these invariants. That is
why I prefer the expression Jones-Reshetikhin-Turaev-Witten invariants,
despite its excessive length.
6. More precisely, a linear combination of pairs of elements.
7. Or a Hopf algebra structure.
WORKS CITED
Adams, C. 1994. The Knot Book. New York: Freeman.
Ashley, C. W. 1944. The Ashley Book of Knots. New York: Doubleday.
Bar-Natan, D. 1995. "On the Vassiliev Knot Invariants." Topology, 34:423-472.
CDL [Chmutov, S. V., S. V. Duzhin, and S. K. Lando.]. 1994. "Vassiliev Knot
Invariants, I, II, III." In Advances in Soviet Mathematics, vol. 21, Singularities
and Curves, ed. V. I. Arnold, pp. 117-126.
Dehornoy, P. 1997. ''rart de tresser:' Pour la science. Special issue, pp. 68-74.
Haken, W. 1961. "Theorie der Normalflachen." Acta Mathematica, 105:245-375.
Jaworski, J., and I. Stewart. 1976. Get Knotted. London: Pan Books.
Jensen, D. 1966. "The Hagfish." Scientific American, 214:82-90.
Mercat, C. 1996. "Theorie des n<:euds et enluminures celtes:' L'Ouvert, no. 84.
Prasolov, v., and A. Sossinsky. 1997. Knots, Links, Braids, and 3-Manifolds. Prov-
idence: American Mathematical Society.
Rouse Ball, W. W. 1971. Fun with String Figures. New York: Dover.
Stewart, I. 1989. "Le polynome de Jones:' Pour la science, 146:94.
Thomson, W. 1867. "Hydrodynamics." Proceedings of the Royal Society of Edin-
burgh,6:94-105.
Walker, J. 1985. "Cat's Cradles and Other Topologies Armed with a Two-Meter
Loop of Flexible String." Scientific American, 252:138-143.
Wang, J. C. 1994. "Appendix. I: An Introduction to DNA Supercoiling and DNA
Topoisomerase-catalyzed Linking Number Changes of Supercoiled DNA."
Advances in Pharmacology, 29B:257-270.
127

ALEXEI SOSSINSKY
TRANSLATE D BY C ISH LE WEI S,.

KNOTS
MATHEMATICS WITH A TWIST
HARVARD UNIVERSITY PRESS
CAMBRIDGE, MASSACHUSETTS LONDON, ENGLAND

2002

Copyright © 2002 by the President and Fellows of Harvard College
ALL RIGHTS RESERVED

Printed in the United States of America Originally published as Nreuds: Genese d'une Theorie Mathematique © Editions du Sew.!. 1999 Illustrations by Margaret C Nelson Design by Marianne Perlak

Library of Congress Cataloging-in-Publication Data Sossinsky. A.B. (Na:uds. English] Knots mathematics with a twist I Alexei Sossinsky ; translated by Giselle Weiss. p. em. Includes bibliographical references ISBN 0-674-00944-4 (alk. paper) 1. Knot theory. 2. Low-dimensional topology. I. Title.
QA612 2 S6713 2002 514' 224-dc21 2002027295

KNOTS AND PHYSICS Xxx? • 2004? NOTES 121 WORKS CITED 127 73 90 105 . ATOMS AND KNOTS 1 Lord Kelvin • 1860 2. FINITE-ORDER INVARIANTS Vassi/iev • 1990 8. JONES'S POLYNOMIAL AND SPIN MODELS Kauffman • 1987 7.CONTENTS PREFACE vii 1. SURGERY AND INVARIANTS 58 Conway· 1973 6. THE ARITHMETIC OF KNOTS 46 Schubert· 1949 5. BRAIDED KNOTS Alexander • 1923 15 3. PLANAR DIAGRAMS OF KNOTS 35 Reidemeister • 1928 4.

.

and physicists. knot theory was regarded as just one of the branches of topology: important. A tentative effort by Alexandre-Theophile Vandermonde at the end of the eighteenth century was short-lived. The French "nouveaux philosophes" (not so new anymore) and post modernists even talk about knots on television.! and a preliminary study by the young Karl Friedrich Gauss was no more successful. Why the interest? Is it a passing fancy or the provocative beginning of a theory as important as relativity or quantum physics? vii . vipers' tangle-knots are familiar objects. hangman's knot. of course. The expressions "quantum group" and "knot polynomial" are used indiscriminately by people with little scientific expertise. Only in the twentieth century did mathematicians apply themselves seriously to the study of knots. Today. Knots-or more accurately. with their typical nerve and incompetence. Knots are trendy. mathematical theories of knots-concern biologists. symbols of complexity.PREFACE Butterfly knot. clover hitch knot. chemists. they were long ignored by mathematicians. but not very interesting to anyone outside a small circle of specialists (particularly Germans and Americans). occasionally metaphors for evil. all that has changed. But until the mid-1980s. Gordian knot. For reasons I do not entirely understand.

This last group of readers suffers from memories of daunting and useless "algebraic expressions. More specifically. dazzling advances. is as emotionally rich as the history of painting or poetry. But mathematics was a vibrant field of inquiry before lackluster teaching reduced it to pseudoscientific namby-pamby. Rather. moreover. and lifeless definitions of geometric entities. and others. young people who like mathematics. usually calls for mathematical sophistication. But it so happens that the mathematical theory of knots-the subject of this book-is an exception to the rule." tautological arguments concerning abstractions of dubious interest.viii PREFACE This book addresses this question. with its sudden brainstorms. when it is not reduced to simple anecdotes. at least to some extent. And the story of its development. more numerous. often imbued with mystery and sometimes with striking and unexpected beauty. It doesn't necessarily take a graduate of an elite math department to understand it. This book is intended for three groups of readers: those with a solid scientific background. it presents specific information about a subject that is difficult to grasp and that. and dramatic failures. who feel they have no aptitude for math as a result of their experience in school but whose natural curiosity remains intact. The hitch is that understanding this history. but its aim is certainly not to provide peremptory answers to global inquiries. crops up in many guises. the reader will see that the only mathematics in this book are simple calculations with polynomials and transformations of little diagrams like these: .

Before I review the topics taken up in each chapter. the starting point will be an original idea. For each topic. the socalled fundamental group. M. Dehn).PREFACE . We will then follow the path of his thinking and that of his followers. knots were associated with a variety of useful activities. Nielsen. Seifert. profound. But I have striven to minimize cross-references (even if it means repeating certain passages). but talking a little about their practical charms will make it easier to glimpse the beauty of the theory. Accordingly. on the contrary. My desire to avoid overly abstract and technically difficult mathematics led me to leave out completely the most classical tool of the theory of knots (and the most efficient at the early stages). in an attempt to understand the major implications of the topic for contemporary science. fiddle with strings and make actual knots. Of course. W. the chapters are ordered more or less chronologically. without going into technical details. and the American J. those activities are not the subject of this book. H. I have not tried to provide a systematic and unified exposition of the theory of knots. G. Given the diversity of the topics tackled in this book. it is worth mentioning that prior to becoming the object of a theory. so that the chapters can be read in whatever order the reader chooses. Van Kampen. as a rule simple. Alexander-were based on the judicious use of this tool. and unexpected. which are almost entirely independent of each other. the work of a particular researcher.x Readers will also have to draw on their intuition of space or. failing that. Their work will barely be mentioned here. various topics are scattered throughout the chapters. H. The first successes with the theory-those of the mathematicians of the German school (N. the Dane J. .

Sailors use knots not only to moor boats. passing from generation to generation (see Adams. the alpinist's chair knot (d). This worksaving device. oral transmission of maritime knot making was supplanted by specialized books about knots. it was even the subject of a detailed article in Diderot's and d'Alembert's Enclopedie. And there are many other knots adapted to these special tasks (see Figure P. To tie a rope to a rigid pole (a mooring or a mast). the engineer's constrictor knot (c). and the knitter's rice stitch (e) are only a few examples among many. rig sails.3a). For each specific task. and the best knots survived.2). to tie two ropes together. In the Age of Enlightenment (in England even earlier).x PREFACE Since Antiquity. One of the first authors in this genre was the Englishman John Smith. the development of knot making was motivated by practical needs. the square knot (d) or the fisherman's knot (f) (when they are the same size) or the sheet bend (e) (when one rope is thicker than the other). The classic reference for knot making is Ashley's famous Book of Knots (1944). one uses the clover hitch knot (see Figure P. unites two major . Sailors were not the only inventors of knots.la). sailors invented an appropriate knot. the terminology associated with knots became codified. especially those of sailors and builders. The fisherman's hook knot (Figure P. a sort of Archimedes' lever with ropes. together with the compound pulley (b and c). A few knots in particular derive from one of the greatest technological inventions of the Middle Ages: the pulley (Figure P.2b). or the camel knot (c). but also to make objects as varied as the regrettably famous "cat 0' nine tails" and straw mats woven in Turk's head knots (Figure P. At the same time. much better known for his romantic adventures with the beautiful Indian princess Pocahontas. the rolling hitch knot (b). 1994).2f). and hoist loads.

Some sailors' knots.PREFACE xi Figure P.l . (f) .2. O~ (b) (d) (e) Figure P. Other knots.

4).3. It is used to pull or to lift all kinds of loads. which in turn were braided in a specific way (generally involving three lines) to make a rope (see Figure P. the rope became the universal technological tool of the age. The procedure for making cables is more com- . but synthetic in our times) had to be twisted into threads that were then braided into thicker strands. Fibers (once made of plants such as hemp. Pullies and a hoist. called lines. The technology for producing ropes (and cables) themselvesbraiding-became very important. Thanks to knots. inventions of Antiquity: the wheel and the rope.xii PREFACE (a) Figure P. usually also attached with the help of suitable knots.

PREFACE XIII Fibers Strand (l ine) Figure P.4. as shown schematically in Figure P. I have in mind the remarkable representations of knots on the megaliths and burial stones engraved by Neolithic peoples. in particular the Celts. and braided ropes. 1996). . plex and involves four (or more) levels of cords. but the geometric technology (based on regular figures) used to create these bewitching designs has been decoded by mathematicians (see Mercat. it is precisely this feature of knots that is responsible for their first traces in our civilization. Actually. mysterious. the technology of braiding is the model for a basic idea in topology (as well as in mechanics)-the braid-which we will discuss in detail in Chapter 2. and magical aspect. Anatomy of a rope. Utilitarian and technological considerations aside. We do not know the mystical and religious meaning of the links represented on menhirs (upright monuments also known as standing stones). lines.C. knots also have an aesthetic. during the fourth century B.5. As far as I know. For the mathematician. these are chains of knots connected to one another (mathematicians call them links).

in illuminated manuscripts. . Neolithic peoples were not alone in using links to decorate their objects of worship. Links are also found in the Middle Ages.S.XIV PREFACE Figure P. Links on a megalith. in the architecture of certain Eastern civilizations (friezes and other ornaments of the famous Alhambra palace in Spain are examples). and in the decorative elements framing icons in orthodox churches in northern Russia.

. The chapter not only deals with this spectacular failure of a beautiful physical theory. models he dubbed "vortex atoms:' To study the theory of matter from this point of view. which was not the work of mathematicians-what a shame for them!-but that of physicists. Kelvin's theory ran aground and was soon forgotten. to give a brief idea of what is to come and to allow those who don't intend to read the book from beginning to end to choose which chapters they wish to take in. it also reviews various aspects of knot theory: Tail's tables of alternating knots. blind geometers. from a mathematical vantage point. the superb wild knots. The starting point (dating from around 1860) was Thomson's idea of using knots as models for the atom. Fortunately for the selfesteem of mathematicians. and Antoine's necklace. more precisely.) The first chapter has to do with the beginnings of the mathematical theory of knots. he had to begin with knots. 1997. 1985). . The chapter ends with a brief discussion of the reasons for the failure of Thomson's theory. 2 (Remember that the chapters are relatively independent. which physicists were unable to solve at the time but mathematicians took care of a century later. ropes that come undone instead of strangling the sexy magician's assistant.. William Thomson (alias Lord Kelvin). Some of these tricks (which amateur magicians can do) are described.PREFACE xv To end this overview of knots on a lighter note. but not without leaving to posterity a series of problems (the Tait conjectures). and so on. Walker. This last object provides us with an opportunity to talk about . think of the essential role they play in the magician's arsenal: knots that aren't. Let us move on to a summary of this book. elsewhere (Prasolov and Sossinsky.

W. Alexander a half-century after Kelvin's abortive start. which was formulated about the same time by the young German researcher Emil Artin. encoding of knots. too. In Chapter 3. is more algebraic (and consequently simpler and more efficient) than knot theory. the functioning of an unknotting algorithm (which is fairly simple but has the disadvantage of futility when it comes to trying to unknot non-unknottable knots) will be explained: there. though it is very complex. also true in real life). Finally. This gives us a chance to talk a little about catastrophe theory. The mathematical theory of braids. We will see that an algorithm invented by Reidemeister's compatriot Wolfgang Haken to determine whether a given knot can be untied does indeed exist. I present a clever but simple geometric construction by the German mathematician Kurt Reidemeister. whose principal theorem (the existence and uniqueness of prime knots) was demonstrated in . but they gave nice results. The connection in question (a geometric construction of childlike simplicity: the so-called closure of braids) enables one to obtain all knots from braids-Alexander's result. which reduces the study of knots in space to their planar projections (called knot diagrams). Efforts in this direction were unsuccessful. and working with knots on the computer. H. Chapter 4 reviews the arithmetic of knots. And because Artin rapidly established the classification of braids. among which are the algorithms and software recently devised by French researchers.XVI PREFACE The second chapter deals with the fundamental connection between knots and braids discovered by the American J. That is because untying a knot often means first making it more complicated (alas. it was natural to try to deduce the classification of knots from it. the modern computer does a better job of unknotting than we poor Homo sapiens.

These calculations provide a very easy and fairly efficient way to show that two knots are not of the same type. whose original definition was far from elementary. But this method is probably not what the reader of this chapter will find most interesting: a biological digression explains how topoisomerases (recently discovered specialized enzymes) actually carry out Conway operations at the molecular level. the Jones polynomial. This description is based on a tool-the Kauffman .PREFACE xvii 1949 by the German Horst Schubert. Conway operations can change not only the appearance but also the type of the knot. one of the most original mathematicians of the twentieth century. in an elementary way. they can even transform knots into links. As in Chapter 3. Oddly. Chapter 6 presents the most famous of the knot invariants. the so-called Alexander-Conway polynomiaP of a knot (or link). Chapter 5 brings us to an invention that seems trite at first. we will be dealing with small geometric operations carried out on knot diagrams. They make it possible to define and to calculate. it is the physical interpretation4 of the Jones polynomial that gives a very simple description of the Jones invariants. and in particular that some knots cannot be unknotted. It is due to the Anglo-American John Conway. In particular. The curious resemblance between knots equipped with the composition operation (placing knots end to end) and positive integers (with the ordinary product operation) excited all sorts of hopes: Could knots turn out to be no more than a geometric coding of numbers? Could the classification of knots be just a plain enumeration? In Chapter 4 I explain why such hopes were unfulfilled. Contrary to Reidemeister moves. it allowed several researchers to establish the first serious connections between this theory and physics. which gave new life to the theory fifteen years ago.

Contrary to what I tried to do in the other chapters. over 130 years ago.xviii PREFACE bracket-that is very simple but that plays no less fundamental a role in modern theoretical physics. Chapter 7 is devoted to the last great invention of knot theory. Vassiliev invariants. and on the dramatic and emotional side of contemporary research. In one of them. the original definition. readers will learn that the main ingredient in the Kauffman bracket was already known in the Neolithic age by the Celtic artists mentioned earlier. The eighth and final chapter discusses connections between knot theory and physics. But I am convinced that even readers closer to the humanities than to the sciences will succeed in getting through this chapter. Instead of complicated mathematical formulas. Here. they can focus on the gist of the discussion. Even if they cannot grasp the precise meaning of the terms and equations. I had to use some new technical terms from mathematical physics without being able to explain them properly. This chapter contains several digressions. The brilliant beginnings of knot theory. on the role of coincidences. which drew on catastrophe theory and spectral sequences. but an elementary description is proposed. too. were marked by a ringing failure-as a physical theory of matter-but the concepts were revived thanks to the repeated efforts of mathemati- .5 was very sophisticated. here I could only sketch out the most rudimentary explanations of what is going on in this area. readers will find abbreviated calculations involving little diagrams. along with a digression on the sociological approach to mathematics.

making the final goal ever more attractive. . the basic role knots are supposed to play in physics has not yet been specified in a convincing way. Finally. and still no one knows whether they possess a complete system of invariants that would be easy to calculate. springing from the imagination of the best researchers. whose only motivation was intellectual curiosity. concrete ideas.PREFACE XIX cians. and we find ourselves at a higher level. But every failure made it easier to grasp the remaining problems. Today we are in a situation similar to that of 1860: some researchers think. Progress required new. that knots playa key role in the basic theory of the structure of matter. But that is not to say that we are back at the beginning: the spiral of knowledge has made a full loop. as William Thomson did. Its major problems are still unsolved: knots continue to elude efforts to classify them effectively. The theory of knots remains just as mysterious and vibrant as ever. often sparking exaggerated hopes. And the ideas came.

.

KNOTS .

.

I ATOMS AND KNOTS (Lord Kelvin ·1860) In 1860. According to Thomson. and those who felt matter to be a superposition of waves dispersed in space-time.1). rigid little bodies that occupy a precise position in space. propagates as a narrow beam bending sharply back on itself-like a snake biting its tail. instead of dispersing in all directions." however. Thomson was looking for a way to combine them. according to which matter is constituted of atoms. His peers were divided into two enemy camps: those who supported the so-called corpuscular theory. And he found one. 1867). thus forming a knot (Figure 1. Thus an atom is like a wave that. the English physicist William Thomson (better known today by the name of Lord Kelvin. The type of knot would then determine the physicochemical properties of the atom. but at the time not yet graced with a noble title) was pondering the fundamental problems linked to the structure of matter. are not pointlike objects but little knots (see Thomson. matter is indeed constituted of atoms. These "vortex atoms. But this snake could wriggle in a fairly complicated way before biting itself. According to this 1 . Each of these theories provides convincing explanations for certain phenomena but is inadequate for others.

2

KNOTS

view, molecules are constituted of several intertwined vortex atoms, that is, they are modeled on what mathematicians call links: a set of curves in space that can knot up individually as well as with each other. This theory will no doubt seem rather fanciful to the reader accustomed to Niels Bohr's planetary model of the atom taught in school. But we are in 1860, the future Nobel laureate will not be born until 25 years later, and the scientific community is taking Thomson's revolutionary idea seriously. The greatest physicist of the period, James Clerk Maxwell, whose famous equations formed the basis of wave theory, hesitated at first, then warmed to the idea. He insisted that Thomson's theory explained the experimental data accumulated by researchers better than any other. To develop his theory, Thomson needed first of all to see which different types of knots are possible; in other words, he had to classify knots. It would then have been possible to classify atoms by associating each type of knot with a specific atom. For example, the three

Figure 1.1. Model of an atom?

ATOMS AND KNOTS

3

~~~

Figure 1.2. Three knots: the trefoil, the figure eight, and the unknot.

knots represented in Figure 1.2, the trefoil, the figure eight, and the trivial knot or unknot, could be models of carbon, oxygen, and hydrogen, respectively. So in the beginning the problem was mathematical (rather than physicochemical): the problem of classifying knots. And it was a Scottish physicist and mathematician, a friend of Thomson, Peter Guthrie Tait, who set out to solve it.

Tait, Kirkman, and the First Tables of Knots
According to Tait, a knot, being a closed curve in space, could be represented by a planar curve obtained by projecting it perpendicularly on the horizontal plane. This projection could have crossings (Figure 1.3), where the projection of one part of the curve crossed another; the planar representation shows the position in space of the two strands that cross each other by interrupting the line that represents the lower strand at the crossing. I have already used this natural way of drawing knots (Figure 1.2) and will continue to use it. Posing the question of how knots should be classified requires

4

KNOTS

Figure 1.3. Planar projection of a knot.

specifying which knots belong to the same class. It requires, in other words, precisely defining the equivalence of knots. But we will leave this definition (the ambient isotopy of knots) for later, limiting ourselves here to an intuitive description. Imagine that the curve defining the knot is a fine thread, flexible and elastic, that can be twisted and moved in a continuous way in space (cutting and gluing back is not allowed). All possible positions will thus be those of the same knot. Changing the position of the curve that defines a knot in space by moving it in a continuous way (without ever cutting or retying it) always results in the same knot by definition, but its planar representation may become unrecognizable. In particular, the number of crossings may change. Nevertheless, the natural approach to classifying

a priori there are 2 10. those in which overpasses and underpasses alternate along the curve (Figure 1. for the task to be doable within a human lifetime. the Reverend Thomas Kirkman. that is.4. But that is not so easy to do.5. exactly two .ATOMS AND KNOTS 5 Figure 1. Indeed. Tait had an initial stroke of luck: he learned that an amateur mathematician. for each crossing of a planar curve. possibilities for making a knot. or 1. For a curve with 10 crossings.2. Tait decided to list only alternating knots. knots in space consists first of making a list of all the planar curves with 1. the curves that represent the same knot in space (Figure 1. and all that remained was to eliminate the duplications systematically.4). In this way.. Two representations of the same knot.024. Of course.3. for example.5). there are two ways to decide which strand in the crossing should be uppermost. Peter Tait stopped at 10. the maximum number of crossings of the knots considered has to be limited.. crossings.4. that is. . then eliminating the duplications from the list. had already classified planar curves with minimal crossings.

" which is analogous to "prime number" in the sense that it cannot be factored. alternating knots corresponded to each planar curve.5. Unfortunately for Thomson. But before continuing this account of Kelvin's and Tait's work. at the century's close. most of the work on classifying knots (with 10 or fewer crossings) had been done. a table of (prime) knots with 7 or fewer crossings. almost no one was interested in knot tables for reasons that will be explained at the end of this chapter. Little. N. . which deals with the arithmetic of knots. The exact meaning of the expression "prime knot. Still. Kirkman. substantially facilitating Tait's task. Little managed to avoid the systematic run-through of the 210 uncrossing possibilities (for each knot) mentioned above.6 shows an example. Little. by the time Little and Tait finished their work. it is worth making a few points about classifying knots. and tables of knots appeared. after six years of work. An alternating knot (a) and a nonalternating knot (b). Figure 1. Nonalternating knots (one with 10 or fewer crossings) were classified in 1899 by C. and Tait. is explained in Chapter 4. Which is not to say that it became simple: he devoted the rest of his life to it.6 KNOTS Figure 1.

Figure 1. .6. Table of knots with seven or fewer crossings. ~.ATOMS AND KNOTS 7 %~%%~~~~ 0.

in fact. or by doing the opposite. and ambient isotopy allows us to twist and move the knot in space as we do with a real string (without tearing it). We define a knot. equivalence being the relation of ambient isotopy. A knot is just a class of equivalent representations of knots.8 KNOTS A Mathematical View of Knot Classification Let us pose the problem in precise terms. it is also a necessary condition for avoiding "local pathologies. the triangle must have no points in common with the polygonal curve other than its sides.7c). Clearly this definition corresponds to our sense of a knot as the abstraction of a string whose ends are stuck together." It is required to avoid the so-called wild knots.7a). Of course. I Representing a knot as a polygonal curve has a motive other than the ability to attach triangles to it (which presupposes that the "curve" is made of segments). as a closed polygonal curve in space (Figure 1. Wild knots are the result of a process of infinite knot- . it is perhaps not satisfying to think of strings that have angles everywhere. which are not topologically equivalent to polygonal curves (or to a smooth curve). An ambient isotopy is just a finite sequence of elementary isotopies (Figure 1. or more precisely a representation of a knot. the very concept of a knot needs to be defined. defined as follows: An elementary isotopy is achieved either by adding a triangle (as we add ABC in Figure 1. 7b) to a segment (AB) of the polygonal curve. then replacing this segment by the two other sides (AC and CB). First of all. From the aesthetic point of view. rigorous enough to satisfy a mathematician (readers disinclined to scientific rigor can skip this part after glancing at the figures). but it is the price to pay for defining a knot in a manner both elementary and rigorous.

.ATOMS AND KNOTS 9 c Figure 1. Before we continue our preliminary investigation of "tame" knots.8. ting: the twists of the curve get smaller and smaller and converge toward a limit point known as a wild point of the curve (Figure 1. A knot drawn as a closed polygonal curve in space (a) and two equivalent representations (b and c) of the same knot (isotopy). Rigorously defining a knot (as a polygonal curve or a smooth curve) makes it possible to avoid these little horrors and simplifies the theory. Figure 1. Wild knots.8).7.

and Blindness The examples of wild knots shown up to now possess a single isolated pathological point. 3 L -==9 -=========-- . But one can go further: Figure 1. then the (smaller) central subintervals 0/9. . Wild knots with several points of the same type can easily be constructed.1 3 .l.2/9) and (7/9.10 KNOTS ~~%%%~%%%%~%% here are a few remarks on their "wild" kin (with a few drawings included).2/3).9. A wild knot converging to Cantor's continuum. for those who know the expression) set of pathological points. This set of wild points is in fact the famous Cantor continuum. Spatial Intuition. 1] that remain after one successively eliminates the central subinterval (1/3.a 9 -II:===-- c::::Jt • Figure 1. o . toward which a succession of smaller and smaller knots converge.1 9 .9 shows a wild knot that has an infinite (even uncountable.l. 9 . Digression: Wild Knots. the set of points in the segment [0.8/9) of the two remaining segments.

. Zuev and represented in the same figure. No. and so on to infinity.. and (25/27. The process continues indefinitely. Begin with a solid torus. but a geometric construction devised by the French mathematician Louis Antoine. Antoine's necklace has some remarkable properties that I shall not dwell on: it will simply aid us to construct a wild knot invented by the Russian mathematician G. I do not mean a gift of the Roman general 2 to Cleopatra.. The knot in question is shown (partially) in the form of the curve that insinuates itself inside the large solid torus. take . T2. dividing each time it penetrates a torus knot to tend toward Antoine's necklace. then inside the smaller tori. Antoine's necklace.20/27). and so on. then into the tiny tori. Let us try to describe this jewel of the mathematical imagination. for example.2/27)..ATOMS AND KNOTS 11 then the four (tiny) central subintervals (1/27. construct a chain similar to the preceding one. in the shape of a doughnut and place inside it four thinner solid tori linked two by two to make a chain. One can show (but the rigorous demonstration is rather tricky) that the curve . the set formed by these four little chains (constituting the 16 tiny tori) is denoted TJ • Inside each tiny torus knot.. (7/27. 8127). A much more beguiling wild knot can be obtained by making a curve pass through a set even more complicated than Cantor's continuum. T1. and the set obtained as the infinite intersection of the sets Tj will be Antoine's necklace: A=TlnT2n .26/27) of the four remaining segments. (191 27. Inside each of these four rings of the chain T2. Va. shown in Figure 1. nT"n . of four rings..10.

So it might surprise the reader to learn that these two mathematicians are . and the set of its wild points is precisely Antoine's necklace.12 KNOTS ~. Inventing monsters such as Antoine's necklace or Zuev's wild knot takes a considerable ability to visualize three-dimensional space. ultimately obtained is indeed a simple dosed curve.10.~~~~~~~~~~~~ Figure 1. A wild knot converging to Antoine's necklace.

are acquired. are innate. little inclined to geometric considerations. Thus. it isn't all that surprising. such as topological structures. on the other hand. The spatial intuition that sighted people have is based on the image of the world that is projected onto their retinas. he immediately sees that the torus is not a sphere. But actually. He. It is also deeper-in the literal as well as the metaphorical sense. the blind person who regains his sight does not distinguish a square from a circle: he sees only their topological equivalence. at first. As for us. like Thomson and Tait. too. another researcherunknown and working in an immense. he based his tables on arithmetical relationships among the various properties of chemical elements. reflecting on the structure of matter. our tendency to consider what we see the "absolute truth" often makes us conceive the world in a very flat and superficial way. The Failure of Thomson's Theory While European physicists were debating the merits of Thomson's theory and Tait was filling in his tables of knots. . since almost all blind mathematicians are (or were) geometers. A blind person's spatial intuition. note that recent biomathematical studies (based on work with children and adults who were born blind but gained their sight afterward) have shown that the deepest mathematical structures. As we leave this digression. thus it is a two.(and not three-) dimensional image that is analyzed in the brain of a sighted person. such as linear structures. This scientist made an unexpected discovery: there are very simple. underdeveloped countrywas.ATOMS AND KNOTS 13 blind. was trying to establish tables of atoms but. is primarily the result of tactile and operational experience. In contrast. whereas finer structures.

relationships among the chemical properties. It was the mathematicians who would take up the subject once again.14 KNOTS until then unnoticed. ashamed and embarrassed. and it was quickly replaced by the arithmetical theory of Mendeleev. forgot about knots for nearly a century. Thomson's theory had not done much for chemistry. And physicists. And he published what today is called the periodic table of elements. Once that happened. It would take some time for this remarkable discovery to be recognized in western Europe. my compatriot Mendeleev buried the idea of atoms as knots. .

the same threads are also attached to nails. a classic twisted rope. but not necessarily in the same order (Figure 2.2). you can imagine a braid of n strands as n threads attached "above" (to horizontally aligned nails) and hanging "down. this means that the ob15 . As with knots.1). it is the formal abstraction of what is meant by a braid in everyday language (a braid in a young girl's hair. a plaited key chain. and of course without tearing or reattaching them) to get a braid that looks different but is equivalent to (or an isotope of) the first braid (Figure 2.' crossing each other without ever going back up. between two beautiful topological objects: braids and knots. some strings tangled in a certain way. a braided dog's leash. Mathematically speaking. The strands of a braid can be rearranged (without detaching the top and bottom. and so on)-in other words.BRAIDED KNOTS (Alexander· 1923) This chapter is devoted to a remarkable connection. what is a braid? Roughly. we do not distinguish two isotopic braids: we think of them as two representations of the same object (from the formal mathematical point of view. at the bottom. discovered by mathematicians. More precisely.

Figure 2. .2.Figure 2. Isotopes of a four-stranded braid.1. Examples of braids.

which by definition consists of only one curve). An attentive reader of the previous chapter will recognize the trefoil knot in Figure 2. which means joining the upper ends of the strands to the lower ends (see Figure 2. The basics of braid theory were developed by Emil Artin in the 1920s. and algorithmic methods. Alexander's theorem can be expressed as follows: Every knot can be represented as a closed braid. W. in contrast to a knot. But we will deal with this theory later.3athough perhaps not at once. Will a knot always be obtained in this way? According to Figure 2. (Actually. The theory is a marvelous blend of geometry. Alexander in 1923.3a). found by J. Closure of Braids A knot can be made from a braid by the operation of closure.3b. the representation of functions of several variables as the composition of functions of a lesser number of variables. since our immediate goal is to understand the connection between braids and knots. Alexander showed that this assertion . explains the importance of braids in knot theory: they all can. and combinatorics. It has a bunch of applications ranging from the textile industry to quantum mechanics by way of topics as varied as complex analysis. H. several curves. not always: the closure of a braid may very well result in a link of several components (that is.BRAIDED KNOTS 17 ject in question is not a concrete braid but an equivalence class of braids). algebra. The following question arises immediately: Which knots can be obtained in this way? The answer.

then. that is. the set of braids possesses a very clear algebraic structure that enables one to classify them. Indeed. Closure of two braids. Is it reasonable. as we will see later. For now. when it always turns in the same direction around a certain point (as the knot in Fig- '--" (a) '--" (b) Figure 2. braids are much simpler objects than knots.Sth%~%% is true for the more general case of links.18 KNOTS ~~~Sth~%%~"~~~\{.3. note that the desired braid can easily be seen when the braid is rolled up in a coil. let us return to Alexander's theorem: How can it be proved? Given a knot.) Alexander probably hoped that his theorem would be a decisive step forward in classifying knots. of which knots are just an example. how does one find the braid whose closure would be that knot? First of all.:~\'%S\'h~%%. to try to use braids to classify knots? How did this idea develop? We will see at the end of the chapter. .

this knot is called a figure eight knot. Another method of braiding knots.4b). In this situation. just move the "fat" part (the thicker line) of the knot (the one "going the wrong way") over the point C on the other side of the curve.5c). this elegant method (transforming any knot into a coiled knot) is universal.BRAIDED KNOTS 19 Figure 2. it is difficult to teach to a computer. But what if the knot is not coiled. Actually. .Sa? (As readers of the first chapter will know. Its weakness-and there is one-is its ineffectiveness from a practical point of view. specifically.Sb) can then be unrolled into a braid as in the preceding example (Figure 2. more doable and easier to program. Unrolling a coiled knot into a braid.4 turns around the center C). all that is required to find the braid is to cut the knot along a line extending outward from the center and then to unroll it (Figure 2.4. ure 2.) In that case. for example. The resulting coiled knot (Figure 2. and it allowed Alexander to prove his theorem. the knot shown in Figure 2.

20 KNOTS (a) (b) Figure 2. was invented by the French mathematician Pierre Vogel.S. Assume that a given knot is oriented. that is to say. To describe it I must introduce some definitions related to the planar representations of knots. The reader not inclined to algorithmic reasoning can blithely skip this description and go on to the study (much simpler and more important) of the group of braids. Vogel's Braiding Algorithm The braiding algorithm transforms any knot into a coiled knot. The planar representation of the knot defines a kind of geographic map in . Coiling a figure eight knot and unrolling it into a braid. the direction of the curve (indicated by arrows) has been selected.

Desingularization transforms each knot into one or more oriented. Because the curve of the knot is oriented. the crossings are marked with arrows.6a. closed curves (without crossings). Desingularized knots and Seifert circles. in other words. the regions or countries being the areas bounded by parts of the knot's curve. the plane. that represent the knot (Figures . Included among the countries is the infinite region-the one located outside the curve. called Seifert circles. these make it possible to "desingularize" the knot N unambiguously.BRAIDED KNOTS 21 Figure 2.6. the border of each country consists of several edges (oriented in accordance to the arrows) that join one crossing of the knot to a neighboring crossing. to replace all the crossings of N by their smoothings as shown in Figure 2. On this map.

the change-of-infinityl operation nests the two circles (Figure 2. In fact. the change-of-infinity move resembles the operation carried out in Figure 2. Note that desingularizing a coiled knot always yields a nested system of Seifert circles. some of which (in this case. but inverting this last circle results in the circle 3'.7b). The result is to create a central country (not in turmoil) and several new countries.6e). PI is not.7c). H is in turmoil. and vice versa (Figure 2. (In this case.8. A country in this map is said to be in turmoil if it has two edges that belong to two different Seifert circles. when Seifert circles are not nested (as in Figure 2. in the smoothing of knot N in Figure 2. two) may be swallowed up by bordering countries. I am sure that now the reader understands the choice of the geopolitical term perestroika. For example. circle 3 does not encompass them. this figure shows that while circles 1 and 2 are nested. which neatly encompasses circles 1 and 2. P2 is not in turmoil either.6d.6b). 2. because its edges go around P2 in opposite directions.5.) Let us now consider the planar map determined by the knot N. whereas regions PI and P2 are not: since the thick edges head in the same direction around H and belong to two distinct Seifert circles. . labeled with arrows going in the same direction around the region. Two Seifert circles are nested if one of them is inside the other and if the orientations of the two circles coincide. finally. because its edges belong to a single Seifert circle. Perestroika consists in replacing the two faulty edges by two "tongues:' one of which passes over the other. On the other hand. forming two new crossings. An operation called perestroika can be applied to any country in turmoil (Figure 2.2.9). the country H is in turmoil.

Change of infinity.BRAIDED KNOTS 23 %%%%~~~~~~%~~%%~%~~%~% (a) (b) Figure 2. Countries in and out of turmoil.8. .7. (c) N Figure 2.

9.24 KNOTS Figure 2. Perestroika of a country in turmoil. but the command Do change i nfi ni ty requires clarification: it means taking one of the small- . Vogel's algorithm can now be presented in the form of a "program" written in a sort of "pseudO-Pascal": Do smoothing While: There is a disjoint region Do perestrOika Do smoothing End while While: The Seifert circles are not nested Do change infinity End while Stop Most of this language was explained above.

Following the first smoothing.4a (and so can admire the desired braid. there are no countries in turmoil. o . I have redrawn it twice (Figures 2.11c). which must be carried out twice-(b) becomes (c) and (c) becomes (d)-to obtain a coiled knot (d).l2b. So we proceed to the command Do change infinity. it is followed by a change-of-infinity move. 2. which appeared earlier as Figure 2.11 shows how Vogel's algorithm coils a knot with five crossings. in fact) to see how changes of infinity occur (Figure 2. Figure 2. Vogel algorithm applied to the unknot. To make sure.10).BRAIDED KNOTS 25 est Seifert circles not nested with the others and sending a point inside this circle to infinity. even if it does not look like one. the loop (in the computational sense of the word) contains two perestroikas (Figures 2.l2c). 2 Following the initial smoothing. 00 C!5 (a) o (b) Figure 2. First let us apply the Vogel algorithm to a very simple knot (the unknot. The reader will have no difficulty recognizing the rolled-up knot from Figure 2.11 band 2. which can then be unrolled into a braid as before (e).10.11d) is indeed a coiled knot. The result (Figure 2. and no nested Seifert circles.4b).

U KNOn ~~~~%~~~~%%%%%%~~%~~ (a) (b) (c) (d) Figure 2. .11. Vogel algorithm applied to knot 52.

to prove that the same is true for the first loop. Of course.12. Proving that the second loop always termi~ nates is elementary. on the set of braids with n . Using the Vogel algorithm to unroll a knot. we are going to define an operation. the composition or product. The Braid Group Let us return to the study of braids. We will come back to the coding of knots in Chapter 4. Yet it does. to transform our "program" into software for a real com~ puter. and very rapidly. Note that it is not at all obvious that the algorithm. we have to know how to code knot representations in such a way that the machine will be able to work with them. will always terminate. On the other hand. First of all. which includes two While loops that are a priori dangerous.BRAIDED KNOTS 27 (a) (b) (c) Figure 2. Vogel had to use fairly sophisticated algebraic topol~ ogy methods.

like the number I.13.I = lin is equal to one. which does not change the class of the braid in any way. whose product with b gives the trivial braid: b .' = e (just as for each number n.28 KNOTS Figure 2. as in Figure 2. strands. b. a braid that. there is a braid known as the unit braid (e). for each braid b there exists an inverse braid. This is the trivial braid. Sure enough. its product with the inverse number n. the inverse braid is the . First.14. The product of two braids. As can be seen in Figure 2.13. b. Second. This operation consists simply of placing the braids end to end (by joining the upper part of the second braid to the lower part of the first). n.I = 1). whose strands hang vertically without crossing. does not change what it multiplies. n .' . It turns out that the product of braids possesses several properties that resemble the ordinary product of numbers. appending a trivial braid to a given braid amounts to extending its strands.

14. The product of braids makes it possible to replace the picture repre- Figure 2. This group will be denoted by Rn. the braid group Rn (for n > 2) is not commutative: the product of two braids generally depends on the order of the factors. indeed. however. beginning at the middle of the product braid. Thus. In contrast to numbers.BRAIDED KNOTS 29 braid obtained by taking the horizontal mirror image of the given braid. each crossing cancels with its mirror image. (b • c) is always true. such that all the crossings gradually dissolve. The third property common to braids and to numbers is the associativity of the product operation: (a . The product of a braid and its inverse. When a set is endowed with an operation that enjoys all three of the properties just described. c = a . b) . two by two. mathematicians call this set a group. I have just shown that braids with n strands form a group. .

. senting a braid by a word-the algebraic encoding of that braid. He found a series of algebraic relations between braid words that gave an adequate algebraic description of their isotopy. moving along a braid from top to bottom.2. But recall that the geometric braids possess an equivalence relation.1 S. n - 1 . So we have replaced braids-geometric objects-by words: their algebraic codes. Algebraic representation of a braid._ Figure 2.30 KNOTS ~~~~~~~~%%~%~ 2 ••• ••• • 2 3 • n-1 n 11I··· X- 1 b. isotopy. • •• .15). What does that mean algebraically? Artin had an answer to this question. . These relations are commutativity for distant braids i. we see that it is the successive product of braids each with a single crossing (Figure 2. namely. . bI . we call these elementary braids and denote them by bl. Indeed.j = 1..I (for braids with n strands). bn.

3 bjbj. B4: .. . . and Artin's relation (or the braid relation) i = 1.16) suffice to replace the geometric manipulations related to isotopy by admissible algebraic calculations on the braid words. also shown in Figure 2.16. Here is an example of an admissible calculation in the group of braids with four strands. that is. . .Q. n .16. Relations among the group of braids.. they do indeed correspond to isotopies.2 Their geometric interpretation is shown in Figure 2.2. ..' = e = bj-'bj. it is immediately obvious that these relations are valid for braids. Figure 2.16 by the other member of that relation.:~> ·IT·· -. To an observer with a little spatial imagination. What is less obvious-and is one of Artin's key findings-is that these two relations (if one adds the trivial relations. each of the admissible calculations consists in replacing a part of a word identical to one of the members of the relations that appear in Figure 2.BRAIDED KNOTS 31 i i+l i+2 i+l i+2 "fl-~' .

nor the one recently discovered by the French mathematician Patrick Dehornoy. it allowed him to find a comparison algorithm for them. But to convince you of the efficiency of algebraic-algorithmic methods in geometry. I chose (more or less at random) an example of a calculation carried out by my little computer (which has in the recesses of its electronic memory some software that drives Dehornoy's algorithm). for each pair of braids. the algorithm says "no" if they are not isotopic and "yes" if they are (as well as providing a set of admissible calculations that lead from one to the other in the latter case). I have enclosed within parentheses the parts of the word that are replaced successively during the calculation. neither Artin's (which he called by the lovely word combing).) More precisely. Artin's theorem is important because it reduces the geometric study of braids to their algebraic study.32 KNOTS (To make this formula easier to read. This algebraic approach allowed Artin to classify braids. This calcu- . which is generally more efficient. which is much simpler and more efficient. Classifying Braids I will not be describing any of the braid-comparison algorithms here. Put in other words. Artin's theorem affirms that Two braids are isotopic if and only if the word representing one of them can be transformed into the word representing the other by a sequence of admissible calculations.

B. shows that a braid with four strands that looks rather complicated is in fact the trivial braid. which takes place in the group B4 and uses the (more readable) notation a.. .BRAIDED KNOTS 33 lation. ABBAAAAA[Abbbbbbbbcba }AccBCaBBBBBBaaaaaaBB = ABBAAAAA[Aba }aaaaaaaBcbaBAccBCaBBBBBBaaaaaaBB = ABBAAAA[Aba}BaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB = ABBAAA[Aba}BBBaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB = ABBAA[Aba}BBBaaaaaaaBcbaBAcBCaBBBBBBaaaaaaBB =ABBA[Aba}BBBBaaaaaaaBcbaBAcBCaBBBBBBaaaaaaBB = ABB[Aba}BBBBBaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB = [ABa} BBBBBBaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB = [ABa} BBBBBBaaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB = b[ABBBBBBBBa}aaaaaaaBcbaBAceBCaBBBBBBaaaaaaBB = bbAAAAAA[Aba}aaaaaBcbaBAceBCaBBBBBBaaaaaaBB = bbAAAAAAb [Aba }aaaaBcbaBAceBCaBBBBBBaaaaaaBB = bbAAAAAAbb[Aba}aaaBcbaBAceBCaBBBBBBaaaaaaBB = bbAAAAAAbbb[Aba}aaBcbaBAceBCaBBBBBBaaaaaaBB = bbAAAAAAbbbb[Aba }aBcbaBAceBCaBBBBBBaaaaaaBB = bbAAAAAAbbbbb [Aba} BcbaBAceBCaBBBBBBaaaaaaBB = bbAAAAAAbbbbbbAB[Bcb }aBcbaBAccBCaBBBBBBaaaaaaBB = bbAAAAAAbbbbbbA[Bcb }CaBcbaBAceBCaBBBBBBaaaaaaBB = bbAAAAAAbbbbbbAcbCC[aA}ceBcbaBABBBBBBaaaaaaBB = bbAAAAAAbbbbbbAcb [CCce} BCaBBBBBBaaaaaaBB = bbAAAAAAbbbbbbA[cbBC}aBBBBBBaaaaaaBB =bbAAAAAAbbbbbb[Aa}BBBBBBaaaaaaBB . b. respectively. . c.. A. b3-I. C for the basic braids bl> bl-I.

the reader can draw the given braid and try to unravel it geometrically-although doing so may instill an inferiority complex with regard to my notebook computer.34 KNOTS = bbAAAAAA[bbbbbbBBBBBBJaaaaaaBB = bb[AAAAAAaaaaaaJBB = [bbBBJ = e For comparison. and we have just seen that braids can be classified. which did the job in less than a tenth of a second. Can Braids Be Used to Classify Knots? Alexander's theorem affirms that all knots are closed braids. Can the classification of knots be deduced from these two facts? Several mathematicians. brimming with new developments. The attempt made for a rich history. But this chapter has gone on too long. that began during the 1930s and perhaps has not yet ended. and I will end it by referring the amateur math lover of nice stories to an article by Dehornoy (1997). .4 have probably nourished this hope (I know some who are still hoping). and not the least of them.

future author of the first book about the mathematics of knots. nor did the combinatorial approach (defining a knot as a closed polygonal curve based on the coordinates of its successive vertices). which enables one to follow the evolution of its positions in space. began to study knots in depth. In practice. the German mathematician Kurt Reidemeister. How could they be classified? The problem of systematizing the possible positions of a curve in space was to prove devilishly difficult. But can the process be inverted? Can the projection be continuously modified in such a way as to obtain all the possible positions of the string in space? That is the question Reidemeister asked himself. Manipulating the string that determines the position of the knot causes its diagram to undergo continuous modifications. seeing a knot means drawing it.3 PLANAR DIAGRAMS OF KNOTS (Reidemeister· 1928) During the 1920s. that is. the famous Knottentheorie. the information did not enable one to see the knot or to manipulate it. 3S . In both cases. The analytic approach (defining a knot using equations) did not help. projecting it onto a suitably chosen plane to obtain what is called a knot diagram.

..... • Q 3: passing a third strand over a crossing.... 1' ... ....2a). Reidemeister moves.} ••••. :'~\ .. they are denoted by the symbols QI' Q2.1. shows how Reidemeister moves are involved in representing the manipulation of a knot.•..~.)(. t•• ". f . ... The process begins with the uncrossing (disappearance) of twin crossings at the top (shaded area of Figure 3.. " :/"R..... ....•• \~ •• ) ••. which represents an unknotting process.. similar to those shown in Figure 3..... Q. 3 •••••• Figure 3.. . while continuously changing the diagram of the knot in the plane without altering the number and relative disposition of the crossing points)./\..36 KNOTS Here is his answer: Just perform a finite number of operations on the diagram. ~..... "":.y~""" . ~\ .. • Q2: appearance (disappearance) of twin crossings..'\ I ' ..... . while doing trivial planar manipulations (that is. . the disappearance of a pair ~""""' ~ . The following figure. Q ...The moves that they allow are the following: • QI: appearance (disappearance) of a little loop.... •.... Q 3...1.~I .. \/ :~ ... followed by the passage over a crossing (b).... The three operations shown in the figure are now called Reidemeister moves...

of twin crossings (c). the diagram of the knot undergoes trivial planar manipulations in preparation for these movements (these trivial manipulations change neither the number nor the distribution of the crossings). the disappearance of a pair of twin crossings (e).2. To understand where Reidemeister moves come from. finally. and why they .PLANAR DIAGRAMS OF KNOTS 37 (e) (d) Figure 3. the disappearance of a small loop (d). Unknotting using Reidemeister moves. and. The reader will have noticed that between the actual Reidemeister moves.

What does it mean to say that? A mathematician would answer that the plane must be selected so that the projection is generic. (2). (3) must be removed by slightly displacing one of the knot's vertices. (2) a vertex (or several vertices) cannot fallon the projection of an edge or line segment (to which they do not belong). I mentioned above that they are projections in a "suitably chosen" plane. singularity. Let us be specific about what all these synonyms mean (catastrophe. but this specification is utterly useless if you do not know the term. A generic projection is a projection without catastrophes. Note that these three catastrophes differ from the crossing catastrophe. which is in fact a basic mathematical conceptI-intuitively clear but difficult to formulate in the general case. we assume that (1) two (or more) vertices cannot be projected to the same point. (3) three (or more) points cannot be projected to the same point. By definition (of a generic projection of a knot). The existence of a generic projection for any knot is obvious and can be demonstrated easily:2 the "forbidden catastrophes" (1).38 KNOTS suffice. or gotten rid of by making tiny changes to the projected object. degeneration) in the case of a knot represented by a closed polygonal curve. singularities. Generic Projection and Catastrophic Projection In introducing the diagrams of knots. which occurs when two points inside two distinct edges project onto a . we will have to spend a little time talking about knot projections. or degenerations that can be avoided.

the catastrophes that occur most often: among all the projections. The shrewd reader. For example. including 5 vertices. There are.3a. rarer catastrophes-also forbidden because they are special cases of catastrophes (1). and (3). of course. Thus.PLANAR DIAGRAMS OF KNOTS 39 single point: this is inevitable. in the upper part of Figure 3. as shown in Figure 3. Reidemeister moves correspond precisely to the most common forbidden catastrophes. the most common among exceptional events.3a. To the right of Figure 3. examining the next panels (b and c) of Figure 3. 17 points. but they are.3. in some sense. will see how a type (2) catastrophe can give rise to move Q2. this move is symbolically represented in keeping with the style of Figure 3. we see a type (2) catastrophe in which the vertex A of edge AB (which moves) projects momentarily (the middle frame) to a point inside the projection of edge Be. This position corresponds to the disappearance of a little loop on the projection (the knot diagram).1. and so on. can project to a single vertex. so to speak. we are in a position to discuss their principal application (often called Reidemeister's theorem or lemma): .3. The Sufficiency of Reidemeister Moves Now that we know the origin of Reidemeister moves. they do represent exceptional events. inasmuch as any little change made to the knot moves the position of the crossing on the diagram somewhat. 7 edges (perpendicular to the plane of projection) can degenerate to a single point. Catastrophes (2) and (3) are. it is the move QI. but cannot totally eliminate it. (2). and a type (3) catastrophe to move Q3. that is.

' ::::.. ...hl :/.... 1 1 ~~ / ' A ' .. i 0 '..... Catastrophes and Reidemeister moves.~..' . . :! i : ' . ...3. . '-~ " . .:: I :. ..' t 1'jk-~~N~Q it~"K ! tf$~ !~~ '... : : : :: :!: : : : r: ::: : : : :: !:: : it:: : : : : :.: 11 I. :: : :::!:! : l:"::: ! ':: ~ i ! ~ .. . '. .: I I C /. : .. :: l-t .. :. ..~ :to "'.. <.. . .J.. ... ' 1 1 .' .....l~(: :::::: : ::: ....' I )'t"': )t ::::..(aj C I ' ! : i ~Ai. il' ... # Figure 3. . .: .. ~\W<1\tl?1 @ ( b) :: r: : T r: :: ! 1 :.!../...... i : ~ l::: :! :: :.........i/ .. '<'.. .. '. ••••• .. !~.. i . . : ': .: i i o.... 1 tY4r-~mS 1'~.. :::: : ::~ •••••• • .......... ' ' • I .: • i • i : ~/. t • • I I I .~ ..... ... I I I i:! 1 I i /:~~~.( :1'" M~: If :.... I.# • I :: ! : T: : : :.1 : . ~~ ~ (~ 1: I I c: I . 0 ! 0 I !: o .:! i:"::: ::::!::::: !: Y 0 I I .. · . /# ~..(.' :: i : :.. . (c) r: :! .. :: t i ! i i": ::1 i ... ' I :::: : y T + :: r: :.. O~! : .# :\. . :! :t t : :: :: ! :: + \ .: :-i-~j ~H H-i-l~1'i7i\! ~-j\i-~f-l i :-i.. .. . i i o 0 " • I I T:: 0 !::.

PLANAR DIAGRAMS OF KNOTS 41 If one knot can be transformed into another knot by continuous manipulation in space. the same result can be obtained by a manipulation whose projection consists uniquely of Reidemeister moves and trivial manipulations of the diagram in the plane. This means that it is possible to examine all the spatial manipulations of knots by trivially manipulating their diagrams in the plane and applying Reidemeister moves from time to time. to devise an algorithm that will determine whether two knots (as sketched) are equivalent. we set out to classify knots-in other words. Reidemeister reduced the three-dimensional and rather abstract problem of knot equivalence to a two-dimensional problem. depending on your point of view). In this way. Caught up in the excitement. let me say a few words about its proof. For readers more at home with mathematics. Does the Reidemeister Theorem Classify Knots? Let us put ourselves in Reidemester's shoes and imagine that we are elated at having demonstrated his theorem. and a more concrete one. the proofs I know are not simple enough to be included in this book. . as we have seen. only catastrophes (2) and (3) can occur and. Unfortunately (or fortunately. in that case. these correspond exactly to Reidemeister moves. I will simply say that all it takes is to analyze in detail a single basic triangular movement (see the beginning of Chapter 1) and to perturb it so that it is in a "general position". to boot. to their classification). Before discussing what the Reidemeister theorem contributes to the study of knots (in particular.

the knots are equivalent.) . compare. we will have to remember the changes to the first knot and apply to it another Reidemester move. among other things. I am sure Reidemeister must have had this thought at one time or another. then compare again. a full day's work. of understanding. the humble "notebook" on which I am writing this text has. software that can untie knots (by comparing them with the unknot. and we have succeeded. say. is it because the knot is nontrivial or because the computer needs more time to find the sequence of operations that will lead to an unraveling? Must we kiss Reidemeister's idea good-bye? Not yet. we go back to the modified knot.3 Does that solve the problem of classifying knots? Of course not. Thus. If not. as in the preceding paragraph). If they are identical. If several moves applied to the first knot don't work. and perhaps it evoked in him that extraordinary feeling that researchers sometimes feel-the feeling of being on the verge of finding out. because there is another possibility that could salvage it. this feeling is often followed by despair when the idea turns out to be inadequate or illusory. and for a good reason: applied to two nonequivalent knots. If not. the algorithm described above never terminates. (I note parenthetically that even for the best of us. If the number of crossings and their relative positions are the same. and so on. The algorithm described above is easy to implement on a computer. and so on. (randomly) apply a Reidemeister move to the first knot and again compare the result with the second. If the two knots are equivalent. sooner or later a set of Reidemeister moves leading from the first knot to the second will be specified. apply yet another move to it. And the user faces a dilemma: if the software does not give an answer after. even a very little one. we have won again.42 KNOTS Take the first knot and compare it with the second. it continues on and on without ever stopping.

The hope of obtaining a simple and effective method of classifying knots using Reidemeister's theorem was too optimistic. To create an algorithm for unraveling (that is.4. . we just need the move Q 3 and allow the moves Q. Unfortunately. nonsimplifiable knot. it is necessary at first to complicate things to make them simpler. An example of a trivial. and Q 2 only when they decrease the number of crossings. A trivial. or none of the permissible moves is applicable4 (and the knot is nontrivial).PLANAR DIAGRAMS OF KNOTS 43 This other possibility is very simple: moves Q\) Q2. increase the number (appearance of a loop or twin crossings). If the process is restricted this way. In reality one cannot always unravel a knot by simplifying it (by diminishing the number of crossings) at each step of the unraveling: sometimes. this argument (however convincing) is erroneous. the number of crossings diminishes. not change the number (Q3 simply passes a strand over a crossing). finally. simplifying the knot). and the algorithm (thus perfected) will always terminate: either there are no more crossings (and the knot was trivial). or. The world is Figure 3. nonsimplifiable knot (that can be unraveled only by first increasing the number of crossings) appears in Figure 3.4. Q 3 may diminish the number of crossings (disappearance of a small loop or of twin crossings).

44 KNOTS Figure 3." made that way: to unravel a knotted situation. Since Reidemeister's theorem was discarded. Moreover.5. often one should begin by tangling it even further.5. for which we can thank Wolfgang Haken. Wolfgang Haken's "Gordian knot. devising trivial knots that are hard to unravel has admittedly become an important exercise in research on unknotting algorithms. but his algo- . only to unravel it better. A particularly barbaric example of such a knot (very hard to unravel-try it!) is shown in Figure 3. 1961). it was Haken who finally solved the problem of un knotting (Haken.

so verification is in general very easy. these particular moves are very simple. the function must never change during the process of manipulating knots. from a theoretical point of view. unless we use the algorithm that Alexander of Macedonia applied with so much success to the Gordian knot: cut it! . Louis Kauffman. or that this theorem is simply yet another example of a spectacular failure and dashed hopes. The theorem occupied a central place in subsequent developments. and their followers (Chapter 6). it is not a real unknotting algorithm (since it can "continue indefinitely" without giving an answer). however. especially in the study of knot invariants by Vaughan Jones. In order that a knot diagram's function proposed in the guise of a new invariant be indeed invariant. or Q 3 are performed. Q2.. But that is not all. proving this requires only verification that the function does not change when the moves QI. What Remains of Reidemeister's Theorem? We should not take the collapse of naIve expectations to mean that the application of Reidemeister's theorem is limited to an algorithm that does not work. From a practical viewpoint. this algorithm and its recent modifications may be used as a relatively efficient tool that often enables a (strong enough) computer to unravel knots that cannot be undone "by hand"5 .PLANAR DIAGRAMS OF KNOTS 45 rithm (too complicated to be put on a computer) is based on a very different class of ideas.. well. According to Reidemeister's theorem. The failure of the unknotting algorithm described above is relative. Of course.

knots in particular. anyone who has read Chapter 2 will say). Thus.1) is the (unique) composition of prime knots. foreshadowed by many researchers. Why should the natural numbers1.. the one drawn at the left in Figure 4. Moreover. . as can be seen at the right in the same figure (it is the composition of two trefoil knots and one Turk's head knot). 4. 2. 5. proved difficult (as in fact did the corresponding principle for numbers) and was not achieved until 1949 (by the German mathematician Horst Schubert). The reader will have grasped that the "composition" of knots consists more or less in setting them end to end (as is done for braids. 3. knots possess an arithmetic very similar to that of the natural numbers. we are going to put knots· in boxes: in 46 .. To explain this operation. every knot (for example.-be the only objects that can be multiplied and decomposed into prime factors? The same features describe other mathematical entities.THE ARITHMETIC OF KNOTS (Schubert· 1949) Arithmetic-of knots? Absolutely. with commutative multiplication (called composition) and a theorem asserting the uniqueness of decomposition into "prime knots:' The demonstration of this fundamental principle. just as every whole number (say 84) factors out in a unique way (84 = 2 X 2 X 3 X 7).

this way. Composing boxed knots.2.THE ARITHMETIC OF KNOTS 47 ~c - Y (b) Figure 4. each knot appears as a knotted string inside a cube. (I will leave to the reader already corrupted by the study of mathematics the task of transforming this intuitive description into a rigorous mathematical definition.2a). .) Making a closed-curve knot from a boxed knot is easy (just join the two ends with a string outside the box). Once all the knots have been boxed. with the two ends stuck to two opposite faces of the cube (Figure 4. Decomposing a knot into prime factors. and vice versa.1. defining their composition is even easier: simply juxtapose the boxes and dissolve the double wall that separates them2 (Figure 4. (a) Figure 4.2b).

This assertion is obvious. being trickier. Juxtaposing such a box with that of any other knot obviously does not change the type of the knot. which (like the number 1) does not change the knot with which one composes it (just as the number 1 does not change the number it multiplies): a#l=a=l#a In its "boxed version. The following property is the existence of the trivial knot or unknot. juxtaposing the three knots (in both cases) and then eliminating the two walls (in a different order." the unknot can be represented as a horizontal rectilinear thread in its cube. of course. for it means. The first is associativity.48 KNOTS Our immediate goal is to study the main properties of the composition of knots. Commutativity of the Composition of Knots Like the multiplication of numbers. which tells us that: (a#b)#c= a#(b#c) where the symbol # denotes the composition of the knots. then composing the knot obtained with a. merits a subheading of its own. indicated by the number 1. roughly speaking. The following property. the composition of two knots is commutative (the result does not depend on the order of the factors): . but the result is the same). This equation means that composing first the two knots a and b and then composing the knot obtained with the third knot c gives the same result as composing first the two knots band c.

'A~ : II d) ®~ -@. but I am sure that its demonstration. the second knot is sent into the first box. which gives us the opportunity for another biological digression.3b). Ila) ~~.-®~Ie) If) Figure 4. tight knot (Figure 4. shown schematically in Figure 4. pulling on the ends of the string that forms the first knot results in a small. will please the reader. and the little knot is blown up to original size. Finally. .3. ~-®. : b) (c) ~. the little knot is slid along its own string.3. then along the string that forms the second knot (c). Voila-there you have it (f)! It may be hard for the reader to understand how a knot can "slide along a string.THE ARITHMETIC OF KNOTS 49 %%%%~%%%%%%%%~%~%~%%~~% a#b= b#a This relationship is not at all obvious. Composition doesn't depend on the order of the knots. Still sliding along the string. What is going on in this figure? First. the little knot traverses the big one and ends up to its right (d)." The simplest way is to take a good length of string (a shoelace will do) and execute the maneuver. Certain organisms are able to perform the process on themselves. Next.

thus spreading the saliva (which it secretes simultaneously) along its length (Figure 4. . The myxine's knot. slime eel).4. It quickly forms a knot with its tail and slides the knot along its body. it will slip through your fingers-not only because it is coated with oily saliva.4a). but also be- (a) Figure 4. more commonly. The myxine inhabits the ocean bottom in temperate latitudes. If you grab a myxine with your hand.50 KNOTS Digression: The Sliding Knot Fish The strange fish in question is called a myxine (or. It has a supple backbone and secretes a very thick acidic saliva that it uses to coat its body when a predator tries to grab it.

5\l. composed with the first. Finally. especially feeding. for a given knot. in general. can a knot be made at the other end in such a way that the two knots cancel each other out when one pulls on the two ends of the string? The analogous response to the question for natural numbers is no: for every natural number n > 1. That is. that would be capable of executing the same maneuver with more complicated knots. The traction it gets from moving its knotted body enables the myxine to carry out other vital tasks. 1966. To my knowledge. Can One Knot Cancel Out Another? Having defined the composition of knots. the myxine does not know how to make any other knots.%%%~%%~~~%% cause of its knot. can one find another knot that. which it achieves by suction (the myxine is a necrophage that leaves behind only the skin and bones of its prey).) Note that the slime eel's knot is a trefoil (that is. gives the unknot? In more geometrical language: If there is a knot at the end of a string. which it advances by pushing forcefully against your fist while its head moves toward the rear and escapes your grasp (Figure 4. with an even more flexible backbone. see Jensen. But let us leave aside these biological knots and come back to their (admittedly more appetizing) mathematical models.THE ARITHMETIC OF KNOTS 51 ~. the simplest nontrivial knot) and. the left trefoil. the myxine gets rid of its layer of saliva (otherwise it would smother in its slimy cocoon) by the same sliding motion of the knot from the tail to the head. there is no natural number m such . but one could easily imagine a longer species of eel. one might ask whether inverse knots exist. once the danger has passed.4b). (For more details about this unusual animal.

. try some experiments with a string. because we can write: C = (a # b) # (a # b) # (a # b) # . ) = a # 1 = a Thus we deduce that a = 1..... In fact. this composition is equal to the unknot. m = l. not a natural number. (Of course.a!: 1.52 KNOTS that n .) We will see that it is exactly the same with knots: no trivial knot possesses an inverse knot. = 1 # 1 # 1 # .. The quotation marks in the preceding sentence indicate that in fact the "proof" is doubtful (that is the least one can say). but then m would be a fraction. On the one hand. we obtain: C = a # (b # a) # (b # a) # . coming back to whole numbers. This assertion is far from obvious.. Indeed. one can take m = 1/ n. one can "prove" in the same way that 1 = 0: ... a :.. Consider the infinite composition: c= a#b#a#b#a#b#a#b#a#b#a#b# . b :.. Perhaps the ensuing failures will provide some clues for reflection.a!: 1 and "shows" that there are no inverse knots.a!: 1) such that a # b = 1. it appears false from the start: Why can't one make a "symmetrical" knot at the other end of the string that would cancel out the first one?3 Why? Before reading the explanation below.. Let us reason by contradiction: Let a and b be knots (nontrivial. = a # (a # b) # (a # b) # . = a # (1 # 1 # . which contradicts the assumption a :. beginning with the trefoil knot. = 1 But by arranging the parentheses differently. that is.

1 + 1 . Prime Knots We have just seen that there are no inverse knots.THE ARITHMETIC OF KNOTS S3 ~~%%%%%~~%~%%% just consider the infinite sum 1 . The error in reasoning is the same in both cases: one cannot manipulate infinite sums or compositions (they must be defined beforehand) as one manipulates finite sums or compositions. of the differential and integral calculus). by the way. and place parentheses in two ways. independently of Newton. a slight modification of the argument makes it totally rigorous. since knot theory did not exist at the time. indeed. that the num- . just as there are no inverse natural numbers (which means. In the case of knots..S)-and replace the questionable arithmetic manipulations by correctly defined topological manipulations. however.1 + 1 .1 + .. in other words. I will skip the technical details. rigorously defines the infinite composition C (Figure 4. Just place copies of the knots a and b one by one in an infinite series of boxes that become smaller and smaller and converge to a point-which.1 + 1 . it is the basis of a rigorous mathematical argument. in order to prove (correctly) a theorem concerning a classical object in the differential calculus-conditionally convergent series. Finally. to end this section. 4 The reader will have to take my word that this clever proof using infinite compositions is more than a brilliant sophism (like that of Achilles and the tortoise). let me add that the author of this clever argument is not a specialist in knot theory but the German philosopher and politician Wilhelm Leibniz. precisely as above. Leibniz discovered this argument in an entirely different context. a great mathematician when he put his mind to it (inventor.

M KNOTI ~~~~%%~~~~S\:t..5. . One knot cannot cancel out another. ~) 1=0 ~(3) Figure 4.

41.5.17. This property (of having no divisors other than itself and 1) is in fact the definition of prime numbers. How can we establish that prime knots exist? How does one prove. But what is the situation for knots? Are there prime knots.. for example.11.37.13. .31. Prime knots and composite knots. and the two alternating knots with five crossings (Figure 4. .53.3. that the trefoil is indeed a prime knot? The notion that comes immediately to mind is to use the minimal number of knot crossings: if the trefoil (which has three crossings) is the composition Figure 4.43..23. the figure eight knot. whereas the square knot (called a double knot by boy scouts) and the granny knot are composite knots.THE ARITHMETIC OF KNOTS 55 ber 1 has no divisor other than itself). in principle taught in the earliest grades at school.6. The set of prime numbers has continued to mystify mathematicians since the beginning of their profession.19. This simple definition gives the following mysterious set: 2.29. knots that cannot be represented as the composition of two other nontrivial knots? The answer is yes: the trefoil.47.6) are prime.7.

in contrast to numbers. the parallelism with natural numbers is complete: every knot decomposes uniquely into prime knots. but its proof is too difficult to include in this book. such that the prime factors of the number are the prime factors of the knot. but there is no natural algorithm that can produce it. 6 is greater than 3. Such a numbering would associate a prime number to each prime knot. Two times 3 makes 6. Unfortunately. How does that work for knots? Unique Factorization into Prime Knots Here. though his proof. The reason is that. too. every knot decomposes into prime knots. It was Horst Schubert who did it at the end of the 1940s. a composite number to each composite knot.56 KNOTS ~%%~%~~~%%%%~%%%%~%% of two other nontrivial knots. This assertion is actually true. at once profound and very technical. which is associated with other properties common to knots and to natural numbers. Obtaining the proof of this marvelous theorem was the ambition of many researchers. Alas! such numbering may indeed exist in principle. is beyond the scope of this book. the decomposition of each natural number into prime factors is unique. this argument is insufficient because we do not know whether the minimal number of crossings of a composite knot is equal to the sum of the minimal number of crossings of the two factors. those would have at least three crossings each (because knots with two crossings or fewer are trivial). Thus. brings us to the obvious idea of numbering knots in such a way as to conserve the decomposition into prime factors. two knots cannot be . Now. and so there is a contradiction: QED. Schubert's theorem.

5 Another reason there is no natural numbering of knots is the lack of order in the set of knots.' since this operation of addition does not exist. this order does not exist (or has not been discovered!) for knots. and of the most exalted kind. it is mathematical art for art's sake.4.5). it is usual to order knots by the number of minimal crossings of their diagrams.2. thus 5 = 1 + 1 + 1 + 1 + l.THE ARITHMETIC OF KNOTS 57 added. but this order is not linear: which of the two knots with five crossings (see the table of knots in Figure 1. it is not possible to obtain all knots by "adding copies of the trivial knot. Natural numbers possess a natural total order (1. . On the other hand. they can only be multiplied (composed).6) is the "smallest"? So the arithmetic of knots has not helped us to classify them. Of course. But there is scant reason to talk of failure here: Schubert's theorem does not need any applications.3. Each positive whole number can be obtained by adding an appropriate quantity of ones ("trivial numbers for multiplication").

it is done by cutting the two strands at the point of crossing and reattaching them "the wrong way" (Figure S.' two ways of modifying a knot near one of the crossings of its strands. running the upper strand under the lower strand (Figure 5. which we will call the flip. Note that when the strands are not oriented. the flip can be performed by cutting the upper strand. consists in eliminating the crossing by interchanging the strands (Figure S.1).2b). "flipping" one of the crossings of a trefoil knot produces a trivial knot (the trefoil unknots-try to draw it. called smoothing. with a string.2a). the English mathematician John Conway discovered the fundamental role played in knot theory by two very simple "surgical operations. then reattaching it under the other strand. Of course. with a string. Conway's second little surgical operation. consists in transforming the chosen crossing (on the planar representation of the knot) into the opposite crossing. there are two ways of reattaching the four ends two by two (b or c).SURGERY AND INVARIANTS (Conway· 1973) In 1973. The first. and you'll see). for example. the flip can change the type of the knot. but the orientation of the knot 58 . that is.

2. )( ( ~ -. .SURGERY AND INVARIANTS 59 x x < (a) / ' ) x > Figure 5. (c) Figure S. . Smoothing (the strands are reattached the wrong way).1.~gi ~ I'~ A (b) -~ \11. . The flip (the upper strand becomes the lower strand)..

the molecule that carries the genetic code. which we will take up a little further in this chapter. and Topoisomerases Watson and Crick's seminal discovery of DNA. We know that this long. How is that possible topologically? Intricate studies have shown that enzymes called topoisomerases specialize in this task.3. as in Figure S. twisted double helix is capable of duplicating itself. W. the importance of Conway's operations extends far beyond the scope of knot theory. topoisomerases perform the three basic operations shown in Figure S. The description of this role-which is quite extraordinary-merits at least a small digression. More specifically. DNA.2a). in particular. then separating into two identical molecules that-unlike the two constituents of the parent molecule-are not bound together but are free to move. presented biochemists with a series of topological puzzles. H. J. among other problems. Flipping and smoothing play an essential role in life itself.60 KNOTS ~%%%%~~%~~% leaves us with no choice as to which pairs of ends should be reattached (this is dictated by the arrows. The flip and the smoothing operation were known and often used by topologists well before Conway. Alexander used them to calculate the polynomials that bear his name (to which we will return later). The reader will immedi- . Actually. Conway's contribution was to show that these two operations can serve as the basis for a fundamental definition of a knot invariant (the Conway polynomial). Digression: Knotted Molecules. as they have been routinely used by nature ever since biological creatures began to reproduce.

is also known in topology. Let us look more closely at how these strange enzymes affect long molecules. whose arrangement on the strand codes the genetic properties of the organism (a little like the way the arrangement of the numbers 0. 9 on a line of text gives the decimal code of a number). 2. C. that the DNA molecule appears in the form of a long double helix. which is called the twist. which is currently very useful in theoretical physics. It is well known that the double strands of DNA ordinarily have free .3.4 is a schematic representation of a fragment of a DNA molecule. each strand of which is constituted of subunits. the bases A. too. Remember. and G. .SURGERY AND INVARIANTS 61 Figure 5... First of all. especially DNA molecules. Figure 5. ately recognize operations (a) and (b): they are the flip and Conway's smoothing! The third operation (Figure 5. .3c). it refers to the mathematical theory of ribbons. note that the rearrangement of the strands occurs at the molecular level and is not visible: even the most powerful electron microscopes provide only indirect information. Topoisomerase operations on DNA. 1. T.

First. These molecules take part in three classical genetic processes-replication. and rejoining. Conversely. then reattach the cut so that the strands change place (Conway's flip).cytosine G . moreover. but it is not always so. transcription. the enzymes nick one of the strands. pass the second strand through the opening.4. . The topoisomerases playa crucial role in this whole process by carrying out the tasks of cutting. there are also molecules with closed double strands (the twice-coiled snake biting its tail) and single-stranded molecules with both closed and open ends. The structure of a DNA molecule.thymine C .adenine T . the double-stranded molecules form supercoils or tangles (this transforms these drawn-out objects into compact ones). and recombination.guanine Figure 5.Bases: A . rearranging. by means of two cuts and two repairs. ends.

the connecting wire gets more and more tangled. eventually becoming a sort of compact ball. The precise mechanism of the cutting. how supercoiling (and the reverse operation) of a closed double-stranded DNA molecule works. When the user twists the cord while returning the handset to the base.5.3c). the corresponding topoisomerase transforms the simple closed curve of DNA in the manner indicated in Figure 5. since the coiling shortens the distance between the phone and the handset. but here the result is useful. which we owe to Gauss. And we have some idea. In making twists (take another look at Figure 5. measures the number of times one of the strands wraps around the other). I In its normal state (not supercoiled). But it is known that there are different types of topoisomerases (they are not the same for singleand double-stranded DNA). whose dimensions are measured in angstroms. from the topological point of view. one of the results of the twist is to change the winding number of the two strands of DNA (this invariant. this result is rather annoying. supercoiling also transforms the long spiral into a compact ball. and rejoining operations is still not well understood. For the phone user.5 successive bases. In the case of DNA.SURGERY AND INVARIANTS 63 the enzymes are able to reattach the two strands "the wrong way" (Conway's smoothing). DNA supercoiling is analogous to what often happens to a telephone whose handset is connected with a long spiral cord. Note that. from the work of James Wang. because transforming the long molecule (several decimeters) into a tiny ball makes it easy for the molecule to enter the nucleus of a cell. rearranging. There . the DNA spiral makes a complete rotation with each series of 10.

64 KNOTS figure 5. but our goal is not to give a detailed description of the findings. double-stranded DNA molecule. the reader is referred to Wang (1994). in the negative to the most obvious question regarding knots. drawings (a) and (e) in Figure 5. and what role do they play in the theory? Roughly. when needed. For a more thorough introduction. which we have called the comparison problem: Given two planar representations of knots.5. knot invariants serve above all to respond. How do they appear. are many other topological phenomena that fascinate biologists.6 represent the same knot-the trefoil. can we say whether they represent the same or different knots? For example. Invariants in Knot Theory Let us return to the mathematical theory of knots. to speak-at lastof invariants. On the other hand. Indeed. this same figure shows how representation (a) can be converted to representation (e). A supercoiled. .

. one can calculate their invariants.SURGERY AND INVARIANTS 65 ~~%%~%%%~%:\~~'G%:3. they would have the same invariant!). When two planar representations are given (for example. a certain algebraic object (a number.6. that is.\(~%s\" Figure 5. Differing values for the invariant prove that the two representations do not define the same knot (if they did. a more clever person. Six representations of the same knot? any attempt to transform drawing (f) into a representation of a trefoil is doomed (try it!). a polynomial) so that this object never varies when the knot is manipulated. as in the first five sketches (a-e) in Figure 5.6). Now suppose that we have at our disposal a knot invariant.6. might be able to do it easily. f and e in Figure 5. or a luckier one. But how do we prove it? The fact that we have not succeeded in turning one sketch into another does not prove anything. a way of associating. with each planar representation of a knot.

66

KNOTS

~%%%%~%%%~~

For example, the calculation of Conway's polynomial (to be described in more detail later) of representations (a) and (f) in Figure 5.6 gives :il + 1 and -,x4 + x-I, respectively; therefore, the diagrams in question clearly represent two different knots. Before moving on to the study of Conway's invariant, let us try to find a numerical invariant for knots ourselves. The first idea that comes to mind is to assign to each knot diagram the number of its crossings. Unfortunately, this number is not an invariant: as one manipulates a knot in space, new crossings may appear on its planar projection, and others may disappear (for example, see Figure 5.6). Those who have read Chapter 3 may remember that the first and second Reidemeister moves change the number of crossings by adding ± 1 and ± 2, respectively. But it is easy, using this idea, to try to find a genuine invariant of the knots: just consider the minimal c(N) number of crossings of all the projections of knot N. This number (a nonnegative integer) is by definition an invariant (it does not depend on the specific projection given, since the definition involves all projections). Unfortunately, it is useless for comparing knots, because it would work only if we could calculate c(N) from a given projection N. Since we do not presently have any algorithm for doing this calculation, we will move on to a more sophisticated but calculable invariant: Conway's.

Conway's Polynomial
To every planar representation N of an oriented knot, Conway associates a polynomial in x, noted V (N), which satisfies the three following rules:

SURGERY AND INVARIANTS

67

[nvariance Two representations of the same knot have the same
polynomial:

N- N

=>

V (N) = V (N)

(I)

Normalization The polynomial of the unknot is equal to 1 (it is regarded as a "zero-order polynomial"): V(O)

=1

(II)

Conway's skein relation Where the three planar representations N+, N_, and No are identical outside the neighborhood of one crossing
and have forms as indicated in the diagrams as follows:

then
(III)

(In other words, No and N_ are derived from N+ by a smoothing and a flip, respectively.) For example, when N + defines the trefoil knot, the Conway notation specifically gives:

The attentive reader will have noticed that in this case the diagram No (on the right side of the equal sign) no longer describes a knot: it consists of two dosed curves instead of one, and it is the diagram of a link (a family of curves in space that can knot separately as well as link to-

gether). But never mind that-Conway's polynomial is actually defined for all links (of which knots are just a particular case). From now on, we will write Conway's skein relation (and other similar relations) in the following symbolic form:

which means that there are three identical links outside the circular areas (bounded by the dotted lines) that each contain one crossing. The second and third links are obtained by flipping and smoothing the first link in this neighborhood.

Examples of the Calculation of the Conway Polynomial
One of the advantages of the Conway invariant is the ease with which it can be calculated. Here are some examples. Consider the link consisting of two separate circles. Then we have V(OO) = O. In fact:

XV(Cf~)(~)V(®) - xv(<r8D)~
~V(O) - V(O) j,g 1 - 1 = 0
Now assume that the links consist of two linked circles, known as the

Hop! link, H =

(jJ). Following Conway's notation,

and since V(OO) = 0 and V(O) j,g 1, we deduce that V(H) = x.

SURGERY AND INVARIANTS 69 Finally. Thus one obtains V(T) = il + 1. we return to equation 5. For that.6fis not a trefoil: .6£. In particular. and by virtue of rule II. where P is the knot represented in Figure 5. x = il. let us calculate Conway's polynomial for the trefoil T. to 1. by virtue of rule I. the flip and the smoothing) and classical algebraic operations (sums and products of polynomials). the calculation of Conway's polynomial for a knot (or link) appears to be a rather original mix of geometrical operations (namely. according to the preceding calculation. Expressed in this way. we now have formal proof that: (1) The two curves of a Hopf link cannot be separated: (2) A trefoil knot cannot be unknotted: (3) The knot represented in Figure 5. The reader with a taste for this type of thing will undoubtedly enjoy calculating V(P).1. the second term on the lefthand side is equal to V(O). is equal to x . the term in the righthand side. Discussion of Results What can we deduce from these calculations? Many things.

Does it always work? In other words. Thus we have in Conway's polynomial a powerful invariant that allows us to distinguish knots. Yet my little notebook computer.7. which has software for calculating Conway's polynomial. for the planar representations of Figure 5. But one can argue that what we have here is a general method that also works in more complicated situations..) = V(K2) ::. where intuition proves of no avail. For example.Of course. my intuition of space (fairly well developed) does not necessarily tell me anything about the knot(s?) they represent. the reader who has not yet been ruined by mathematics will say that there is no point in having a formal proof of something so obvious as statements (1). which proves that the knots represented by A and B are different. does the equality of the polynomials of two knot representations imply that they are two representations of the same knot? Does one always have V(K. Two representations of the same knot? .7. K. or (3). trumpeted after a few seconds' reflection that V(A) = 1 and V(B) = :xl + 1. (2). = K2? Unfortunately. the answer is no: a calcula- Figure 5.

In .SURGERY AND INVARIANTS 71 tion similar to this one shows that the Conway polynomial for the figure eight knot (Figure 1. M = Millet. They are H = Hoste. 2 The simplest way to define the Homfly polynomial P(x.2) is equal to X2 + 1: it is the same as that for the trefoil. But-the sceptical reader will counter-what tells us that the trefoil and the figure eight knot are not. The Homfly Polynomial "Homfly" is not the name of the inventor of this polynomial: it is an acronym for the six(!) researchers who discovered the same polynomial at the same time and published their results simultaneously (in 1985) in the same journal. F = Freyd. and III with Pin the place of V and with the following modification to the skein relation (axiom III): xp(X)-yp(X)=p()(). The Conway polynomial does not distinguish the trefoil from the figure eight knot. L = Lickorish.O = Ocneanu. y) (with two variables x and y) is to use Conway's axioms I. II. in fact. (III') The reader who has grasped how to perform the little calculations of Conway polynomials of knots will perhaps enjoy redoing these calculations with the new skein relation III' for the same knots and links. One example is Jones's famous two-variable polynomial (the topic of the next chapter) or the Homfly polynomial. We will only be able to answer it conclusively when we have an invariant more sensitive than Conway's. the same knot? Good question. and Y = Yetter. which can also be obtained using Conway's method and with which we will end this chapter. it is not refined enough for that.

That is why the search for a complete invariant continues in the fol· lowing chapters.8 shows two different knots that have the same Homfly polynomial. . the answer is no: Figure 5. he will then see that the Homfly polynomials for the trefoil and the figure eight knot are not the same. The Homfly polynomial is thus more sensitive than Conway's. But is it a complete invariant? Can it distinguish all nonisotopic knots? Unfortunately. Figure 5.72 KNOTS particular.8. Two knots that have the same Homfly polynomial.

and it exceeds. has the double advantage of being very easy and of clearly showing the relationship of the polynomial with statistical physics. this theory is far from elementary (see Stewart. the assumed mathematical sophistication of the wide readership I hope to reach. Of that there can be no doubt. braid theory) and especially to physics (statistical models and quantum groups). But it so happens that another approach to Jones's polynomial. It is on this branch of physics that I shall base an explanation of the theory. But the significance of the famous polynomial extends far beyond the scope of knot theory: it is popular because of its connections to other branches of mathematics (the algebra of operators. Unfortunately. as conceived by its author at the outset.6 JONES'S POLYNOMIAL AND SPIN MODELS (Kauffman· 1987) The discovery by Vaughan Jones of the polynomial that bears his name reinvigorated the study of knot invariants. by a long shot. 73 . and so I will begin with a few of its fundamental notions. that of Louis Kauffman of the University of Chicago. 1989). It thus makes sense to devote this chapter-a key one-to Jones's theory.

each atom (represented in the figure by a fat dot) is characterized by its interactions with its neighboring atoms (an interaction is represented by a line joining the two atom-dots) and its "internal state"-what physicists call its "spin:' a parameter l that has a finite number of values (in the model here. two).ES 2: e[s(a j). statistical models and in particular the famous Ising model have captured the attention of both mathematicians and physicists. . For the model to be wholly deterministic. while e[s(aj). The two spins in this model are up and down and are represented by arrows pointing up and down.1) where the external sum is taken over the set S of all states.. and k a coefficient called Boltzmann's constant (whose value depends on the choice of units).s(a)) (. What is it all about? It has to do with theoretical models of regular atomic structures that can adopt a variety of states.74 KNOTS Statistical Models For a good thirty years (and especially since the publication in 1982 of Roger Baxter's book on this subject). T is the temperature.1)./)EA (6. This is an expression of the form: Z(P) = 2: exp ~. . each state being determined by the distribution of spins on the atoms (a very simple example is shown in Figure 6. and the internal sum is taken over all edges (interactions).. s(aj)) is the energy of the interaction between the atoms aj and aj (which actually depends only on their spins). respectively. we must specify its partition function. At any given instant.

locally.JONES'S POLYNOMIAL AND SPIN MODELS 7S ~~~~~ Figure 6. in the case of the Potts model of freezing water. Using the partition function Z. we can calculate a given model's total energy and the probability that it is in a given state and.2. each intersection divides . study its phase transitions-for example. 2 I do not intend to delve further into the study of statistical models. such as the one shown in Figure 6.1. especially. The little bit I have just said about it will be enough for the reader to understand where Louis Kauffman got his funny idea of associating a statistical model with each knot. Kauffman's Model Take any (unoriented) knot. Look closely at one of the knot's crossings. its change from the liquid state (water) to the solid one (ice) and vice versa. The spin model.

2b). Let us denote by S(K) the set of all the states of the knot K.76 KNOTS the plane into two complementary angles.2. and defined by the equation: '" (K) = L.Ja a(. In other words. The type-A angle is the one we see to our right when we start moving along the upper strand. where a(s) and f3(s) denote the number of type-A and . as well as Figure 6.)-p(.2a. we could have written up or down next to each intersection. It will be denoted by (K). we can associate with each crossing the word up or the word down.) ( -a 2 -a -z)y(. (The direction chosen for crossing the intersection can be either of the two possibilities: the resulting type of angle does not depend on this choicecheck that!) In Figure 6. We will need to make this choice right away.2) where the sum is calculated for all the possible 2" states s E S(K) of the knot K.l-I (6. To completely define the Kauffman model associated with the knot K. called the Kauffman bracket. for a given knot. we can choose at each intersection what might be called a Kauffman spin. This notation also has the advantage of dearly indicating the choice between one of two ways of smoothing a crossing (of an unoriented knot) by exchanging strands "following the little stick" (Figure 6. but I prefer to draw a little stick inside the chosen angle (look again at Figure 6. A knot with n intersections thus has 2" possible states. we need only define the corresponding partition function. and B angles are left blank. We say that such a choice (at all crossings) is a state of our knot.3b). A angles are shaded. Thus. one of which is type A (or up type) and the other type B (or down type). To represent the knot in a specific state. and then (after we cross over the lower strand) to our left.

State of a knot. illustrating angles of type A and B. .3. A ((=3 {J=ly=2 A Figure 6. Smoothing a figure eight state.JONES'S POLYNOMIAL AND SPIN MODELS 77 ~%%~~~% (a) Figure 6.2.

Without going into details.3 shows a possible state for the figure eight knot diagram (a) and the result of the corresponding smoothing (b). that is. Let us begin by comparing Figures 6. One might ask where Kauffman got this unusual formula (which is quite unlike its prototype in equation 6.78 KNOTS type-B crossings. let us pause a moment to compare the result we get with a classical model. a(s) = 3.1 and 6.2 is also valid for links of more than one component. /3(s) = 1. for this example. Figure 6. I would simply say that he found it by backward reasoning (and trial and error). Note in addition that equation 6. Be that as it may. whereas y(s) denotes the number of closed curves obtained when all the knot crossings are smoothed following the little s-state sticks. In this way. we get a polynomial3 (in a). such as the Potts model. respectively. by starting from the result he wanted to obtain. and y(s) = 2 (two closed curves appear after all the crossings are smoothed). Consequently. the application of this formula is very simple (though very laborious if the knot has many crossings).1). the graphical similarity re- .2. Obtaining the value of Kauffman's bracket for the figure eight knot diagram requires drawing all 16 possible states of the diagram (16 = 24) and summing the 16 results for the equation above (which describes just one of the 16 states). Aren't they awfully similar? Of course. Before continuing with our study of the Kauffman bracket. which is the value of the Kauffman bracket for the diagram of the given knot.

and Kauffman's equation (6. The three principal rules are the following: (I) (KU O) = (-tal . indicates how the value of (K) changes when one adds to it a trivial knot unlinked with K: this value is multiplied by a coefficient equal to (-tal . which give the partition functions of the models. in which K is any knot (or link).2) has no physical interpretation.2). are totally different.JONES'S POLYNOMIAL AND SPIN MODELS 79 ~~%%~~% sults from a judicious choice of the knot shown in Figure 6. Moreover. the zero-order polynomial with the constant term 1).a. On the other hand. Rule II.2.2)(K) (II) (III) (O) = 1 Let us begin at the end. but generally one can say that the state of the knot and the state of a planar regular atomic structure are more or less the same.1 and 6.a. The third and simplest rule tells us that the Kauffman bracket for the diagram of the unknot 0 is equal to 1 (that is. we will see later on that true statistical models (in particular Potts's model) can in fact be used to construct other knot invariants. equations 6. Properties of the Kauffman Bracket Our first goal is to indicate certain properties of this bracket to see how to deduce an invariant of knots from the bracket.2. . Kauffman's model is thus not a "true" statistical model-which does not in any way diminish its usefulness for knots.

.. the bracket of the link of N disjoint circles-which is obviously equal to (-a 2 . '" . .. Rules I-III make child's play of calculating a Kauffman's bracket from the diagram of a knot (or link): just apply rule I (taking care to note the intermediate equations obtained) until all the crossings are gone.80 KNOTS ~~~~~~~~~~~'C\1. is the fundamental rule underlying Kauffman's theory and this chapter-describes the connection between the brackets of the three links (or knots) symbolized by the three icons: which differ only by a single small detail." {X:..~% The first rule-which. the knots considered by Kauffman are not oriented (there are no arrows). like Conway's skein relations. nothing more than a very simple equation related to a local surgical operation. . More precisely. Having said that. despite its simplicity. rule I of Kauffman's theory is. Let us simply recall what the icons for the skein relations look like: (X) . there is only one type of crossing (two according to Conway)... but two ways of smoothing (Conway's arrows impose a single smoothing. Those who have read the chapter devoted to the Conway surgical operations will doubtless have noticed the analogy that exists between Kauffman's rule I and skein relations. :::5"C .. the icons denote three arbitrary links that are identical except for the segments of the strands indicated inside the dotted circles....a-Z)N-I. then calculate.. the same for the two different crossings).. Consequently. with the help of rules I and II..' How are they different from those in rule I? First.

the Kauffman bracket can only be used in knot theory if it is invariant.I )( -a2 - a. we get: For trefoils. For example. (00) = a(O) + a-I(OO) = a . (00) = -a- 3 For the Hopf link.JONES'S POLYNOMIAL AND SPIN MODELS 81 %~%~%%~~~~%%%%~%%%%%~ Then use that value to find the given knot step by step. using the preceding calculations.2 ) = -a- 3 In a similar way. we get: (0) (00) =1 = (-a 2 - a.2 ) Following rule I and the preceding result. for the unknot (using rule III) and the trivial twocomponent link (using rule 11). that is. still using the preceding results. we obtain these formulas: Of course. employing the intermediate equations. 1 + (a. if two diagrams of the same knot always have the .

we get: Comparing the first and the last member of this series of equalities. The reader will perhaps remember that there are three of those moves. This basic question is treated in detail in the following section. Using rule I several times and rule II once. Invariance of the Kauffman Bracket Thanks to Reidemeister's theorem. The reader who has not (or who does not like mathematical proofs) will miss little by going directly to the subsequent paragraphs. The attentive reader will have noticed the miraculous disappearance of the . where I finally introduce the Jones polynomial. Let us begin with the second move.82 KNOTS same Kauffman bracket. O 2. proving the invariance of the brack~t requires only showing that its value does not change when the knot (or the link) undergoes Reidemeister moves. we see that the invariance with respect to O 2 has been established. which is intended more specifically for those who have read the chapter on Reidemeister moves.1 in Chapter 3. take a look at Figure 3.

2 for Kauffman's bracket is motivated precisely by this calculation. Of course.3) Obviously. we can write: (6.g::) = (§) Comparing the righthand sides of the two equalities in equation 6. the most complicated of the Reidemeister moves. we obtain: (®) = (. But it is precisely this equality that expresses the invariance of the bracket with respect to Q 3! . Inspired by our little victory. That is therefore also true for the lefthand sides. and 0 for the undesirable icon:-g::. Now. this is no coincidence: the selection (a priori bizarre) of the coefficients in equation 6. Still using the basic rule I. twice applying the invariance relative to Q 2 (which we have just demonstrated).3 shows that they are equal term by term.JONES'S POLYNOMIAL AND SPIN MODELS 83 ~~~~~ different powers of a to give 1 for the coefficient at the desired icon ) C. (~)=(®) since these two diagrams are isotopic in the plane. let us move on to verifying the invariance relative to Q3.

%% A Little Personal Digression God knows I do not like exclamation points. Or better yet (in the case of a discovery).a.) = A( St. Using rules I and II. the simplest of the three. we get the coefficient a. the euphoria that the conductor must experience when all the musicians and the choir. . For example: . QI. we obtain: ( :. Invariance of the Bracket (Continued) To prove the invariance of Kauffman's bracket. For everyone else: the emotion a mathematician experiences when he encounters (or discovers) something similar is close to what the art lover feels when he first looks at Michelangelo's Creation in the Sistine Chapel.:ji·.3 ).···) . in the same breath that he instills and controls. I generally prefer AngloSaxon understatement to the exalted declarations of the Slavic soul..%~~~~.84 KNOTS ~5\l..··. it remains only to verify its invariance relative to the first Reidemeister move. ~. Failed again! The Kauffman bracket is not invariant relative to Q I and thus is not an invariant of knot isotopy. Why? Lovers of mathematics will understand. Yet I had to restrain myself from putting two exclamation points instead of just one at the end of the previous section.~%5\l. repeat the "Ode to Joy" at the end of the fourth movement of Beethoven's Ninth.)(') = a('o .5\l. where A = a(-a2 .• ) + a-I( .2) + a-I = -a3• Disaster! This blasted coefficient a3 has refused to disappear (for the other little loop.

and like many other mathematicians. The Jones polynomial and skein relations had made their appearance. and thus we should get: ()(») =()(») =(0) =1 Should we plunge into despair? Another Little Personal Digression That's exactly what I did fifteen years ago when I was working on these same questions. and I remembered having been stymied by this same move 0 1 (unable to get rid of a stubborn coefficient that just would not disappear) and dropping it. Later. I was playing with variants of these relations in the hope of finding invariants more sensitive than Jones's. we will have to add a supplementary factor to our bracket whose purpose would be to rid us of this tiresome a:!:3. But how? . Kauffman's Trick and Jones's Polynomial The starting point is obvious: since the coefficient a:!:3 refuses to go away.JONES'S POLYNOMIAL AND SPIN MODELS 85 ~~%%%%%%%%~%~~~ But the two diagrams both represent the trivial knot. among my scribbles. I retrieved some formulas very close to Kauffman's rule I. But Kauffman persevered. even if perseverance doesn't always pay off-I am not at all sure that by continuing I would have managed to discover the fantastic little trick that allowed Louis Kauffman to succeed. Budding young researchers will grasp the moral of the story.

and everything goes smoothly with 0 1 as well (lazy readers will have to take my word for it).86 KNOTS ~~%%%%%%%~% ("i"' ) Let us calion a classical tool of knot theory. the writhe w(K) is the integer equal to the difference between the number of positive crossings and the number of negative crossings. I am going to take advantage of the fact that the basic rule (I) of I I . the writhe. which does a superb job of killing off the annoying coefficient a±3 resulting from move 0 1.4) where the non oriented diagram K is obtained from the oriented diagram Kby forgetting its orientation (erasing the arrows) and where ( . we have seen that the bracket (. let us now define the Jones polynomial4 of an oriented knot (or link) K by writing: I KI) (6. move 0 1 changes the writhe: it adds 1 or -1 to it according to whether the eliminated loop is negative or positive (:"i":)· X(K) = (_a)-3w(K)( Still following Kauffman's lead. ) therefore follows from Reidemeister's theorem. (I leave to the "mathematized" reader the pleasure of unraveling the details of this sophisticated murder. Before describing what Jones's polynomial contributes to knot theory. which is defined in the following way: for every oriented knot K. It is easy to see that the writhe is an invariant of the Reidemeister moves O 2 and 0 3. ) is the same Kauffman bracket that has caused us so much joy and sorrow.) It is now obvious that the Jones polynomial is an isotopy invariant of knots (and links). Kauffman's trick is this factor (-a)-31<\K). worthy of Agatha Christie. the isotopy invariance of X( .) as well as the factor (-a)-31<\K) are invariant with respect to Reidemeister moves O 2 and 0 3. Indeed. In contrast.

Rules for the Jones Polynomial I have just demonstrated the first fundamental rule of Jones's polynomial: (1) Two diagrams of the same knot (link) have the same Jones polynomial. only a few years ago. an insignificant-looking little formula whose fundamental character was immediately obvious. The second fundamental rule-whose proof derives from a fairly easy calculation based on the "Celtic rule" (I) of Kauffman's bracket and equation 6.JONES'S POLYNOMIAL AND SPIN MODELS 87 ~%~~~%%%%%~%% Kauffman's bracket is still fresh in our minds to linger a little on the history of this equation. A New Digression-On Menhirs No mathematician would deny Louis Kauffman the honor of having invented rule I.5).4-is the skein relation for the Jones polynomial: . Yet. Kauffman learned that he was not the first to come up with the formula: a specialist of ancient Celtic culture explained to him that sculptors who worked on menhirs six thousand years ago used exactly the same rule to alter the structures of connected ribbons (thus of knots and links) that decorated these burial stones. Readers will find motifs from rule I (can I still call it Kauffman's rule?) in the links of the ribbons carved on the menhir shown in the Preface (Figure P.

) Let us do the calculation for the trefoil (to simplify the notations.2 )X(K). I have written a.88 KNOTS The two other rules are obtained directly from rules II and III for the bracket: (3) X(KUO) = (-a .q-1/2) = q-'/2 X(&) = q-2 = q-I + q-l(q-1/2 + q-S/2)(qI/2 _ + q-3 _ q-4 q-1/2) Readers who have developed a taste for these calculations can check that the same result is obtained if we use equation 6. (In fact.4 = q). These rules are sufficient to calculate the Jones polynomial for specific knots and links. We have obtained the trivial knot and the Hopf link. . for the trefoil: q-l(qI/2 . (4) X(O) = 1. X( (0)) = q-2(qIl2 + q-1/2) = _q-1/2 _ Thus. Let us do the calculation for the latter: Following rules (3) and (4).4 and the preceding calculation for the Kauffman bracket of the trefoil. one can show that rules (1)-(4) entirely determine the Jones polynomial.a.

'(.~%%%%Sl. Do not think that proving this fact has no purpose other than to satisfy our mathematical pedantry. Jones himself and his followers found new versions of his polynomial that were even finer.8). .JONES'S POLYNOMIAL AND SPIN MODELS 89 S\. This application (which caught the attention of specialists in knot theory) led Vaughan Jones to hope that his polynomial would be a complete invariant.\hS\'~h%~'S\('~S\\' Similar calculations show that the knots in the little table of knots presented in the first chapter are all different. for example. Nonetheless. the right trefoil from the left trefoil. based on totally different ideas. Another effort. It distinguishes. When Jones's polynomial appeared.lh~(. and a closer comparison of the diagrams of these knots (which look quite different) showed that they were in fact isotopic diagrams (diagrams of the same knot): the table was wrong. among others. Unfortunately. Figure 5. is described in the next chapter.:~3. This result was suspect. Having said that. Moreover. its calculation for the knots with 13 crossings or fewer gave different values for all the knots. for knot theory. at least for prime knots. the role of the Jones polynomial remains very important: it is a sensitive invariant-more sensitive. I must point out that no one has succeeded in finding a complete invariant along these lines. for example. which Alexander's polynomial cannot do.'%~%~:\. except for two specific knots with 11 crossings. than Alexander's polynomial. that wasn't to be: there are nonisotopic prime knots that have the same Jones polynomial (see.

and consequently a specialist in the theory of singularities (better known in the West under the media-friendly term catastrophe theory).)-. The concept is of disarming simplicity. "Since the singularity doesn't exist.FINITE-ORDER INVARIANTS (Vassiliev· 1990) Victor Vassiliev should never have worked on knots.<::). these differ from true knots in that they possess double points. Together with proper knots. without the least hint of a catastrophe. he was unable to apply the techniques of this theory directly to knots-objects with a regular local structure. You might say that moving a true knot in space produces a "catastrophe" 90 . it must be invented:' Be that as it may. Perhaps a wise humanist whispered in his ear. ((Z)) or ( :. smooth and continuous. the appearance of double points barely differs from that of crossings. X. A student of Vladimir Arnold.::). Vassiliev did invent it. where one part of the knot cuts another part transversally (::. Vassiliev explains. one must consider singular knots. In the planar representation of a knot.

2.. the stratified set <ji :::> Io U II U I2 U . C. double points. The points of <ji thus represent knots (singular or ordinary).. but one that may be different from the initial knot.. As an example..1 shows how the trefoil knot is changed (as it undergoes a catastrophe) into a singular knot with a single double point and then becomes a trivial knot. the mobile point cuts through the stratum II (the stratum of the singular knots having exactly one double point) . the knot becomes singular. shamelessly employing crude drawings..... At the first catastrophic moment (when the double point 1 forms on the figure eight knot H).. constituted of singular knots with 1. which show the process of deformation of a knot (the "figure eight knot") in Euclidean space 1R3. .. 3. around the square. Inside the square.FINITE-ORDER INVARIANTS 91 when one part of the knot crosses another. Ordinary knots thus form a subset of <ji denoted by Io. 2. A . we see the path taken in the space <ji (a "symbolic" representation of the same deformation) by the mobile point H. G... at that instant. D. I 2 I 3.. whereas the others form what is called the discriminant I.... 0 corresponding to the knot undergoing deformation. is of infinite dimension. Figure 7. so it is difficult to visualize. we see more "realistic" representations of some of these knot-points... I shall describe it very geometrically (though not very rigorously).. Nonetheless.. ••• .. then immediately reverts to an ordinary knot. It is in the vicinity of these strata that we are going to pursue our study of the invariants of knots. Unfortunately. This is bro. Vassiliev jumbles together in one set (denoted by <ji) ordinary knots with singular knots.. respectively. where the space (of infinite dimension!) <ji will be represented by a square: the one at the center of Figure 7... ken up into strata II. F. B.. which may possess any (finite) number of double points.

..2. The trefoil becomes singular. then unknots. Deformation of a knot in space 1R3 and in '?l'.Figure 7.. .. ~ @o i ~~--H~~~__~~ ~ t 2 (t L---L--_-'--_-'--_----JF A i d)D ~ (~c ~ ~B Figure 7.1. dJ G +-- ~ .

2-0 .. and follow the curve shown in Figure 7. a question arises. We will begin by giving a simple example of a specific Vassiliev invariant. begin at point 0 (corresponding to the unknot 0).. and the result is knot A.1 = -1 Right away. and denoted F) and continues to deform until the second catastrophic instant... To define it.. forms and disappears immediately. H-by cutting across the stratum l:l three times.. to obtain: vo(H) =0 + 1 - 1 . and the un knot changes into a trefoil..... Every Vassiliev invariant assigns to each knot (and in particular to singular knots) a certain numerical value. F . C) through stratum l:l' but at another location (D). the path shown in a dotted line in the figure? Happily... Thus one can easily calculate the value vo(H) of the selected Vassiliev invariant for the figure eight knot. This event corresponds (in the symbolic representation inside the square) to a new crossing (F .. when a new double point... D . once in the positive direction and twice in the negative direction.. which we will call Vo. the value of vo(M) increases by 1. G . B .. For that. the answer is . A).. assuming vo(O) = 0.FINITE-ORDER INVARIANTS 93 at point G. A .. point 2. D . which is in fact the trivial knot (0). we will assume it to be equal to zero for the trivial knot [vo(O) = 0] and postulate that each time the moving point M (which represents our knot) cuts across the stratum l:l in the positive direction (that of the arrows l issuing from l:l).. The knot then becomes trivial (unknotted. Finally.. for example. Is the invariant in question correctly defined? Doesn't its value depend on the choice of the path that connects points 0 and H? Will one get the same result if one takes.. B . a new crossing through another part of the same stratum occurs (C . C .

descending to the neighborhood of stratum l:2. we must move a little more deeply into the strata. for that.1 = -1) and in the general case. for example. true that it is not a complete invariant: it cannot tell all knots apart. in this case. and it took all of Vassiliev's cleverness (and very sophisticated techniques from algebraic topology) to prove it. it is not all that difficult to find another Vassiliev invariant that can distinguish two trefoils. for example). But this is not the only Vassiliev invariant. he considers the space of all knots (singular or not) in which knots are . Instead of considering knots individually (as Vaughan Jones does. however. Digression: Mathematical Sociology Vassiliev's approach to knots could be called sociological.2 So this Vassiliev invariant does a good job of its basic task: it succeeds in distinguishing certain knots. both in this particular case [Vo( H) = 0 . and also that the figure eight knot is not equivalent to the trefoil [vo(H) -1 :¢: 1 = = vo( C)). simple calculations show that the value of Vo for right and left trefoils is the same: the invariant cannot tell the difference between a trefoil and its mirror image. On the other hand. It is. There are infinitely many of them! In particular. we have seen in passing that the trefoil is not trivial either [since vo( C) = 1 :¢: 0).94 KNOTS yes. I want to take a little rest from mathematical reasoning by making a short digression about the method used here. But before proceeding from examples to general theory. that the figure eight knot cannot be unraveled. But this fundamental fact is not at all obvious. since its invariant is different from that of the unknot (-I:¢: 0). What do these calculations tell us? First.

This sociological approach in mathematics is not Vassiliev's invention. . and others. it seemed ready to devour mathematics whole. or other stratification. in a more striking manner. In the theory of singularities. and it remains the weapon of choice of Vladimir Arnold and his school.. (Fortunately. Alexander Grothendieck. Moreover.. by studying the strata l:" with greater and greater indices n. U l:2 U . Like Vassiliev. we are going to try to find all of these invariants. to lay the basis of category theory.) But let us come back to Vassiliev and his singular knots. readers who do not have it can skip directly to the conclusion of the chapter. Since that requires a certain ease with mathematical reasoning. Vassiliev does not go looking for just one invariant-he wants to find all of them. perhaps to exorcise it: in the beginning. . focusing only on their position in the social.. political. the sociological approach-which we will soon delve into in detail-seems especially fruitful. to define the entire space of invariants. Saunders MacLane. Much earlier it was used by David Hilbert to create functional analysis (functions lose their own personality and become points in certain linear spaces). that is. economic. In the same way that classical sociology makes an abstraction of the personality of the people it studies. In this specific situation. it is dear today that nothing like that actually happened. here the mathematical sociologist focuses only on the position of the point with respect to the stratification in the space gjP: gjP ::) l:o U l:.FINITE-ORDER INVARIANTS 95 only points and therefore have lost their intrinsic properties. All the information that is needed to define the invariants of knots can be found by exploring the strata l:" l:2. This theory was ironically called "abstract nonsense" by more classically inclined mathematicians.. by always going deeper and deeper. it is due to Rene Thorn. and by Samuel Eilenberg.

are isotopic if there is a homeomorphism of jR3 (preserving the orientation) that sends K. Ordinary knots (a) and singular knots (b). points where a part of the curve transversally cuts another part. to K2. (Then the terms knot or singular knot may stand both for a specific object and a class of isotopic equivalence-the reader can decide depending on the context. like ordinary knots (see Figure 7. ®@ o. respecting the arrows (and the cyclic order of the branches with double points).3). and K2. K.96 KNOTS A Brief Description of the General Theory To recap: a singular knot K is any smooth curve3 in space jR3 whose only singularities are double points (in finite number).) We will denote by l:o the set of (ordinary) knots and by l:n the set of singular knots with n double points. are oriented (marked with arrows). as for ordinary knots. Note that singular knots.-~ (a) ~ (b) tJ Figure 7.. . there is a natural relation of equivalence called ambient isotopy: two knots (possibly singular).3. For singular knots.

The function v must be defined on the equivalence classes (elements of ?f). but they are the same for the three knots.1) which means that the function v is applied to three knots identical everywhere but inside a little ball. 4 We say that a function 11: ?f -+ IR is a Vassiliev invariant (in the broad sense) if.%%~~%% Slightly displacing one of the branches of a singular knot near a double point smoothes the singularity with two different crossings: .. From definition 7..2) v(%) ..) ..v(®) + v(~) ."'o\) = ··b·· and the four-term relation: 0 (7.3) . the other negative..FINITE-ORDER INVARIANTS 97 ~~5\'h%%~~5\t.v(®) = 0 (7. ' . \:x········· :X······>-+'. where the knots appear exactly as shown in the three little dotted circles..5\t. for each double point in a singular knot... Recall that the smoothing to the left is called positive. )·· '.. it satisfies the following relation: (7. the parts of the knot outside the ball are not shown explicitly... " .1 we immediately deduce the so-called one-term relation: v(. X·· ..../... and thus v(K) = v(K') if K and K' belong to the same class. <+-:..

98

KNOTS

Actually, deducing equation 7.2 simply requires applying definition 7.1 once

and noting that the two little loops obtained by solving the double point can be eliminated isotopically, giving two identical knots for which the difference between the invariants will indeed be zero. Providing relation 7.3 requires smoothing the four off-center double points, which gives eight terms (each with a single double point) that neatly cancel out two by two. We say that a function 11: 'JF -+ IR is a Vassiliev invariant of order less than or equal to n if it satisfies relation 7.1 and vanishes on all knots with n + 1 double points or more. s The set Vn of all Vassiliev invariants of order less than or equal to n possesses an obvious vector space structure and has the inclusions

VO CVt CV2 CV3 ••• Lemma The value of the Vassiliev invariant of order less than or equal to n of a singular knot with exactly n double points does
not vary when one (or several) crossings are changed to opposite crossings. The idea behind the proof is very simple: according to equation 7.1, changing a crossing into an opposite crossing causes the value of the invariant

vto make a jump equal to v( [.X:;); but here this jump is

zero, since the argument of v in this case is a singular knot with n + 1 double points. An obvious consequence of the lemma is that zero-order invariants

FINITE-ORDER INVARIANTS

99

are all constants (in other words, Vo = IR, the set of real numbers) and thus uninteresting. Indeed, we know that all knots can be un knotted by changing a certain number of crossings; and since these operations do not change the value of any zero-order invariant (according to the lemma), their value is equal to the value of the invariant of the unknot. It can be shown almost as easily that there are no nonzero first-order invariants (in other words, Vo = VI), but, fortunately, the theory becomes nontrivial from the second order onward. To illustrate, we will distinguish among the elements of V2 a specific invariant, denoted vo, by taking it equal to 0 on the unknot and equal to 1 on the singular knot with the following two crossings:

£.

Using the lemma, one

can show that Vo is well defined. The calculation, which uses definition 7.1 three times and the equality vo(O) = 0 three times, is shown in Figure 7.4. In fact, what we have is the same invariant Vo whose value for another knot. the figure eight knot, was calculated (without the validity of the calculation being rigorously established) at the beginning of this chapter. But this time our calculation is quite rigorous. Readers

Figure 7.4. Calculating a second-order invariant of the trefoil.

100

KNOTS

who like these things can redo the calculation for the figure eight knot, as well as for other knots.

Gauss Diagrams and Kontsevich's Theorem
We are now going to divest ourselves of the geometry underlying our study of knot invariants and consider them in terms of a purely combinatory theory. The lemma of the previous section tells us that the value of an nth-order invariant of a knot with n double points is unaffected by changes in the crossings. Thus, its value does not depend on the phenomenon of knotting; it depends only on the order (a combinatorial concept!) in which the double points appear when following the curve of the knot. We propose to code this order in the following way. Consider knot K: SI ... 1R3 with n double points. Proceeding around the circle SI, we will label all the points sent to double points by the mapping of K, then join all the pairs of labeled points sent to the same double point by chords (Figure 7.5). The resulting configuration is called the Gauss diagram or the chord diagram of order n of the singular knot K. Figure 7.6 shows all the Gauss diagrams of orders n = 1, 2, 3. Note that all the nonsingular knots have the same diagram (the circle without any cords). A good exercise for the reader who is hooked is to draw eight singular knots for which the eight diagrams of Figure 7.6 are the corresponding Gauss diagrams (there are of course many knots that correspond to the same diagram). Now we will rewrite the one-term (equation 7.2) and the four-term (equation 7.3) relations in the language of Gauss diagrams. In this notation, a Gauss diagram actually stands for the value of an nth-order invariant (always the same one in the given formula) of one of the sin-

Gauss diagrams of order n :s 3. we get: 0=0 flj . I have omitted the terms 11( ••• ).FINITE-ORDER INVARIANTS 101 3 4 2 3 2 3 Figure 7.4) How are we to understand this notation? The first formula means that the value of each nth-order invariant for a singular knot with n double points that contains a little loop-with a double point (see equation 7. Gauss diagram of a singular knot. in this formula. gular knots with n double points that corresponds to the diagram (which specific knot is chosen is immaterial because of the lemma).QJ + fb -f6 = 0 (7. thus. but we will understand that the undrawn chords are identical for all the diagrams. we will not draw all their chords. In this way. When there are several diagrams.2)-is zero. . Figure 7.6.5.

Moreover.5) and since the third diagram vanishes-by virtue of the one-term relation (see the first equality in equation 7. one can consider the vector space 211 n = 1R(~n) of all finite linear combinations of Gauss diagrams 211 E ~n. Similarly. so that dimsI 3 = 1.6 expresses one of the two nonzero vectors remaining in terms of the other. We obtain a vector space. for n = 3.4)-we get: (7. the other n . but the one-term formula cancels the last three "basis vectors" of 2113 shown in Figure 7. for n = 3. More generally.2 chords are exactly the same in all four. moreover.1 chords of the diagram.102 KNOTS and I have not drawn the other n . it is understood that these supplementary chords cannot terminate on the little fat arc. but I have only drawn two in each diagram.6.and four-term relations and take the quotient of211 n by these relations. For example.6) This relation can be envisaged as an equality in the vector space IR of Gauss diagrams with three chords.) . it is understood that none of these chords can terminate on the little fat arc in the diagram.6). the four-term relation specifically gives: (7. equation 7. For example. (The reader can check that dimsI 4 = 3. which we denote by sIn. one has dim 2113 = 5 (Figure 7. the second formula describes the alternate sum of the values of the same nth-order invariant for four diagrams with n chords. one can then write for 211 nall the relations that follow from the one.

The proof of this theorem.2. which is even more remarkable than the theorem itself. this calculation is far from easy.0. the left trefoil. which means that two . 1994) is not limited to calculating the dimensions of Vassiliev spaces. The values of these dimensions are 1. . Conclusion: Why Vassiliev Invariants? Given Jones's polynomial invariants and those of his followers. respectively. Kontsevich's theorem The vector space VJV" .. Dror Bar-Natan at Harvard succeeded in finding the dimensions of the spaces .. 4. . the theory can also be used to find the values of specific invariants of specific knots. 9. But with the help of a supercomputer. and 44. 1. 3. is there really any need to invent more? Of course there is: all the polynomial invariants known to date are not complete.27.l(@) = 1. 1995).1 of nth-order Vassiliev invariants is isomorphic to the space . is unfortunately too long and difficult to be presented here (Bar-Natan. 1. 9. True. For example.l( 0) = 0 and V. defined by the formulas V..<An of Gauss diagrams with n chords modulo the one-term and four-term relations. I'll leave this calculation to the deft reader./ Vn 1 for n = 0. But from it we see that the study of the space of nth-order Vassiliev invariants (and the determination of their dimensions) can be reduced to a purely combinatorial calculation.FINITE-ORDER INVARIANTS 103 ~%%~%%%%%%%%%%%%% The main result of this combinatorial theory is that the space . can be used to show that the right trefoil is not equivalent to its mirror image. The usefulness of this combinatorial theory (a more detailed study can be found in CDL. 14.<A 4 completely describes nth-order Vassiliev invariants. the invariant v} E V}.<An $I! V.1.

In contrast. for example. but at a more elementary level. Another justification for Vassiliev invariants is their universality: all the other invariants can be deduced from them. Finally-and this side of Vassiliev's approach seems the most interesting to me. demonstrated that the coefficients of Jones's and Kauffman's polynomials can be expressed in terms of Vassiliev invariants. Similarly. for it is still developing-there are obvious and natural links (perhaps more than with the Jones and Kauffman polynomials) with physics.104 KNOTS nonequivalent knots can have the same polynomial invariant. that is. that the coefficient at X. of Columbia University. But this aspect of the theory is beyond the scope of this book. Today there are many other examples showing that Vassiliev's method makes it possible not only to obtain previously known invariants of knots. but also to define invariants-classical and newfor many other objects (and not only knots). Joan Birman and Xiao-Song Lin. this conjecture has neither been proved nor disproved. . for Vassiliev invariants. I would suggest that readers of Chapter 5 prove. But I will talk more about that in the next and concluding chapter. for each pair of nonequivalent knots KI and K2 there is a natural number n E ~ and an invariant v E Vn such that v(K1} ¢ v(K2 }. Thus.in Conway's polynomial V(N} for any knot Nis a second-order Vassiliev invariant. For the moment. the following assertion holds: Conjecture Finite-order invariants classify knots.

This final chapter. who understood everything but did not dare. Of course. one cannot make serious predictions about future scientific discoveries.KNOTS AND PHYSICS (Xxx? . deals with research still ongoing.' Is there "something in the air" today with regard to knots? It seems 105 . 2004?) This last chapter differs radically from the preceding ones. and the unbelievable failure of Carl Gauss. All the cases had to do with popularizing research results that have taken on a definitive shape. Their aim was to present the history of some of the (generally simple) basic ideas of knot theory and to describe the varied approaches to the central problem of the theory-that of classifying knots. But sometimes researchers working in a particular area have a premonition of an event. In everyday language. most often tackled by means of different invariants. which was anticipated by many others. "The idea was in the air:' The classical example-perhaps the most striking-is that of the independent discovery of non-Euclidean geometry by Janos Bolyai and Nicolai Lobachevski. even research that is only barely sketched out. on the other hand. we say of such a situation (and usually after the fact).

Coincidences Connections linking knots.Baxter equation . according to certain "specialists"). • Reidemeister's third move (the focus of Chapter 3). and quantum physics are based on a strange coincidence among five relationships that stem from totally distinct branches of knowledge: • Artin's relation in braid groups (which I talked about in Chapter 2). But first. which I talked about in Chapter 6). • one of the fundamental relations of the Hecke operator algebra. These coincidences (visible to the naked eye without having to understand the relationships listed below in detail) are shown in part in Figure 8. At the left of the figure. nor to name the future Lobachevski.1. I will return briefly to predictions at the end of the chapter. • the Yang-Baxter quantum equation (which governs the behavior of elementary particles in certain situations). statistical models. braids.106 KNOTS to me that there is. • the classical Yang-Baxter equation (one of the principal laws governing the evolution of what physicists call statistical models. I am not going to chance naming the area of mathematical physics where the event will occur. nor predict (at least in any serious way) the date of the discovery: that is why the title of this chapter refers to Xxx with a question mark as the future discoverer and the fictitious date of 2004 (the end of the world. I wish to explain the sources of the remarkable symbiosis that already exists between knots and physics. we see the Yang.

Figure 8. in algebraic form {bib j + I bj = bj + I bib. the Englishman W. as are the two drawings (look closely!). tJ>~ i · · · . the Frenchman Pierre Vogel. II ~f<1 ~ II . and Vladimir Drinfeld. the Russians Vladimir Turaev. and at the right. . a drawing showing Reidemeister's third move. . B.KNOTS AND PHYSICS 107 ~~~~~~~~~~~~~ b. The two equations are in fact similar (just replace b by R. Raymond Lickorish. the American Edward Witten. + d and in graphical representation. at the center Artin's braid group.1. Three aspects of a single relationship. Nickolai Reshetikhin. A strange kind of statistical model devised by Louis Kauffman en- . or vice versa). Oleg Viro.+1 1----. b. RjR j + IR j = Rj + IRjR j + I. i · b. It was while exploiting these coincidences that the New Zealander Vaughan Jones. and others discovered certain (profound? fortuitous?) connections between knot theory and several branches of physics. i b.

In Kauffman's approach (a version of which is presented in Chapter 6). it is the third Reidemeister move that plays the key role. Turaev discovered a whole series of polynomial knot invariants. Mathematicians have no specific object of study in material reality. natural or social. Should I go on? Wouldn't it be more reassuring to offer an explanation more specific and logical than "coincidence" for all these interdisciplinary links? Unfortunately. 2 "Everything" because one can apply them to anything. such as numbers. polynomials. Yet there is indeed a more general explanation in the context of connections between mathematics and reality. if a specific explanation exists. I am not going to try to . which allowed him to rediscover his own polynomial in another way. "Nothing:' because in mathematics one studies only abstractions. Digression: Coincidences and Mathematical Structure All the sciences. What is the object of mathematics? The response is paradoxical: everything and nothing. Using certain solutions to the Yang-Baxter equation. have an object: they purport to describe a certain part of reality. geometrical figures.108 KNOTS abled him to describe the knot invariant actually discovered earlier by Jones-the famous Jones polynomial. any object that has the same structure as the abstraction in question. Jones's original definition (not elaborated here) was based on braids and Hecke algebras (and thus on the coincidence between the Hecke and Artin relations). differential equations. I do not know it. of real life. Jones described a version of Potts's model (a statistical model very different from Kauffman's) based on the Yang-Baxter equation.

crystals) made up of atoms (with spins. Thus. Statistical Models and Knot Polynomials I talked about statistical models at the beginning of Chapter 6. A (perhaps unexpected) consequence of this state of affairs is the importance of coincidences: if the structures of two objects happen "accidentally" to "coincide" (even if these objects have completely different origins). they are described by "the same mathematics.KNOTS AND PHYSICS 109 explain the meaning of the expression that appears above in italics. if the trace of an operator belonging to a Hecke algebra has the same properties as those of a knot invariant. for example (looking at Figure 8. why not produce a knot invariant by using this trace (which is what Jones did)? And if quantum particles. Recall that they have to do with regular structures (for example. Such a system X must have a partition function Z(X) (which is the sum over all possible states of X of certain expressions that depend on the energies oflocal interactions). in the hope 3 that the reader will understand.Baxter equation.1). that the Yang-Baxter equation has "the same structure" as the third Reidemeister move. the general digression is therefore ended. this function permits one to calculate the principal . why not invent a theory of quantum particles using knot invariants (as did Sir Michael Atiyah. satisfy an equation that coincides with the Yang.' by the same theory. like knots. about whom I shall say more later)? We have returned to concrete physical considerations related to knot theory. say) that have simple local interactions (symbolized in the figures by the line segments joining the interacting atoms). in particular Ising's and Potts's models.

this function does not correspond to any real statistical model-rather. When the model is in a well-determined state s E S (S denotes the set of all possible states).110 KNOTS global parameters of the system (temperature. we saw how a kind of partition function allowed us to define and to calculate the Jones polynomial for knots. define the state of the system as an arbitrary function that assigns a spin to each atom. it is the fruit of the fertile imagination of Louis Kauffman. which physicists call up and down. the (local) interaction energy E[s( VI). the spin taking only two values. without going into too much detail.2. Actually. take the black regions for the vertices of the graph (or the atoms of the model) and join two vertices by a line or an edge (the interaction) if the black regions possess a common crossing.2. as noticed by Jones himself. total energy) and to study phase transitions (such as passage from the liquid state to the solid state). Moreover. Given a planar diagram of a knot (or a link). In Chapter 6. Next. you do that by alternately painting in black and white the parts of the plane delimited by the knot projection (taking care that the outside part be white). for directly constructing his polynomial. begin by drawing its dual graph (or dual statistical model of the knot). as shown in Figure 8. declare the edges (interactions) to be positive or negative according to a convention that the reader can identify by looking at Figure 8. s( V2)] of the two atom-vertices joined by the edge [VI' V2] is assumed to be equal to ± 1 if they have the same spin and to a:!: I if the spins are opposite. But it is most surprising that there exists a real statistical model. My immediate aim now is to describe this construction. with a genuine partition function. choose the plus sign or the minus sign depend- .

ing on whether the edge (interaction) is positive or negative.s(v j)] sES (v.4 Thus we see that Potts's model for freezing water brings us rather .v.KNOTS AND PHYSICS 111 Figure 8. (It is this specific choice of the interaction energy of the atoms that is unique to Potts's model.2.) That done. the model of the phase transitions between water and ice. we can define the partition function of the model by the formula: Z(K) = (~ )In E[s(vj ).. Dual graph of a knot. lEA where A is the set of all the edges. here. a is the name of the variable of the polynomial (in a and a-I) that you want to obtain. Deriving Jones's polynomial from this partition function requires applying to it a variation of "Kauffman's trick" (described in detail in Chapter 6).

recall that Jones's original constructionpurely mathematical-preceded the "physical" construction just described. Let us move on to another coincidence.) Of course. where it was used to define Jones's polynomial. a priori. it is statistical physics that produces a construction that can be applied to mathematics. the most famous invariant of knots. one that really deals with an application of knot theory to physics. Kauffman's Bracket and Quantum Fields I described Kauffman's bracket in Chapter 6." Curious analysis! Knot theory has nothing to contribute to physics here-on the contrary. .112 KNOTS easily to the most famous invariant of knots. very far apart. In analyzing this construction in surveys or popular articles. It has already been noted that this equation (which we have no need of here) has no physical interpretation. what is important here is not the rivalry between physicists and mathematicians. defined by a precise formula inspired by the partition functions of statistical models. at least in the framework of a realistic statistical model. We are going to see that it can be used for something else entirely. mathematicians have a tendency to enthuse over «the application of knot theory to statistical physics. Recall that this bracket associates to every planar knot diagram K a polynomial (K) in a and a-I. which are. but this unexpected coincidence between two areas of knowledge. (To spare the self-esteem of mathematicians. It is in another branch of physics-topological quantum field theorythat it plays a role.

instead of a plane. In this context. a typical ex- . where 2 is the dimension of "space" and 1 that of "time. electromagnetic fields. Specifying this theory. Vogel and his coauthors constructed a whole series of examples of TQFTs. which in this context physicists call Wilson lines. the topological one. They are thus topological invariants. and so on) in the most general context. generally designated by the acronym TQFT. It was Witten's idea to use a generalization of the Jones polynomial (often called 5 the /ones-Witten invariant). on which the diagram of a knot (or a link) is drawn. In this context. I will restrict myself to the context that features the bracket.KNOTS AND PHYSICS 113 This theory. in which Kauffman's bracket actually plays the key role. There is no question here of explaining this theory and these examples-the mathematics required are too sophisticated. they must conserve the same values for every topological transformation of the coordinates.' these three coordinates being mixed. like knot invariants. This TQFT was simply a model whose dimension is 2 + I. Michael Atiyah (also a Fields medal winner. The strands of the knot may have ends at the boundary of the surface. the physical magnitudes that one studies-the observable5-Cannot depend at all on the system of coordinates under consideration. think of a surface with a boundary. Later on. as relativity demands. seeks to formalize the quantum version of classical field theory (gravitational force fields. but for earlier work) rethought Witten's model from a mathematical viewpoint and generalized it to create an axiomatic theory of TQFTs. he deserves the credit for finding it (and it won him the prestigious Fields medal) and for using it to construct a TQFT. The model is thus a three-dimensional one that may contain knots.

Each of these diagrams is associated with a polynomial in a and a-I that satisfies two very simple rules (already seen in Chapter 6): (KU O) = (-a 2 - a.114 KNOTS ample is shown in Figure 8. Physicists take them very seriously. Figure 8. Diagram oflinks on a bounded surface. that a special case of this construction (when the surface is a disk) gives the so-called Temperley-Lieb algebra. .3. and so on. Reidemeister. Baxter. I do not want to judge the interest of these TQFT models from the point of view of physical reality. Yang. an algebra of operators that satisfies the rules of Artin.Z) (K) Readers who recall this chapter will immediately recognize two fundamental properties of Kauffman's bracket.3. Heeke. to be done with coincidences. Note.

I will focus on their physical significance. The operations (of multiplication and comultiplication) must satisfy some very obvious axioms (such as associativity) that endow Q with what mathematicians call a bialgebra structure. A multiplication. has little appeal: a quantum group is a set of abstract elements that must satisfy a whole list of formal algebraic axioms whose real meaning is not very obvious. The comultiplication does the opposite: it associates a pair of elements6 from this set with a single element from Q-its coproduct.4. and to the splitting of a single particle into two. these two operations correspond respectively to the fusion of two particles into a single one. the study of which (in the context of its connection to knots) we will move on to next.KNOTS AND PHYSICS 115 but perhaps not as seriously as the (mathematical) idea of the quantum group. From the point of view of physics. despite their name. however. note that. the class of . 7 These axioms are not all that restrictive. First off. Quantum Groups as Machines for Making Invariants Quantum groups appeared twenty years ago and today are an object of intense study by both mathematicians and physicists." That means that two operations are given for any set Q: a multiplication and a comultiplication. quantum groups are not groups at all. they are algebras. and even "bialgebras. Their formal definition. I have tried to represent this correspondence graphically in Figure 8. of course. for example. and there are so many quantum groups that one is led to consider a narrower class. associates to each pair of elements a well-defined element from Qtheir product. Rather than trying to explain quantum groups in detail.

ending up on top)? No. The thing is that the set V of invariants (which is in fact a vector . ideologically close to catastrophe theory. one after the other.. The underlying physics is not evident in Vassiliev's work. and this. it is buried in the algebraic and combinatorial structures characterizing invariants.. both new and familiar. in the course of which the lower strand of a knot breaks through the upper strand. . Vassiliev invariants are obtained by applying a very general construction. the representations of these quantum groups allow us to define a lot of invariants. it seems. provides the link between quasi-triangular quantum groups and knots... The axiom of quasi-triangularity implies that the YangBaxter equation holds for this class. ~ / Figure 8. Can one give a fundamental physical meaning to the flip (the principal catastrophe..). quasi-triangular quantum groups defined by Drinfeld (another Fields medalist!). are the truly scientific way to mass-produce knot invariants..4. ~ . Quantum groups.. to knots. the reader will of course have guessed. More precisely. Product and coproduct of two particles.0 0 .. as it were.116 KNOTS ~ ~ . Vassiliev Invariants and Physics As we saw in the preceding chapter. at least not in any obvious way.

at the analytical level. But there. the interpretation of Vassiliev invariants by chord diagrams (which I talked about in the previous chapter). which are ordinary numbers).KNOTS AND PHYSICS 117 %~%~~%~%%~~%%%%%~% space) is not only endowed with a multiplication (obtained by multiplying the values of invariants. several even. right from the start. also still not understood. Dror Bar-Natan has exploited the fact that this formula is none other than a form of the classical . The latter is defined by means of the connected sum # of two knots via the following obvious formula: It is easy to see that these two operations make a bialgebra out of V. This bialgebraic structure is inherent in Vassiliev invariants. the algebra of Chinese characters (which until recently was called "the algebra of Feynman diagrams") is dose to physical theory. as its (former) name indicates. another contribution from Maxim Kontsevich (yet another Fields medalist!). Thus. at the combinatorial level. Next. this "very physical" structure (fusing and splitting of particles) obviates the need to go looking "outside" for another algebraic object (such as the quasi-triangular quantum group for Jones-Witten-type invariants) to make "physical" invariants. we are still at the stage of hope and speculation. First. and thus should have a physical interpretation. it is a generalization of the Gauss integral in electromagnetism. too. but also a comultiplication: Ll: V -+ V ® V. the Vassiliev invariants of a knot can be expressed via the admirable Kontsevich integral. is the four-term relation that I spoke about in Chapter 7. In particular. A final important point. also lends algebra a physical orientation. In a certain sense. But there is more. What interpretation? Nobody knows.

118 KNOTS Jacobi identity to construct Vassiliev invariants using representations of Lie algebras. According to Einstein. According to specialists in string theory. more generally speaking). we live in four-dimensional space-time. From the viewpoint of actual physical reality. Atiyah. many areas remain unexplored. For example. there is still a lot to do. Vogal. became the basis for physics-oriented theories. the significance of TQFTs (in the style of Witten. there are some less-studied "generalized knots. Will interest in the connections between physics and knots be short-lived? For specialists in knot theory. Yetter) remains unclear. there is still no unknotting algorithm simple and efficient enough to be taught to a computer. to say the least. in particular Kauffman's bracket. Crane. Will this coincidence between fundamental mathematical relations have a "physical" development? Conclusion: Nothing Is Finished At the beginning of this book. do not forget that aside from classical knots (threedimensional curves in space). Where are we now? Thomson's idea was ephemeral. Finally. For researchers in mathematical physics who look at knots from the side. Very recently. knot invariants. and many other important questions have been shelved. the propagation of a particle can be modeled by sur- . such as topological quantum field theory.' such as (two-dimensional) spheres in four-dimensional space (or surfaces. we saw how William Thomson's idea of using the knot to make a model of the atom almost a century and a half ago was the start of the theory of knots. particularly concerning Vassiliev invariants.

in the context of knot theory. and hope. .KNOTS AND PHYSICS 119 ~%~%~%%%~%%%%%%~%% faces. I hope that the reader (and myself as well!) will again experience. Is a quantum theory of gravitation hiding in there somewhere? Do Vassiliev invariants (which also must exist in this context) have a real physical interpretation? Research always begins with a question. To conclude. the incomparable joy of understanding inspired by a great discovery.

.

simpler and perhaps no less efficient. 2. 5. like that of theoretical physics. Another strategy. 3. he is indeed the Vandermonde of the determinant. 2. a symmetry with respect to a small circle whose center is situated in 121 . is to thumb the book and choose which chapter to read according to the most interesting pictures. scientific-sounding words. 1. Antoine is French for Antony. Here I mean precise mathematical terms and not descriptions of TV horror films. 2. Atoms and Knots 1. Differential geometry defines knots more neatly as "smooth closed curves. For those who know.' but this involves using the calculus. The first definition of a polynomial invariant for knots. Braided Knots 1. 4.NOTES Preface 1. Modern mathematical terminology. In this case the physical interpretation uses the ideas of statistical physics. This terminology derives from what geometers call an inversion (in our case. was based on mathematical ideas that were very sophisticated for the time: homology theory and covering spaces. due to Alexander. tends to favor everyday words over serious.

there are several effective ways of "coding" knots. and perhaps also William Thurston. Moreover. And Emil Artin himself. proof of the analogous assertion is much more difficult and requires special techniques. Nicholas Pippinger) implies that the number of Reidemeister moves needed for unknotting is bounded.122 NOTES one of the regions bounded by the given Seifert circle). So called because they show up in all groups. My computer understands it. This inversion sends the center "to infinity" (and transforms the region into an infinite region). Can the clever reader guess the principle behind the coding? Perhaps I need to say here that for a projection of a given knot. The reader may wonder how a computer can "see" knots.-1. recent work by three American mathematicians (Joel Hass. Jeffrey Lagarias. 3. the one I use in my unknotting software gives the following description of the left trefoil: 1 + -2 . which means that in principle this algorithm either unknots a knot or declares that it cannot be unraveled . 4. Vladimir Arnold. the basics of which were set down by Hassler Whitney between the First and Second World Wars and subsequently developed by Rene Thom. the number of different applications of (. the software may trigger a useless and infinite loop (in the computational sense of the term). and their followers. This knot is numbered 52 in the tables of knots (Figure 1. 5. Markov. Especially in "singularity theory" (often also called "catastrophe theory"). 3. In fact.h is finite. 2.6). 4. To the mathematically knowledgeable reader I should note that for a knot represented by a differentiable curve. Joan Birman. Planar Diagrams of Knots 1. Among whom are certainly A. A. not only in the group of braids Bn. For example. 3. 2. Otherwise.-3 + -1 . Actually.-2 + -3 . be careful that the software doesn't make the mistake of following a specific application of "h by the inverse of the same application.

S. 2. were defined as closed curves in three-dimensional space. remember. let us leave to him or her the task of modifying the definition so that it will not be open to criticism. say with confidence that. for which this construction works perfectly well. and hence this argument is important only from the theoretical viewpoint. in particular. 2. For the reader more familiar with mathematics. 4. The Arithmetic of Knots 1. the bound is huge. 4. if it exists. Knots. 5. or the type of knot. It would be safer to say that we do not know whether an appropriate operation of addition exists for knots. Surgery and Invariants An angstrom (A) is one ten-billionth of a meter. the geometric addition of knots is not simpleif it were. one that is based on the notion of homeomorphism. . and thus that the composite knot is not a true boxed knot. This acronym is a flagrant injustice to two Polish mathematicians. she/he should redefine the equivalence (isotopy) of knots and say that the knot. The pedantic reader will say that the box obtained is no longer cubic. For practical purposes. however. 3. a fast and usually effective unknotting software has recently been devised by the Russian I. Unfortunately. is an equivalence class. someone would have discovered it. Jozef 1. note that adopting the argument represented in Figure 4.NOTES 123 ~~~%%%%~%~% (since it hasn't succeeded in unraveling within the prescribed bound for the number of moves). By way of reward. That is correct. however.5 specifically requires using another definition of knot equivalence. Dynnikov. since we know only that those who have searched for it have not managed to find one. This idea will certainly occur to the reader who has assimilated the chapter on braids. We can.

this polynomial can contain negative powers of a. however. Generally. Or "intrinsic angular moment. Mathematicians call it a Laurent polynomial. X(·} is not the Jones polynomial." Of course. not to mention many Russians who didn't publish because they considered the polynomial to be just a variant of the Jones polynomial. Another acronym. From a mathematical point of view. but this is essentially only a change in notation." 2. called coorientation. 6. it must be emphasized that . Finite-Order Invariants 1. We use "flat water" not only to simplify the drawing but also because it is the model most studied by physicists. who made the same discovery at the same time but published later. there is a more realistic three-dimensional model.124 NOTES Przytycki and Pawel Traczyk. respectively. Lymphotu. But especially-as we will see later-we use it because the two-dimensional model relates to knots. 7. At issue here is a very theoretical model: "two-dimensional water. to obtain it requires changing the variable of the polynomial by writing q = a4. Strictly speaking. Jones's Polynomial and Spin Models 1. in such a way that the follOwing catastrophes (cutting across II) are positive and negative. 4. 2. He gave credit to the Poles and to others (U = unknown discoverers). was proposed later by Dror BarNatan. One chooses the direction of the arrows. but the acronym didn't stick. 3.

as though he had suddenly been struck blind? 2. but we know nothing of their actual configuration. 4. Knots and Physics 1. on which modelling hyperbolic geometry is mere child's play for a professional of Gauss's level. have missed the discovery. And lack of recognition drove Bolyai to drink. every smooth mapping K of the circle 51 into Euclidean space 1R3. The problem is that our argumentation is based on the configuration of strata as it appears in Figure 7. quickly became the laughing stock of his contemporaries. because they were discovered independently by M.2. respectively).) When Lobachevski's publication came out. at least for anyone used to reasoning mathematically. Mathematized readers desiring a rigorous account of the preceding calculations will find one a little further on in this chapter. Petersburg (but he published his results only much later). Vassiliev invariants are frequently also called finite-order invariants and sometimes Gusarov-Vassiliev invariants. Gusarov from St. That is. 5. The positive (or negative. Gauss discovered the first principles of hyperbolic non-Euclidean geome- try well in advance of Lobachevski and Bolyai but lacked the courage to publish this scandalous theory. N. Barring the conviction-from a strictly Platonic point of view-that ab- . respectively) resolution is well defined: it is the one for which the traveller moving along the upper branch (following the arrow) sees the arrow of the lower branch pointing to the left (or right. The others will have to take my word for it: the rigorous version of this calculation is really very simple. and not only because we have not proved that Vo is well defined. How could the brilliant Gauss. 3. with all the tools at hand.NOTES 125 we have not shown anything at all. Gauss had already created the differential geometry of surfaces. who did publish it. 8. (Lobachevski.

That is why I prefer the expression Jones-Reshetikhin-Turaev-Witten invariants. 6. Or a Hopf algebra structure. despite its excessive length. . 5. it is easy to show that the polynomial Z(K) is invariant relative to the second and third Reidemeister moves.126 NOTES 3. Witten's construction lacked mathematical rigor. stractions. are more real than the material world. A famous attempt was made by Nicolas Bourbaki. Turaev who succeeded (thanks to another approach) in providing a correct mathematical definition for these invariants. More precisely. Reshetikhin and V. 7. whereas the first move gives a superfluous factor that can be got rid of owing to another factor dependent on the writhe. Unfairly. Actually. I have nothing to say about it. exactly as shown in Chapter 6. It was N. 4. a linear combination of pairs of elements. which belong to the world of ideas.

Knots. "Theorie des n<:euds et enluminures celtes:' L'Ouvert. 1989. I. 117-126. J. 146:94. W. III. ed. C. pp. C. Walker. "Appendix. "Cat's Cradles and Other Topologies Armed with a Two-Meter Loop of Flexible String. "Hydrodynamics. S. W. New York: Freeman. I. 84. 105:245-375. v. 1985. Special issue. and I. 1995. Dehornoy. Jensen. Arnold." Acta Mathematica. C. 1994. Jaworski. 1997. 34:423-472. Get Knotted. Fun with String Figures. Providence: American Mathematical Society. J.WORKS CITED Adams. Wang." Topology. The Knot Book. and A. 29B:257-270. 1997. Stewart. 1971. W. Lando. I. vol. 1961. Stewart.]. Ashley.6:94-105. Haken. 1994. S. pp. 21. and S. V. D. and 3-Manifolds. Singularities and Curves." In Advances in Soviet Mathematics. 214:82-90. 68-74. II. V. 127 . W. 1944. Bar-Natan. D. no. Duzhin. 1996. CDL [Chmutov. "On the Vassiliev Knot Invariants. 252:138-143. "Vassiliev Knot Invariants. 1867." Scientific American. Rouse Ball." Advances in Pharmacology. Sossinsky. Links. W. Braids.. 1994." Proceedings of the Royal Society of Edinburgh. New York: Dover. Prasolov. "Theorie der Normalflachen. Mercat. I: An Introduction to DNA Supercoiling and DNA Topoisomerase-catalyzed Linking Number Changes of Supercoiled DNA. London: Pan Books.." Scientific American. New York: Doubleday. P. K. ''rart de tresser:' Pour la science. 1976. J. V. "The Hagfish. C. "Le polynome de Jones:' Pour la science. The Ashley Book of Knots.. Thomson. 1966.

Sign up to vote on this title
UsefulNot useful