The Fractal Geometry of Nature. by Benoit B.
Mandelbrot
Review by: J. W. Cannon
The American Mathematical Monthly, Vol. 91, No. 9 (Nov., 1984), pp. 594-598
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2323761 .
Accessed: 01/01/2014 10:07
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 support@jstor.org.
Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.
http://www.jstor.org
This content downloaded from 178.91.253.11 on Wed, 1 Jan 2014 10:07:51 AM
All use subject to JSTOR Terms and Conditions
594 J. W. CANNON [November
(Sz. -Nagyand Foias). A generalquestionwhichariseshereis theexistence ofinvariantsubspaces,
and althoughthesolutionin generalseemsbeyondreach,severalimportant advanceshavebeen
made.In the1930'svonNeumannprovedtheexistence ofproperinvariant subspacesforcompact
operators(althoughhisproofwas neverpublished).Thisresultwas subsequently re-examined and
generalizedby a numberof workers, mostnotablyby Lomonosovin 1973 who extendedit to
operatorswhichmerelycommutewitha nonzerocompact.Anothernotableresultwas Scott
Brown'sproofin 1978of theexistence ofinvariant subspacesforsubnormal operators,a class of
operatorsfirststudied by Halmos (see Ch. 21 of theHilbert Space ProblemBook).
A recurringthemeis theperturbation ofan operator by a compactoperator, beginningwiththe
Fredholmalternative and a 1909 resultof Weyl on the invarianceof the limitpointsof the
spectrum undersuchperturbations. Thislineofdevelopment leadsto somesurprisingconnections
betweenHilbertspaceoperators and topology, beginningwiththefact(Atiyahand Janich,1964)
that the Fredholmoperatorsforma classifying space for complex K-theory.The studyof
operatorsthatarenormalmodulothecompacts(Brown,Douglas,and Fillmore, 1973)has led to a
good deal of interestin "noncommutative topology."An exampleof recentprogressis the
astonishinglysimpleproof,due to J. Cuntz,of theBottperiodicity theorem-madepossibleby
extending thecontextto C*-algebras.
Because of thediversityof topicsthatit includes,singleoperatortheoryis ideallysuitedto a
book of thenatureof Halmos's(we can imaginea HomotopyTheoryProblemBook beingless
successful).The secondeditioncontains250 problemsas against199 in thefirst.It remainsan
compactsurveyof singleoperatortheory,
excellent, as wellas a valuablesourceforHilbertspace
techniques.The form,content, and styleof thesecondeditiondeservethesamehighpraiseas the
first.
Freeman,San Francisco,1982.460 pp.
ofNature.By BenoitB. Mandelbrot.
The FractalGeometry
J.W. CANNON
University
Department,
Mathematics Madison,WI 53706
of Wisconsin,
booklessbythetextthanbythequotations....W. H. Auden[1,p. 471.
He judgesa... critical
criticwhohas everhad to reviewa newvolumeof poetry
Again,anyconscientious in a limited'
space
comment....W. H.
without
knowsthattheonlyfairthingto do wouldbe to givea seriesof quotations
Auden[1. pp. 11-12].
setor concrete
is a mathematical
A fractal at all scales... . B. B.
or fragmented
objectthatis irregular
[3].
Mandelbrot
... I claimthatmanypatterns andfragmented,
ofNatureareso irregular that,compared withEuclid-a
termusedin thisworktodenoteallofstandard geometry-Nature exhibits
notsimply a higher
degreebutan
altogether different ofthesepatterns
. Theexistence
levelofcomplexity... challengesus tostudythoseforms
thatEuclidleavesasideas being"formless," themorphology
to investigate of the"amorphous." B. B.
Mandelbrot [4].
All pulchritudeis relative....We oughtnottobelievethatthebanksoftheoceanarereallydeformed,
becausetheyhavenottheform ofa regular
bulwark;northatthemountains areoutofshape,becausethey
arenotexactpyramids orcones;northatthestarsareunskilfully at
placed,becausetheyarenotall situated
uniform butwithrespectto our fanciesonly;norare they
distance.Theseare notnaturalirregularities,
incommodious to thetrueusesof lifeand thedesignsof man'sbeingon earth.... QuotedfromRichard
Bentley byB. B. Mandelbrot [4].
BenoitB. Mandelbrot,in hisbook entitledTheFractalGeometry sets
ofNature,systematically
can be
thereader'seyeand mindthatmanyof Nature'sapparentirregularities
about convincing
modelledbymathematical
and beautifully
efficiently and somewitha
objects,somedeterministic
This content downloaded from 178.91.253.11 on Wed, 1 Jan 2014 10:07:51 AM
All use subject to JSTOR Terms and Conditions
1984] REVIEWS 595
randomcomponent,but all highlyirregular whichMandelbrotchoosesto call
or fragmented,
fractals.
Each newlycreatedcomputer-generated makestheargument
illustration morecompelling.On
therearjacket of thebook,a Brownianearththatneverwas risesovera moonscapethatnever
was,bothcreatedby thecomputer artistry
ofone of Mandelbrot's coworkersat IBM, RichardF.
Voss. The programsare based on simplegeometricinvarianceprinciples.More recentprograms
totallyfake,
createmountainranges,valleys,rivers,clouds,and mists,all visuallycompelling,
againbased on a fewsimpleinvarianceprinciples.
Computer-generatedfractallandscape surfaceof Hausdorffdimensionapproximately2.1 realized by Richard F.
Voss. From The Fractal Geometryof Nature,W. H. Freemanand Company?,Benoit B. Mandelbrot1977, 1982,
1983.
Mandelbrotsuggeststentative firstfractalmodelsforcoastlines,galaxyclusters,turbulence,
island clusters,trees,thelungs,Brownianmotion,drainagesystems, mountainranges,irregular
textures,errorburstsin data transmission,
stockmarket and manymore.The bookis a
variations,
rich source of beautifulpictures,interestingnew mathematical models,and imaginativenew
terminology.
Mandelbrotpresentsa widerangeof experimental evidenceby otherscientiststo supporthis
viewsthatstandardmodelsfallshortofexplaining thesamephenomenawiththesameefficiency,
clarity,and adherenceto fact.Mandelbrot describesthebook as a manifestoand a casebook.It is
fullof strongly fullof historical
held opinionsand claimsof priority, anecdotes,apt illustration,
and thebestin computer art.
The classical mathematical continuousfunctionsof
fractalsare the nowhere-differentiable
Weierstrass, curvesof Peano,and thefamousCantorsets.
thespace-filling
The classical view was that theseexampleswere forcedupon us by the abstractionsof
mathematics and thattheyborelittlerelationship
to reality.
Here are someof theopposingviewsexpressedby Mandelbrot.
As regardsnowhere-differentiable
functions. MandelbrotquotesJeanPerrin,whoseworkon
Brownianmotionwon PerrintheNobel Prize.,
... A good teacher... draws a well-definedcurve and says, holdinga ruler,"You see that thereis a
tangentat everypoint."... Mathematicians, however,are well aware thatit is childishto tryto show by
drawingcurves that everycontinuousfunctionhas a derivative... . Curves that have no tangentsare the
rule.... Those who hear of curveswithouttangents,or of functionswithoutderivatives, oftenthinkat first
thatNaturepresentsno suchcomplications... . The contrary howeveris true... . Consider,forinstance,one
This content downloaded from 178.91.253.11 on Wed, 1 Jan 2014 10:07:51 AM
All use subject to JSTOR Terms and Conditions
596 J.W. CANNON [November
of the whiteflakesthatare obtainedby saltinga solutionof soap. At a distanceits contourmay appear
sharplydefined,but as we drawnearerits sharpnessdisappears.The eye can no longerdrawa tangentat any
point... . The use of a magnifying glass or microscopeleaves us just as uncertain,forfreshirregularities
appear everytimewe increasethe magnification.... An essentialcharacteristic of our flake... is thatwe
suspect... thatany scale involvesdetailsthatabsolutelyprohibitthefixingof a tangent... . [4, pp. 6-7]
Accordingto NorbertWiener,the paths of Brownianmotionmay best be modelledby
continuous,
nowhere-differentiable, curves.[4,p. 9]
locallyplane-filling
As regardsPeano curves,Mandelbrotwrites,
it must penetrate
... if a tree made of riversof vanishingwidth is to drain an area thoroughly,
everywhere. motion... . [4, p. 59]
One who followstherivers'combinedbank performsa plane-filling
Thus a curve of positivearea has intuitivevalue as a model and the analogoushigher
withthree-dimensional
surface
object,a two-dimensional
dimensional volumeisjust as important:
... Harveyled to a viewof thecirculationof theblood whichassertsthatboth an arteryand a vein are
foundwithina smalldistanceof everypointof thebody-except of coursethatpointswithinan artery(or a
vein) are preventedfrombeingveryclose to a vein(or an artery).
Stateddifferently .. .: everypointin nonvasculartissueshouldlie on theboundarybetweenthetwoblood
networks.
A second designfactoris thatblood is expensive.Hence thevolumeof all thearteriesand veinsmustbe a
small percentageof the body volume,leaving the bulk to tissue... . (Therefore)tissue... is a fractal
surface:.. . (it) mustbe topologically two-dimensional, because it formsthecommonboundaryof two shapes
that are topologicallythree-dimensional, but it is requiredto have a volumethatnot onlyis nonnegligible
comparedto the shapesit bounds,but is muchlarger.
(Thus) fractalmonsters are theverysubstanceofourflesh... . [4, pp. 149-150]
Mandelbrotnotesthatthebestmodelfortheclustering massof thestars
of the3-dimensional
and galaxiesin theuniverseseenfromafarseemsto be not 3-dimensional butrathertheCantor
set (or "dust" as Mandelbrotprefersto call it) of topologicaldimension0; Mandelbrotthen
recallsan argument of E. E. Fournierd'Albewhichsuggests nontopologi-
thatthecorresponding
cal, geometricor fractaldimensionof theclustering 1. We quote
mass shouldbe approximately
onlytheconclusionof theargument.
... To paraphraseFournier,an importantconclusionmaybe drawnfromtheobservationthatno stellar
velocityexceeds 1/300 of thevelocityof light.It is thatthemass comprisedwithina worldball increasesas
its radius,and not as its volume...: [4, p. 89]
Experimental of about1.23.
evidenceseemsto suggesta moreprecisefractaldimension
In conclusion,Mandelbrotarguesthatall of theclassicalfractalmonstersare good intuitive
modelsof real and importantobjects.
Though Mandelbrotspends the firsthalf of his book (200 pages) on the classical and
deterministic it seemsclearto me thathisloveis attachedto therandomfractalmodels.
fractals,
phenomena
The principal modelledare landscapes,galaxies,and coastlines.
... The fullmeaningand practicalrelevanceof these(basic fractal)themesare not apparentuntilone
tacklesrandomfractals... . Let us go back to thequestion"How longis thecoast of Britain?"... One might
improvethe (deterministic fractal)model by invokingmorecomplicateddeterministic algorithms. However,
thisapproachwouldbe notonlytedious,but doomedto failure,because each coastlineis moldedthroughout
the ages by multipleinfluencesthatare not recordedand cannotbe reconstituted in any detail... . The key
lies in statistics... . [4, pp. 200-201]
out of thisdifficulty
... Can chance bringabout the strongdegreeof irregularity encountered,say, in coastlines?Not only
does it, but in manycases it goes beyondthedesiredgoal... . [4, p. 201]
... The precedingsectionarguesthatthetheoryof changeis not reallydifficult. it is not
Unfortunately,
really easy... . Indeed, rules that generateacceptable randomcurvesare veryhard to describe,because
geometricsets are imbeddedin a space. By merelyvaryingat randomtheshapes,thesizes,and theorderof a
coastline'sparts,one tendsto be leftwithpieces thatwill not fittogether... . [4, p. 201]
This content downloaded from 178.91.253.11 on Wed, 1 Jan 2014 10:07:51 AM
All use subject to JSTOR Terms and Conditions
1984] REVIEWS 597
enough.... [4, p. 205]
... The nonrandomfractals'essentialfailingis thattheyare not symmetric
Mandelbrot'sgoalsin introducing randomfractalsaretwofold.First,he wantsmodelsthatare
Secondly,
yetvisuallycorrect.
efficient he wantsmodelsthatsatisfy statistical
therathersurprising
invariancepropertiesthathave been suggestedby experiment. Amazingly,he discoversthat
thesecondproperty
satisfying thefirst.
also seemsto satisfy
The mostimportant thingto noteabouttheinvariance Mandelbrotdescribesis that
properties
manyare demonstrably incompatible withstandardGaussiandistributions.
... The studyof scalinglaws led me to recognizethatnonstandardcentrallimitbehavioris in factpartof
nature.Unfortunately, as soon as the centrallimittheoremargumentis recognizedto have more than one
possible outcome,it ceases to be persuasive... . [4, p. 423]
Mandelbrotrecognized apparently
thata largenumberofreal-world, randomprocessesseemed
or scaling.
generalnotionof randomself-similarity
to satisfythefollowing
... Any naturalfluctuationcan be processedto be heard-as impliedby the termnoise. Tape it, and
listento it througha speakerthatreproducesfaithfullybetween,say,40 Hz to 14,000Hz. Thenplay thesame
tape fasteror slower than normal. In general,one expects the characterof what is heard to change
considerably.A violin,forexample,no longersounds like a violin.... There is a special class of sounds,
however,that behave quite differently. Afterthe tape speed is changed,it sufficesto adjust the volumeto
make the speakeroutput"sound thesame" as before.I proposethatsuch soundsor noisesbe called scaling.
WhiteGaussian noise... is scaling.But otherscalingnoisescan be made available formodelmaking....
[4, pp. 249-250]
And themembers ofone important familyof suchscalingnoisesare distinguishedby a certain
fractaldimension[4, p. 250]. It is the membersof thisfamilythatMandelbrotoftenuses to
wonderful advantage.
Fractalshaveexistedin factifnotin namein mathematics eversincemathematicians beganto
thinkseriouslyabout thefoundations of theirsubject.Fractalshave dominatedcertainareas of
mathematics fordecades.But Mandelbrot'sthoughtand the interestevokedby his book are
symptomatic a realchangein themathematical
of milieu.Whatis thatchange?
I believethatthechangehas beenwrought by theevolutionoftheelectronic computer and the
development of marvelouscomputer graphics.As Mandelbrotarguedin a quotationcitedin the
previoussection,any streamof data can be processedas sound or noise. But also it can be
processedas geometric shapesor pictures.One discoversthatalgorithms have intrinsicshapes,
thatthoseshapes are complexand beautiful, thattheyhave underlying geometries,underlying
or invariance
self-similarity properties. Mathematicians are in thenew positionof beingable to
exploreextensively an almostunchartered the
area, geometry of algorithms.
What's in a name?That whichwe call a rose By any othername would smellas sweet... . Shakespeare,
Romeoand Juliet,II, ii, 43.
Or wouldit? Let us chooseas a nameforeach individualthealgorithm encodedin its genetic
structureor in thegoverning lawsofitscreation.If we changethename,then
physicalor spiritual
we changethecode,changetheindividual, changetheshape,smell,and feelof theindividual.
If we thinkin abstractterms,thenit amazes us thatalgorithms have shapes,but whenwe
realizethatmuchoflifeaboutus is algorithmic,thenwe see and smelland hearin everyface,bud,
flower,and tree,theshape,smell,and soundof an algorithm, a marvelouscreation.
It seemsto me thatwhatMandelbrotis studying in his FractalGeometry of Natureis the
of
asymptoticgeometricshape algorithms. fragmentation,
Irregularity, fractal dimension,and
otherimportant which
properties he emphasizes are thensimply important measuresof
diagnostic
this asymptotic shape.Therefore I would redefinefractalgeometry forthe purposeof mathe-
maticsas follows.
FRACTAL GEOMETRY is thestudyof theasymptoticgeometricshape of algorithms.A FRACTAL is
theasymptoticgeometricshape of a particularalgorithm.
This content downloaded from 178.91.253.11 on Wed, 1 Jan 2014 10:07:51 AM
All use subject to JSTOR Terms and Conditions
598 LETTERS TO THE EDITOR [November
Mathematiciansare likeFrenchmen:whateveryou say to themtheytranslateintotheirown languageand
forthwith
it is somethingentirelydifferent...Goethe.
It is clearthatmanybeautiful in fractalgeometry.
thingsare cryingto be discovered To what
extentwe shall succeedis not clear.As we attemptto understand complex,seemingly random
processes,we oftenfeellikeMarkStuddockwhowas beingtrainedin Objectivity byhis diabolic
mentorin C. S. Lewis'ssciencefictionnovel,ThatHideousStrength:
... Then (Mark) noticedthespotson theceiling.... Theyweredeliberately paintedon: littleroundblack
spotsplaced at irregularintervalson thepale mustard-coloured surface... . He determinedthathe would not
fall into the trapof tryingto countthem.Theywould be hard to count,theywereso irregularly placed. Or
weren'tthey?Now thathis eyesweregrowingused to them(and one couldn'thelp noticingthattherewere
five in that littlegroup to the right),theirarrangementseemed to hoveron the vergeof regularity. They
suggestedsome kindof pattern.Theirpeculiaruglinessconsistedin theveryfactthattheykepton suggesting
it and thenfrustrating theexpectationthusaroused.Suddenlyhe realisedthatthiswas anothertrap.He fixed
his eyes on the table.
There were spots on the table too: white ones. Shinywhite spots, not quite round. And arranged,
apparently,to correspondto thespotson theceiling.Or werethey?No, of coursenot... ah, now he had it!
The pattern(if you could call it a pattern)on thetablewas an exact reversalof thaton theceiling.But with
certainexceptions.He foundhe was glancingrapidlyfromone to theother,tryingto puzzle it out... . [2, pp.
297-298]
References
1. W. H. Auden,The Dyer's Hand, VintageBooks,New York, 1968.
Macmillan,New York, 1969.
2. C. S. Lewis, That Hideous Strength,
3. B. B. Mandelbrot,Fractals:form,chance,and dimension,Freeman,San Francisco,1977.
4. _ _, The FractalGeometry of Nature,Freeman,San Francisco,1983.
LETIERS TO THE EDITOR
Materialfor thisdepartmentshouldbe preparedexactlythesame wayas submitted
manuscripts (see theinside
P. R. Halmos,Department
frontcover)and sentto Professor ofMathematics, Bloomington,
Indiana University,
IN 47405 (afterDecember15, 1984: Department ofMathematics, ofSanta Clara, Santa Clara, CA
University
95053).
What followswas NOT submitted as a letterto theeditor.It is an articlethatappearedin
somewhatshortenedformsin severalnewspapers, and severalreadershave suggestedthatit be
made morewidelyavailableto the membersof theAssociation.Thereis no standardway to
includenon-standardcomments suchas thisin theMONTHLY: thedecisionto haveit appearin
thisdepartmenthas theauthor'spermission.
Editor:
In 1968, when I was an undergraduate, manystudentswere advisedto major in Library
oflibrarians
fora newgeneration
Science:therewas a needin largelibraries withthe
conversant
modem conceptsof data storageand retrieval.Withina fewyears,theneed forsuchlibrarians
driedup; indeedmanyfindit hardto believethatthereeverwas sucha need.
Today the hot collegemajoris ComputerScience.At UCLA nearlyone thirdof entering
freshmenexpressan interestin majoringin ComputerScience.The situationis similaron
campusesacrossthenation.ComputerScienceis verysexystuff thesedays.Glossymagazinesare
This content downloaded from 178.91.253.11 on Wed, 1 Jan 2014 10:07:51 AM
All use subject to JSTOR Terms and Conditions
