Review (Fernand Braudel Center)

Research Foundation of State University of New York

What Is Chaos Theory?

Author(s): Ivor Ekeland
Source: Review (Fernand Braudel Center), Vol. 21, No. 2 (1998), pp. 137-150
Published by: Research Foundation of State University of New York for and on behalf of the
Fernand Braudel Center
Stable URL: http://www.jstor.org/stable/40241422
Accessed: 11-01-2016 04:37 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact [email protected].

Review (Fernand Braudel Center), Fernand Braudel Center and Research Foundation of State University of New
York are collaborating with JSTOR to digitize, preserve and extend access to Review (Fernand Braudel Center).

http://www.jstor.org

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
 Whatis Chaos Theory?*

Ivar Ekeland
SCIENTIFIC THEORIES

theoryis like Euclidean geometry:it is not a scientific

Chaostheory;it is, rather,a setof mathematical statements.
There seems to be a paradoxhere.Whatis moreassured;what
is a betterguaranteeagainstanyriskof errorthanmathematics? So
why thendeny to it the statusof scientific
theory? Well,a scientific
theorytellsus something abouttheworld,whilemathematics tellsus
somethingabout mathematics. Science entailsa riskof error,but
applies to thereal world. Mathematics is necessarilycorrect,butthat
certitudeonlyappliesto mathematics, and notto theusage one may
makeof it.
One expectsa scientific theoryto teachus somethingabout the
world around us, about one of the physical,biological,or social
aspects of thatmultiform realityin whichwe live. Galilean theory
says thatabsolute of
points reference existin space and ifone relates
themovementof a bodyto them,itsaccelerationis thequotientof
the forceexertedon it dividedby its mass. In physics,Newtonian
theorysaysthatbodiesare attracted to one anotherin directpropor-
tion to theirmass and in inverseproportionto thesquare of their
distance;developingthatidea, it can be demonstrated thata single
planet orbitingaround a sun describesan ellipse,and thisis in
keepingwithKepler's observations.In biology,Darwiniantheory
says thatanimal formsare not fixed,but thatspecies evolve by
naturalselection;thishas been understooddifferently dependingon
thetimeor theauthor,butwe continueto speakofitas thestruggle
forlife.

*
Originallypublishedin Le Chaosby Ivar Ekeland,in the seriesDOMINOS, Paris:
Flammarion,1995, 85-111. Translatedby RichardE. Lee.

review,xxi,2, 1998,137-50 137

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
138 IvarEkeland

Whatall of thesetheorieshavein commonis thattheyare true

or false;moreproperly,theyare situatedon a scale goingfromthe
rigorouslyexact to the completelyfallaciousand theyare appreci-
ated accordingto theirplace on thatscale. That appreciationis
partlysubjectiveand perpetuallyreassessed.Galileo,forexample,
struggledhiswholelifeto havehis ideas accepted,whiletodaythey
seemso naturalthatitis difficult forus to comprehendthoseofhis
predecessors. As forDarwin, hisideas stillprovokefierceopposition
on thepartofChristianfundamentalists and in certainstatesin the
UnitedStatestheteachingof creationismon an equal footingwith
the theoryof evolutionis obligatory.All of thissignifieswhat is
particular theory,thatitis permanently
about a scientific subjectto
a judgement:it is trueor false.The twogo together.A theorycan
onlybe truebecause it could also be false.The touchstonewhich
distinguishes theoryfromreligiousmythor rantings
a scientific pure
and simpleis thefactthatit can be refuted,thatis,it maybe called
intoquestionbyexperiments. The famousexperiment byMichelson
and Morleyshowedthatthespeed of lightwas thesame no matter
thepointof reference.This experiment led to theabandonmentof
the Galilean theoryof absolutepointsof referencefor Einstein's
specialtheoryofrelativity.To invalidatethetheoryofevolution,one
could constitutetwo populationsof animalsof the same species,
place themin differing climaticor biologicalconditions,and note
that,generationaftergeneration,individualsremainedidenticalto
one another even though the two populationswere separated.
Darwinhimselfconsideredthatsuchan experiment had been real-
ized on the Galapagos Islandswherelong-isolated animalpopula-
tionspresentedsubstantialmodifications withrespectto theirlikes
on terrafirma.
Bywayofcontrast, evidently,theidea thatGod createdtheworld
fivethousandyearsago just as we see it todayis intellectuallymore
comfortablebecause it does not run the riskof beingquestioned.
Whatwouldhave stoppedHim fromfillingcertainrockswithdino-
saur fossilsevenifsuchbeastshad neverexistedor keptHim from
creatingdifferent animalson the Galapagos than those on terra
As
firma? long as no one has inventeda timemachine,no conceiv-
able experimentcould everdifferentiate betweensucha worldand
one in whichdinosaurshad actuallyexisted.The scientistlivesin a
world where the only certitudeis provisional.His universeis a
cemeteryof mistakenideas and outmodedtheoriesand even the

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
 CHAOS THEORY 139

theoryhe maypresently be usingis onlytherewaitingfora better

one to takeitsplace.
But purelymathematicalstatements, like Euclideangeometry,
are not part of thisplay of trueand false,even if theyare often
designatedby the word "theory."To speak of the theoryof num-
bers1or the theoryof Galois2is not to talkabout somethingthat
could eventuallybe shownto be inexactor thatcould be put into
questionbysome kindof experiment. The equality,2 + 2 = 4, is not
of the order of physicallaw but of logical necessity.It can be de-
ducedfromothermathematical statements,suchas 1 + 1 = 2, and an
adequatedefinition ofthenumbers1,2, 3, and 4. It is nottrue,then,
in the sense thatit tellsus somethingabout theworldin whichwe
live,but in thesense thatit is consistentwiththerestof mathemat-
ics. It cannotbe refuted;whatexperimentcould everconvinceus
that2 + 2 = 5? It cannotbe contestedor improved.We do notexpect
some futuremathematicians, more intelligentthatwe, to finda
+
bettervalue for2 2 than4. No law has been passed in theUnited
Stateson thevalue of 2 + 2, and theauthorofan alternative theory
accordingto which2 + 2 wouldnot equal 4, wouldriskbeingcom-
mittedto an asylum.

MATHEMATICAL THEORIES

In mathematics, theword"theory" designatesa setofstatements

whichhave in commoneitherthesame domainof application(the
theoryof numbers)or can be attributedto the geniusof a single
person(thetheoryofGalois).These are purelylogicalconstructions,
limitedto deducingthenedessary consequencesfromcertainmathe-
maticalstatements whichare consideredfundamental (theaxioms)
butwhichdo notpretendin anywayto be applicableto thephysical,
biological,or social world. In this sense, theyare not scientific
theories. Euclidean geometry,for example, deduces necessary
consequencesfromcertainaxioms,nine to be exact,includingthe
famousaxiomofEuclid(througha pointpassesonlya singleparallel
to a givenline). It tellsus, forexample,thatifwe accepttheseaxi-

1Mathematical
disciplinewhichstudiesthe propertiesof integers.
2
Theorywhichstudiesthe propertiesof algebraicequations; continuationof the
workof Nils Abel (1802-29) and ÉvaristeGalois (1811-32).

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
140 IvarEkeland

oms, we mustaccept thatgivena righttriangle,the square of the

hypotenuse is equal to thesumofthesquaresof thetwoothersides
(the Pythagoreantheorem).Unless some errorof reasoninghas
escapedus duringthepasttwothousandyears,Euclideangeometry
is necessarilytrue.
Whetherthe externalworldconformsor not is anotheraffair
whichconcernsthephysicist, not themathematician. If we tracea
righttriangle whose sides measure 3 and 4 on a piece paper with
of
rulerand square,and thenfindthatthehypotenusemeasures5, we
havenotverifiedthePythagorean theorem(32+ 42= 52)butGalileo's
theorybywhichthedistancesbetweenpointsin space can be calcu-
lated accordingto theformulasof Euclideangeometry.If one day
we wereto findthatthiswas notthecase, itwouldnotbe Euclidean
geometry whichwe wouldhaveto abandon,butGalileo'stheory.We
wouldhaveto call on othergeometriesthen(theyexistand arejust
as "true"as Euclideangeometry and forthesame reasons)to calcu-
late thedistancebetweenpointsin space.Moreover,thatis whatthe
theoryofrelativity is all about:experiment provesthatwhenobjects
are travelingclose to the speed of light,Euclideangeometryis no
longerapplicable,whatis expressedawkwardly as "speed contracts
distance."Naturally, speed does not compressanything, itis simply
thatanothergeometry, otherthanEuclidean,is applicableto that
situation.But thatdoes not mean thatthePythagorean theoremis
any less even
true, if physicalreality refuses to conform to it-rather
as the rules of chess are trueindependentof the factthatone is
playingcheckers.
Chaos theory, likeEuclideangeometry or thetheoryofnumbers,
is a setofmathematical statements,whichhas itsownexistenceinde-
pendentof whetherit is applicableto anyobservedphenomena.It
simplydrawsthelogicaland necessaryconsequencesfromcertain
ideas formodellingtimeintroducedin the seventeenth century.It
if
teaches us that we accept the mathematical models commonly
used overthepastfourcenturieswhichpermitted Newton,Poincaré,
and Einsteinto do theircalculations, thenwe also haveto acceptthe
possibilityof chaoticphenomena:instability withrespectto initial
conditions,existenceof strangeattractors. It teachesus as wellthat
ifthesechaoticphenomenaoccur,theywillalso have consequences,
which althoughlogical and necessary,may nonethelessbe unex-
pected.

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
CHAOS THEORY 141

Chaos theoryis thenabove all else anotherstepin thecontinual

progressofmathematics. The mathematician in me can but rejoice.
How manytimeshaveI been askedwithan air ofstupefaction: "You
do researchin mathematics? How is it stillpossible to findsome-
thingout aboutmathematics?" as thoughthecase of equal triangles
was the last word in science,and thatwe have been drivellingon
about the same thingsforthepast twothousandyearsor twohun-
dred yearsdependingon the charityof our interlocutor. That im-
pression is sustained by books written for the generalpublicwhich
almostnevergetup to thetwentieth and
century dwellindefinitely
on thesamepropertiesofprimenumbers.In reality, mathematics is
an extraordinarily livelyscience witha of
quick rhythm progress.
Practicallyall of theresultsconsideredimportanttodayhave been
discoveredduringthetwentieth century; as forcurrentfindings, the
bread and butterof theresearcher,theybecome outmodedwithin
fiveyears.A researchpaperin mathematics willcitebetween20 and
50 articleson whichit is based; 10% of thesewillbe morethanfive
yearsold, and rarelywillevenone be morethat20 yearsold.
Fromthispointofview,forthemathematician chaos theoryis a
like
theory anyother, born in the twentieth centuryand in constant
progress,markedeveryyearbythepublicationof dozens of books
and hundredsof technicalarticles.The onlythingparticularabout
itis theinterestthatitarousesbeyondthefrontiers ofmathematics,
among scientistsin otherdisciplines,and even in the publicat large.
Onlyinfrequently do the results of mathematics go beyonda circle
of initiates;the rare timesthatthishappens it has a tendencyto
arousesuspicionratherthanenthusiasm.In effect, itis to be feared
thattheinterestelicitedbychaos theorymightin partbe due to its
nameand thatsomepeople mightbe lookingfora theoryofgeneral
disorder,not to sayof themuddleof therealworld.This of course
willbe gravely disappointing and no help to theprogressof science.
Butwe are goingto see thatthereare betterreasonsto be interested
in chaos theoryand thatpublic confidence,foronce, is not mis-
placed.The majorimpactofthistheoryis yetto come and itwillnot
be limitedto mathematics butwillextendto all of science.

MODELS AND MODELLING

In orderto understandwhy,we have to keep in mindthestruc-
tureof scientific
theories,or at leastin thedomainof physics.It is

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
142 IvarEkeland

composed of-or up untilnow has been composed of,because we

are goingto see thatchaos theorymaychangethatstructure- two
parts,parallel to one another. On one side thephysicalsystem and
on theotherthemathematical modeland betweenthetwoa mysteri-
ous correspondence: thestateofthesystemis describedbythevalue
ofcertainvariablesofthemodel,and theinternallogicofthemodel
constrainsthesystem.The prototypeof all suchphysicaltheoriesis
Newtoniantheory.On one side we have thesolarsystem, thesun at
thecenter,and theothercelestialbodies,planets,asteroids,comets,
orbitingaroundit.On theotherside,we have some pointsin three
dimensionalEuclideanspace and some equationswhichdetermine
theirmovement.Betweenthesetwo,we havethepermanentmiracle
thatthemathematical movementoftheseabstractpointsin abstract
space coincideswiththeobservedmovementof thecelestialbodies
in concretespace and thatone can thenpredictthepositionof the
planetsbycalculating thesolutionsto Newton'sequations.We have
become so used to thismiraclethatwe no longereven thinkabout
it,but thatdoes not makeit anyless incomprehensible, at leastfor
thisauthor.
Be thatas itmay,thisapproachto modelling3 is at thebase of all
of modernscience.We do not knowifwe shouldadmiremorethe
mathematicalsophisticationor the power of predictionof the
modelspresentedto us bymodernphysics.Butbiologyand econom-
ics have also developedmodels,whichalthoughless sophisticated
and less powerful,representnonethelessimportantstagesin the
development ofscience.Of course,all of thesemodelsare as varied
as the situationstheycoverand the stateof mathematics permits.
One can,however,place themin twograndcategories,stochasticor
deterministic, accordingto whethertheyappeal to the elementof
chance.
A modelis stochastic4ifat somepointithas someonethrowdice
and makesuse of theresultthusobtained.In modernphysics,one
calculatesprobabilities:theprobability foran electronto pass from
one atomicorbitto another,theprobability fora nucleusto disinte-
grate,the probability for a vacuum to give birthto a couple of
As forknowingwhetherthephenomenonwillactually
antiparticles.

3 Intellectualconstructionsof a mathematical
model, that is to say, of a set of
equations whichare supposed to describephysicalreality.
4 A model is stochasticifit calls forthe calculationof
probabilities.

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
CHAOS THEORY 143

occur,ifsome electronwillchangeitsenergylevel,ifsome nucleus

willdisintegrate or ifwe willsee a positronappear in suchand such
a place, we do not know;thatdepends on a drawingin whichthe
physicistdoes not participate.This is theprototypeof a stochastic
modeland itshockedEinsteinwhoaskedhimself, who threw
rightly,
thedice. It is truethatclassicalphysicsgotus used to deterministic
models,whereno one throwsdice,and thaton thispointrelativity
conformsto the Newtoniancase. In a deterministic model,5the
evolutionof the model is entirelydeterminedbyitsactual state;if
one can resolvethe equations,one can predictfuturestatesand
reconstitute previousstates.
The firsteffectof chaos theory(let me say immediately thatI
believe it the least important)is to enlargethe paletteof models
availablefortherepresentation ofirregularor randomphenomena.
Up until now, when the physicist, biologistor the economist
the
encountereda phenomenonof thistype,he looked fora stochastic
model withthe idea thata deterministic model led necessarilyto
and
regular predictablebehavior, in contradictionpreciselywiththe
phenomenon he was tryingto model. If chaos theoryseems so
interesting to researchersin thesedisciplines,it is because it opens
up anotherpossibility, thatofproposinga deterministic, butchaotic,
model.One can thenhope to givean accountof thephysicalsystem
withoutappealing,as withstochastic models,to a dice throwing deus
exmachina.
One ofthephenomenaat thebase ofeconomicsis theexistence
of cyclesof activity, in whichperiods of recessionand expansion
alternate.Theyare moreor lesspronounced(we stillrememberthe
Great Depressionof 1929) and more or less long-lasting (afterthe
Second WorldWar Europe experienced30 yearsof uninterrupted
expansion),but theyare alwaysthere.Experienceshowsthatthey
are hardlypredictable;we cannot even recognizewhethersome
declinein activityis temporary or ifitsignalstheonsetofa recession
and the beginningof a new cycle.They are not totallyrandom
either;on the contrary, historicalanalysesrevealperfectly logical
linksbetweenthe expectationsof actorsand theirbehavior,and
revealtheimportancethatcertainpoliciescan have,in one sense as
in another.Incontestably, one oftheprincipalproblemsof econom-

5 A model is deterministic
ifit does not call forthecalculationof probabilities.

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
144 Ivar Ekeland

ics is the explicationof these cycles.Should one search for the

reasons in the internalworkingsof the economy?Should theybe
attributed to factorsexternalto it,be thattheincompetence ofthose
who governor industrialprogress?
Up untilnow, the second typeof explanationhas been privi-
leged. Because of theirregularcharacterof thecycles,attemptsto
representthemhave used stochasticmodels;thatis to sayattempts
to explain cycleshave been by recourseto exogenousshocks,of a
randomnature,to whichtheeconomywas subject.A certainnum-
ber ofmodels,forexample,appeal to technological innovation, each
newinventiontranslating intoa qualitativejump, a brutalproductiv-
ity gain propagated through the whole economy (technological
shock).Butchaos theoryopensup newexplanatory possibilitiesand
in thepastseveralyearswe havewitnesseda flowering ofdeterminis-
tic models linkingeconomic cyclesto the expectationsof actors
(consumersand producers),thealternationof generations, or even
to thesearchof a social optimumon thepartof a benevolentplan-
ner.Whatthesemodelshavein commonis seeingin theeconomya
chaotic system.The cyclesthen are only a consequence of this
hypothesis; thereare manyothersand I thinkwe are farfromhaving
finishedexploringthem.
In general,chaos theoryconsiderably broadensthepossibilities
foremploying deterministic models.Up untilnow,theiruse seemed
limitedto completelypredictablesystems, perfectly transparent to
the scholar,who could penetrateequallythe remotestpast or the
most distantfuture.This is the visionof Laplace who, fromthe
simplefactthattheuniverseobeyedNewton'slaw,concludedthat
everything was determinedfromnow on, and thatan intelligence
that knew the exact positionsand speeds of each particlein the
universecould calculateeverything in thepastand thefuture.In the
great book of the universe, everything is writtentoday,ifone is able
to read it.
Chaos theorydeliversus fromthatsuffocating mantle,fromthat
closed universewhere nothingcan happen, where there exists
neithertheunknownnorthenovel.The visionitproposes,inspired
bytheLorenzattractor, is entirelydifferent.Certainly, theuniverse
is governedby deterministic models, whether it be Newton's or
Einstein's.But thatdoes notimplythatthefuturecan be calculated,
anymorethanthepast:we haveenoughexamplesofchaoticsystems

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
CHAOS THEORY 145

to understandthatthecharacteristic timescale6poses strictlimitson

predictability.But thatdoes not implythatbeyondthecharacteristic
time scale thereis nothingmore to be said. Thanks to Lorenz's
system,we learned of the existenceof strangeattractorswhich
systemsapproachnaturallyand thendo not leave. Fromwherever
the systemstartswe knowwhereitwillgo: onto thestrangeattrac-
tor.And finally hereis a predictionthatwe can makewithcertitude,
well beyondthecharacteristic timescale: thesystemwillbe on the
strange attractor.
Proposinga deterministic modelfortheuniverseaffirms thatit
obeys strictlaws whichconstrainits evolutionover its entireexis-
tence;itmustnotleave itsstrangeattractor. Fromthepointofview
ofphysics, thatmeansthatall theoretically possiblestatescan notbe
practically realizedand thatnaturalstates,thosewhichmayappear
as the universeevolves,musthave veryparticularcharacteristics.
That is how thelaws of physicstakeform:theyare thesetsof rela-
tionswhichcharacterizethe naturalstatesamong all the possible
states.Thus,gas moleculesoccupyinga givenvolumecan in princi-
ple be distributed in verydifferent ways;one can imagine,forexam-
ple, thattheymight all be piled up in a corner,locallyachievinga
veryhighpressureand leavingmostof thevolumeempty,wherean
absolutevacuumis momentarily realized.But sucha state,although
theoreticallypossible, is not naturalin thesense thatthesystemwill
leave it spontaneouslyand rapidlyforits naturalstate,thermody-
namic equilibrium;the pressureand the temperaturewillbe uni-
formwithintherecipientand willbe relatedto thevolumeaccord-
ingto Mariotte'slaw.
But-and here is thecontribution of chaos theory-to propose a
deterministic modelalso leavesa space forchance,a dimensionfor
theunpredictable. Admittedly, thesystemis confinedon itsstrange
attractor, but itsmovementon theattractorescapes us. More pre-
cisely, characteristic
the timescale T poses a limitto thepossibilities
forprediction;rememberthatthisis thetimenecessaryfora posi-
tioningerroror a perturbation ofthemovementto be multipliedby
ten. For a timeperiodinferiorto T, thereis no problemfollowing

6 A chaotic
systemamplifiesanydifferencein initialconditions.The characteristic
timescale of a systemis thetimeit takesforthatdifferenceto be multipliedby ten.Two
trajectoriesinitiallydistantd fromone anotherwillbe distantlOd fromone anotherat
the end of the characteristic time.

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
146 IvarEkeland

the systemwithcalculations.For timeperiodssurpassing10T, its

trace is completelylost; all thatone can say (and it is alreadyan
importantdetail) is thatit is someplace on the strangeattractor.
Whereexactly,we do notknow.
Admirableand subtlemixof chanceand necessity!Here is the
resolutionin one fellswoopofan armyoffalseproblemsconcerning
humanfreedomin a deterministic universe.We no longersee, like
Laplace, a cloudless skyopen to an infinitehorizon,so clear,how-
ever, thatit givesus the impressionthatwe could reach out and
touchit.Neitheris itan overcastsky,shroudedin a fogwhichblocks
our gaze and concealsfromus anyhorizonat all. Whatwe see are
the twotogether,as whengustsofwindin a stormyskycontriveto
giveus a glimpseof distantsun-drenched horizons.

BETWEEN MODEL AND REALITY: CALCULATION

But even thisvision,despiteitsprofundity and beauty,is not to

our eyesthemostimportant contribution of chaos theory.The thin
marginwhichseparatesmathematicalzero fromalmostnothing,
absoluteexactitudefromthebestapproximation, slipsbetweenthe
mathematical modeland thephysicalsystemwhichit is supposedto
represent. fourcenturiesitwentunnoticedbecause themeans
For
of calculation available strictlylimitedthe use of deterministic
models and theircomparisonwithphysicalreality.It is thanksto
chaos theorythat thismarginhas finallybeen detectedand the
futuredevelopmentof sciencewillaccord it increasingattention.
Betweenthemathematical modeland physicalrealityan intermedi-
ate space has been discovered,thatof calculation.
Let us explain.Untilthe inventionof computers,onlycalcula-
tionswithpaper,penciland erasurecouldbe performed. Exceptfor
very rare exceptions,to which we willreturn,that means thatonly
linear equations could be solved. In nature there are a certain
numberof systemswhichare governedby thistypeof equation.
These systems,called linear,alwaysexhibitverysimplebehavior;
theyare neverchaotic,theirtrajectories are alwayspredictable,and
theirmovementsare alwaysregular.But up untilabout 1950, they
weretheonlyones forwhichone couldcalculatethetrajectories and
study the movement. It was then quite natural forthe attentionof
scientiststo be focusedon them,and overfourcenturies,a multi-

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
CHAOS THEORY 147

tude of linearmodels fordiversephenomenafromphysicsto the

economyweredeveloped.Duringthisperiodthelinearmodelwas
king.It was even used to studychaoticsystems, like meteorology,
because theexactmodel includednonlinearequationswhichcould
not be solved. Unable to use the exact model, more and more
complicatedlinear models were constructedto betterand better
approximatethesystemunderconsideration, but ofwhichwe now
knowtheywillnevergivean exactidea.
up untilthetwentieth
Briefly, century, thenotionof a determin-
isticmodel was bound up withthat of a linear model.That is what
explains,forexample,Laplace'serror.He, and hispredecessors, had
constructeda highlyrefinedlinear model to representthe solar
system.He believedthatthepropertiesofhis modelweretheprop-
ertiesof thesystemand thatit was thereforestable.But his model
was onlyan approximation, preciselybecause it was linearand the
systemis nonlinear.We nowknowthattheapproximationceases to
be validbeyondthecharacteristictimescale,approximately one hun-
dred millionyears.That,Laplace could not know,because he did
nothavethemeansofcalculating directly fromthenonlinearmodel
and thereby the
ascertaining quality of his approximation (excellent,
we have to say,at thehumanleveland withthemeansavailable).
Onlyin thetwentiethcenturyhas itbeen understoodthatnonlin-
ear models have propertiesfundamentally differentfromlinear
models. The centuryopened withthatgreatworkof Poincaré,Les
Méthodesnouvellesde la méchaniquecéleste,where he shows that the
linearmodelsemployedbyhispredecessors, as refinedas theymight
be, could nevergivean exact idea of thelong-term behaviorof the
planetaryorbitsand thatthe exact,nonlinearmodel can conceal
trajectoriesof a complexityunsuspecteduntilthen.The century
closes as chaos theorybrilliantlyconfirmsPoincaré'sintuitionand
fromnow on our knowledgeof nonlinearmodelsrestson a multi-
tudeofexperimental results,numericalsimulations, and mathemati-
cal theorems.In themeantime, the data processingrevolution finally
allowedus to calculatethesolutionsto nonlinearequationsand to
representthem graphically, withoutwhichwe would still know
nothingof the Lorenz attractorand the chaoticcharacterof the
solarsystem.
Thanks to computers,nonlinearmodels can finallybe used.
Numericalsimulationrevealstheirpeculiarcharacteristics, verydif-
ferentfromthoseoflinearsystems, and mathematical analysisgives

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
148 IvarEkeland

us theirconfirmation. Chaos theorywas bornof thatconfrontation

betweenthe computerand the mathematician. The computerre-
vealsto themathematician thephenomenato studyand themathe-
maticianexposes the limitsof the computer.In a chaoticsystem,
certaincalculationslackphysicalsignificance: one can ask thecom-
puters of the weather bureau what theweather willbe likein Parisin
twoyearsand, ifleftto worklong enough,theywillgivean answer,
but thereis no hope thatthiscalculatedresponsemightbe any
betterthanwhatI could guessbased on seasonalaverages.
This is thebeginningof a revolutionin theconceptionof scien-
tifictheories.Fromnow on, the correspondencebetweenphysical
realityand mathematical modelis no longerimmediate;itpasses by
way of Never
calculation. againwillone say:suchand suchan equa-
tionrepresents suchand sucha phenomenon.Itwillbe necessaryto
add: the systemis chaotic,its characteristic time scale is of this
duration, be aware thatbeyond that time certain calculationsno
longer representanything, and if you wantto calculatesuch and
sucha quantity, use suchand sucha methodratherthansome other
method.In otherwords,itwillno longerbe possibleto pronounce
a scientifictheorywithoutstatingwhatcan be, and cannotbe, calcu-
latedin thattheoryand withoutindicating in each case theappropri-
ate meansofcalculation.Itwasalreadyknownthatscientific theories
have limitsto theirvalidity. The domainof classicalmechanics,for
example,is limitedon one hand byquantummechanicsand on the
otherbythetheoryofrelativity. Nowwe have to getused to thefact
thatthereare also numericallimits.
I believethatthisrevolutionwillalso be extendedto theteaching
of mathematics whereproblemslinkedto calculationwilltakeon a
considerableimportance.To takeonlyone example,studentsare
taughtto solvethesecond degreealgebraicequation:
ay? + bx + c = 0

bylearningthecelebratedformula:
x = - b ± Vb2- 4ac
2ap

Is it not finallytime,afterso manycenturies,to realizethatthis

solutionis not a solution?Because, in short,how can we calculate

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
CHAOS THEORY 149

thatsquare root?If,forexample,resolving:

J + 4x-3 = 0

we get:

-2±/T

How muchfurther
alongare we; nowwe have to resolve:

*?-7 = 0.

Justbecausewe callthesolutionv7 does notmeanthatwe know

howto calculateit.As forlookingup thevalue in a table,or getting
out our calculator,thatsimplyevades thedifficulty and relieson a
previous How
calculation. and by whom?
All thathas been accomplishedis to reduceone second degree
equation to another.That is not "solving"theequation,and ifit is
leftthere,studentsgeta falseidea ofwhatmathematics is and what
can be expectedofit.To solvetheequationis to be able to calculate
thesolutions,thatis to be able to givetheirvalueswithall thepreci-
sion one wants.And thatis wherethe problemsreallyget inter-
esting.The methodsused forcalculatingthevaluesof numberslike
v7 are in factproceduresforimproving an approximation. Starting
withan initialvalue,3 forexample,theprocedureis thenapplied
one time,a second time,a thirdtime,and so on. If nine decimal
placesare desired,theprocedurecan be interrupted whenitreturns
twonumberswhichcoincideout to theninthplace.Fromtherearise
a multitudeof questions:will the methodalwaysend up witha
result?Can both solutionsbe calculatedthatway?Whatinfluence
does thechoiceoftheinitialvaluehave?This is a domainveryclose
to the practicalwherethe computercan supplyappreciableassis-
tance in visualizingthe problemsand stimulating intuitions.But
even so thequestionsbeingasked are no less complexand inspire
veryinteresting developmentsin mathematics. It sufficesto widen
theframework everso slightly,to workin thecomplexplane or with
higherdegreeequations,forthe calculationprocedureto become
chaoticand generateincoherentseriesof values insteadof nicely
convergingto a solution.

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions
150 IvarEkeland

In a well-knownarticle,thephysicist David Ruelleaskedhimself

ifour mathematics was natural:wouldlittlegreenmen,livingon a
planet orbiting around threered suns,and thushavingan experi-
ence of the physicaluniversecompletelydifferent fromour own,
have developed the same mathematics?Perhaps we shall never
know.But whatis happeningrightnow is analogous:thepowerof
calculationavailableto menfromnowon is changingtheiruniverse.
It is transforming theirenvironment, it is transforming theirsocie-
ties,it is transformingthem,it is transforming theirscience.Chaos
theoryis a beginning,notan end.

This content downloaded from 170.140.26.180 on Mon, 11 Jan 2016 04:37:36 UTC
All use subject to JSTOR Terms and Conditions