A Disorienting Look at Euler's Theorem on the Axis of a Rotation

Author(s): Bob Palais, Richard Palais and Stephen Rodi

Source: The American Mathematical Monthly, Vol. 116, No. 10 (Dec., 2009), pp. 892-909
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/40391249 .
Accessed: 21/03/2013 23:54

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].

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 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
 A Disorienting
Look at Euler's Theorem
on theAxisofa Rotation
Bob Palais,RichardPalais,and StephenRodi

1. INTRODUCTION. A rotation in twodimensions(or otherevendimensions)does

notin generalleave anydirectionfixed,and evenin threedimensionsit is notimme-
diatelyobviousthatthecompositionof rotationsaboutdistinctaxes is equivalentto a
rotationabouta singleaxis. However,in 1775-1776, LeonhardEuler [8] publisheda
remarkable resultstatingthatin threedimensionseveryrotationof a sphereaboutits
centerhas an axis,and providinga geometricconstruction forfinding it.
In modernterms,we formulate Euler's resultin termsof rotationmatricesas fol-
lows.

Euler's Theorem on the Axis of a Three-DimensionalRotation. If R is a 3 x 3

orthogonalmatrix(RTR = RRT = I) and R is properfdetR = + 1), thenthereis a
Rv = v.
nonzerovectorv satisfying

This importantfacthas a myriadof applicationsin pureand appliedmathematics,

and as a resultthereare manyknownproofs.It is so well knownthatthe general
conceptof a rotationis oftenconfusedwithrotationaboutan axis.
In the nextsection,we offera slightlydifferent formulation, assumingonly or-
thogonality,butnotnecessarilyorientation preservation.We give an elementary and
constructiveproofthatappearsto be new thatthereis eithera fixedvectoror else a
"reversed"vector,i.e., one satisfyingRv = - v. In thespiritof therecenttercentenary
ofEuler'sbirth,followingourproofit seemsappropriate to surveyotherproofsofthis
famoustheorem.We beginwithEuler's own proofand providean Englishtranslation
fromtheoriginalLatin.Euler's construction relieson implicitassumptions of orienta-
tionpreservationand genericity, and leaves confirmationofhis characterization of the
fixedaxis to thereader.Our currenttastesprefersuch mattersto be spelledout,and
we do so in Section4. There,we again classifygeneraldistancepreserving transfor-
mations,thistimeusingEuler's sphericalgeometry in moderndressinsteadof linear
algebra.We notethatsome constructions presentin Euler's originalpapercorrespond
to thoseappearingin our proofwithmatrices.In thefinalsection,we surveyseveral
otherproofs.

2. EULER'S THEOREM FOR ORTHOGONAL MATRICES: A CONSTRUC-

TIVE LINEAR ALGEBRA PROOF. We will see below that,in Euler's original
oforientation
paper,preservation is assumedimplicitly.
In thissectionwe omitanyas-
sumptionthatourmap R is eitherproperor improper(detR = -1), and inferit from
theconclusion.

Euler's Theoremfor3 x 3 OrthogonalMatrices. If R is a 3 x 3 orthogonalmatrix,

Rv = v, or a nonzerovectorv satisfying
thenthereis a nonzerovectorv satisfying
Rv = -v.
doi:10.4169/000298909X477014

892 © THE MATHEMATICALASSOCIATION OF AMERICA [Monthly 1 16

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
Remark.Theproofwillfindall suchvectors andwilltherefore
constructively, iden-
tifythe of
dimension thesubspaceof reversedvectors
v Rv
satisfying = - v. R is
or
proper improper to
according whether dimension
this is evenorodd.

Proof.Webegintheproof thatifA := ' (R - RT)is theskew-symmetric

byobserving
R
partofR, then commutes A:
with

RA = AR. (1)

To seethis,weusethefactthatR commutes andwithitsinverse

withitself R"1 = RT,
so itcommuteswithR - RTandhencewithA. As a consequence, ifv is inthekernel
ofA, A(Rv) = R(Av) = 0 showsthatRv is alsointhekernelofA.
Ourproofproceedsbytreating twocases:a generic A
case, ^ 0, anda nongeneric
case, A = 0. ■

1. The GenericCase. NotethatA takestheform:

an -a3ì '
-an 0 a23 I .
(0
Ifweletv := (a23a3i an)T thenv ^ 0 andAv= 0 is anidentity,i.e.,A hasnontrivial
kernel.TheformofA guarantees itsrankis at least2. SinceA
that,ifitis nonzero,
hasnonzero kernel, itsrankmustbe exactly2 andthereforev spansitskernel. Since
=
||Rv|| ||v||, shows
this Rv = ±v.

Remark.In thegenericcase, therecan be no vectorsindependent of v satisfying

Rv = ±v. ForRv = ±v impliesthat(R - R-1)v = 0, so v is inthe(one-dimensional)
kernel ofA. SinceR preserves orthogonality, theorthogonal complement ofv,vx,is
an invarianttwo-dimensional subspace on which R actsorthogonally. We can findits
matrix withrespectto an orthonormal basisforyL by observing thattheimageof
thefirstbasisvectoris somearbitrary unitvectorin v1, (£), andtheimageof the
secondis oneofthetwounitvectors orthogonal totheimageofthefirst, ± ( ~*). If
wechoosetheminussign,theresulting ) whichis a reflection
R is ( %_?a acrossa fixed
vector.In thiscase,we see thatR is nongeneric, becauseitis symmetric withrespect
to an orthonormal basis,or alternatively, =
sinceR2 I so Rr = R"1 = R. Sincewe
haveassumedR is generic, R mustact as a properrotation of vx, and we can see
thatR preserves theorientation of a right-handed if
frame v is fixed, and reverses
itsorientationifv is reversed.We willsaymoreaboutorientation preservation after
treatingthenongeneric case.It is possibletoshowthatA hasnontrivial kernel in any
odddimension n,byappealing todetA = detAr = det(-A) = (-1)" detA.

The abovecase is genericbecausetheorthogonal withA = 0, i.e.,the

matrices
symmetricones,havemeasure zero,corresponding 0 andn. Still
toanglesofrotation
andwe havea similar
theyareimportant, forthem.
result

thenwe can writeanyvector

2. The NongenericCase. If R is an n x n matrix,
x € Rnintheformx = x+ + x_,wherex+ = ±(x + Rx) andx_= ¿(x - Rx). When
A = 0 andR = Rr = R~' we haveR2 = I, andin thiscase Rx+ = +x+, Rx_=
foranyx andy,x£y_= 0. In otherwords,we havean orthogonal
-x_. Moreover,
decomposition = V+ © V~ intothefixed-point
Rn ofR
setV+ (the+1 eigenspace)

December
2009] theaxis of a rotation 893

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
and the space V" of reversedvectors(the -1 eigenspace) of R, and the operators
P+ = ±(I + R) and P_ = ¿(I - R) are theprojections( P| = P±) ontothesespaces.
Fromthepointof view of operatornotation,thisis a consequenceof R2 = I, which
- I) = (R - I)(R + 1) = 0.
impliesR2 - I = (R + I)(R
For n = 3, we may now classifysymmetric orthogonal3x3 matricesintofour
cases accordingto thecomplementary dimensionsd+ andd- of thefixedand reversed
subspaces.As in thegenericcase, we notethatR is properor improperaccordingto
whetherd- is even or odd. If d+ = 0 and d- = 3 thenR is the antipodalmap (im-
proper).If d+ = 1 and d_ = 2 thenR is a half-turn rotationaboutsome axis (proper).
If d+ = 2 and d- = 1 thenR is a reflection if
in some plane (improper).And finally,
d+ = 3 and d_ = 0 thenR is theidentity map (proper).
thefollowingconverseto theclassification
It is possibleto formulate of nongeneric
3x3 orthogonalmatricesas well. If R is an orthogonal3x3 matrixthatfixesor
reversesatleasttwoindependent vectors,thenR is symmetric. The proofis elementary
and leftto thereader.
Includingtheproperand impropergenericcases, we have in totalsix cases, which
arecharacterized by thedimensionsof theirfixedand reversedsubspaces.

Examples.
1. If
sino 0'
-sino coso 0 I ,
o
(coso o ')
then
sino 0' / 0 '
-sino 0 0 I , andv= I 0 I .
(0 0 0 0/ 'sin6>/
8 ^ 0,n, weareinthegeneric
With andR isproper.
case,thez-axisisfixed,
2. If
/ 2/3 2/3 -1/3'
R = RG= -1/3 2/3 2/3 ,
' 2/3 -1/3 2/3 )
then
/ 0 1/2 -1/2' /l/2'
A= I -1/2 0 1/2 I andv= I 1/2I .
' 1/2 -1/2 0 / 'l/2/
Inthiscasev isfixed
andR isproper.
3. If

/-1/3 2/3 2/3 '

R = R„= 2/3 -1/3 2/3 ,
V 2/3 2/3 -1/3/
894 © THE MATHEMATICALASSOCIATION OF AMERICA [Monthly 116

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
 thenA = 0, so R is nongeneric.
Thuswe form

/2/3 2/3 2/3' /-4/3 2/3 2/3 '

R + I= 2/3 2/3 2/3 andR-I = 2/3 -4/3 2/3 .
'2/3 2/3 2/3/ ' 2/3 2/3 -4/3/

We findthatx+ = (2/3 2/3 2/3)r is fixed,whilex.,, = (-4/3 2/3 2/3)r

andx_2= (2/3 -4/3 2/3)rspanthereversed subspace.
4. If
/I 0 0'
R = Ro = I= 0 10
Voo V
thenA = 0, and
is theidentity,

0 0'
0 2 0.
0 0 2/

givesv = (2 0 0)r, (0 2 0)r, (0 0 2)r as fixedvectors.

Ourconstruction
5. If
/-I 0 0'
R = R;r= 0 -10,
Vo ° V
aboutthez-axisby an anglen, thenA = 0, so R is non-
whichis a rotation
Weform
generic.
/0 0 0' /-2 0 0'
R + I = (0 0 0 andR-I= 0 -2 0,
'0 0 2/ ' 0 0 0/
and we findx+ = (0 0 2)r is fixedwhilex_,i= (-2 0 0)r and x_,2=
(0 - 2 0)r arereversed.
Compositionwith-I givesa one-to-one between
correspondence properandim-
proper We
transformations.
orthogonal may use to
this obtain
improper from
examples
theproperonesabove.

Remarks.Notethatnomultiplications arerequired tofindanaxisineither

thegeneric
ornongeneric thecosineoftheangleofrotation
case.Similarly, aboutthisaxiscaneas-
ilybe found bytakingthetraceofR. Thetraceis independent ofbasis,andina proper
orthonormal basiscontainingtheaxis,thematrix forR is givenbythematrix Rö in
example 1 above,whose traceis 1
clearly + 2 cos6. Forcomparison,otherapproaches
havebeensuggested tocompute theaxisthatinvolve from1 to3 cross-products
orup
followed
to 18 multiplications, by more cross-products, and
normalizations, projec-
tionsforup toabout40 multiplications tofindthecosineoftheangle[6,Chapt.12].
OthershaveusedtheEuler-Rodrigues formula, derivedwiththeassumption thatan
axisexists,toobtaina varietyofsimilar formulasfortheaxis[1,9, 11,13],yetnone
oftheseclaimtobe proofs ofitsexistence.

December
2009] the axis of a rotation 895

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
 The formulation and proofwe have givenare independent of whetherR is proper
or improper.We have also shownthattheevennessor oddnessof thenumberof inde-
pendentreversedvectorscorrespondsto our intuitive notionof a properor improper
transformation or
preserving reversingright-handed frames. If we so wish,we can
takeevennessor oddnessof thedimensionof thereversedsubspaceas our definition
of a properor improperorthogonaltransformation. As a corollary, properorthogonal
transformations of R3 alwayshave a nonzerofixedvector.
We could also bringorientation intoEuler's theoremby rewriting (1) in theconju-
gation form RAR"1 = RART = A, observing thatA acts as a cross-product, Aw =
v x w, and invokingthetransformation ruleforthecross-product underrotation.Any
furthercorrespondences would involvedevelopingmoreadvanceddefinitions of ori-
entationpreservation, e.g., determinants.
Somewhat in the spirit of Axler [4, 5], we
preferto leave thisto other treatments.

3. EULER'S GEOMETRIC PROOF: TRANSLATION AND COMMENTARY.

Since we have notfoundan Englishtranslation of Euler's originalprooffrom[8] in
theliterature,we provideone here,followedbya discussionof itsmeaning.Paragraph
numbers25 and 26 aretakenfromEuler'soriginalarticle,as is Figure1 below(Euler's
Fig. 2). The authorshave preparedan interlineartranslationof Euler's text(Latinand
verbatimEnglishtranslation on alternatelines) thatmay be viewed online at http:
. math. uci . edu/euler/interlinear
//vmm . htm.

•<^y
Figure 1. (Euler'sFig. 2)

Theorem. In whateverwaya spheremightbe rotatedarounditsowncenter,a diame-

wouldcoincide
tercan alwaysbe chosenwhosedirectionin therotatedconfiguration
withtheoriginalconfiguration.

Demonstration.
25. Let (Fig. 2) circleA, B, C referto an arbitrarygreatcircleof a spherein an initial
configuration, which aftera rotationwill attainthe a, b, c, in sucha way
configuration
thatpointsA, B, C shouldbe rotatedintopointsa, b, c' also, at thesame timeletthe
pointA be an intersection pointof thesetwocircles.Withthissetup,itis to be proved
thata pointO alwaysis givenwhichis relatedin thesame wayto circleA, B, C as to
circlea,b,c. Forthis,therefore, itis necessaryfirstthatdistancesOA and Oa be equal

896 ASSOCIATIONOF AMERICA [Monthly116

© THE MATHEMATICAL

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
to each other;andnext,moreover, itis also necessarythatarcsOA and Oa to thosetwo
circlesbe equallyinclined,or equivalently, thattheangle Oab = theangle OAB: and
therefore also will be thecomplements in two rightangles [i.e., theirsupplementary
angles],thatis, theangles OaA and OAolmustbe equal. However,since thearcs Oa
and OA are equal, angle OaA will also equal angle OAa, and therefore OAa = OAa.
Fromwhichclearly,iftheangleaAa is bisectedbythearc OA,thenthepointO which
is soughtwill be situatedsomewhereon thisarc AO. That point,therefore, will be
discoveredifthearcaO is constructed in sucha waythatangleAaO comesoutequal to
angleOAa. In fact,theintersection of thesearcswill givethepoint0, through which,
ifa diameterof thesphereis constructed, itspositionin therotatedconfiguration will
stillbe thesame as it was in theinitialconfiguration.

26. To definethepointO moreeasily,arcAa can be bisectedbypointM, fromwhich

is constructedarcMO normaltoAa. Thencertainly arcAO can be drawnin sucha way
thatitbisectsangleaAa' and so theintersectionof thesearcs will revealO, whichis
thepointsought.Hereitis observed,ifarcaa [Aa ! !] is takenequal to arcaA, a willbe
thepointof thespherewhich,aftertherotation, attainsthepointA, forwhichreason
angleaAa oughtto be bisected,as opposedto itsadjacentangleaAB.
In paragraph25 Eulerimplicitly uses thefactthatthedirectionand distanceof an
arbitrarypoint,relative to a reference pointon a referencecircle,uniquelycharacterize
itsimagerelativeto theirimagesundera rigidrotationT, usingthisto characterize the
axis O of T by equatingT(0) = O. If we call thereferencepointA, call its image
a, and call some otherpointon the referencecirclea, thenEuler characterizesthe
axis as theuniquepointon thebisectorof angleaAa thatmakestheangles OAa and
AaO equal. The image of thegreatcircleI bisectingangle aAa containsboth O and
a = T(A), so thisprescription can be viewedas thefindingO as theintersection of /
withitsownimage.
In paragraph26, Eulerpresentsa secondmethodforconstructing an axis in which
thebisectorof angleaAa is intersected withtheperpendicular bisectorof thearcAa.
The midpointofAa throughwhichit passes is designatedM in Euler's Figure2. At
theend of thisconstruction Eulernotesthatwhena is further specifiedby T(or) = A,
thenthelengthsof thearcs aa and aA will be equal. This appearsin bothprintings,
althoughsurelyhe means 'Aa' = 'aA'. In thiscase, thebisectorof angleaAa agrees
withtheperpendicular bisectorofa anda. Undera different arrangement ofreference
points- used by Euler in another context- the anglecorresponding to aAB is bisected
to construct theaxis, so thefinalline emphasizeswhichof two anglesformedby the
intersectionof a greatcircleand itsimageshouldbe bisected.1
Euler's characterizations of fixedpointsare indeedcorrectforgeneric,lengthand
orientation preserving motions of a sphereaboutits center.Froma modernpointof
view, we would require some further detailin a completeproof.In particular,Euler's
firstconstruction breaksdown whenT is nongeneric, in whichcase theintersection
pointof a circleand its image,A, is on the equator,and a = T(A) is its antipodal
point.The conditionthatthe angles OAa and OaA are equal holds forany point O
on thegreatcirclebisectingtheanglebetweentheoriginalcircleABC and theimage
circleabc, and cannotbe used to locatetheaxis uniquely.
1We are to a reviewerforpointing
outthesignificance ofthepaper[7] thatEulerpublishedthirteen
grateful
yearsearliersinceit illuminateshis argumentabove. In it,Eulerfindstheinstantaneous axis of a rotational
motionofa rigidbodyaboutitscenterofgravity, alongwhichthevelocityis zero.Sincetheconstruction of [8]
aboveis basedon a somewhatsimilarcharacterization oftheaxis,itis interesting
tocomparethesetwopapers,
andtheinterestedreadercan finda linktotheoriginalpaper[7] (inFrench),ourtranslation,
anda discussionof
howtheconstructions ofaxes inthesetwopapersarerelatedat:http : math. utah . edu/~palais/E177 . html

December2009] the axis of a rotation 897

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
 Eulerstartsfromtheassumptionthata fixedpointO exists.Based on thisassump-
tion,he findstwowaysto characterize itslocationin termsof constructible directions
along which one can proceed from certain points, for a given distance, or untilone
intersects another such great circle. Once this has been accomplished, our modern sen-
sibilitieswould want a demonstration that these constructions are always well-defined,
and thatthepointthatitfindsis indeedfixed.Otherwise,itis possibleto be misledby
a compellingfigureor self-fulfilling assumption.As we havejust pointedout,there
are certaincases where the condition specifyingthe distanceof the axis fromA is
degenerate.
As notedabove, thefactthatT is orientation preserving is no moreexplicitlyin-
voked in thisproof than in that of [7]. There, Euler makes use of distanceand angle
and
preservation, proximity to theidentity, which combine to identify rotations. In [8],
only distance and are
angle preservation visiblyused, so it might seem to applyto
improperorthogonal transformations as well, although we know that it cannot. The
principle that a distance and angle relative to a point on a circle uniquely determines
a locationbreaksdownifwe ignoretheorientation of theangle,and thisis equivalent
to ignoringthedistinction betweenproperand impropertransformations. Because of
we
this, mightexpect Euler to distinguish carefully between angles oppositeorien-
of
tation, since consistent angle orientation is equivalent to orientation formaps of the
sphere. But in fact he equatesangle OAa with the oppositely oriented OaA and OAa at
variouspointsof his demonstration. We see thisas simplyan opportunity, or perhaps
a suggestionfromEuler,thatit maybe possibleto recasthis idea in a settingwhere
bothorientations are treatedsimultaneously as we have done below.Withoutmodifi-
cation, Euler's second construction fails to find thereversedaxis of a genericimproper
orthogonal transformation, since it is no longer theperpendicular
on bisectorofthearc
a
joining point and its image (see Figure4, below). Since the reversed pointsare on
thebisectorof theangleaAa, thefirstconstruction will workiftheconditionlocating
themby equatinganglesis modifiedappropriately.
Euler's theoremis thefirsttopicdiscussedin E. T. Whittaker's classic treatiseon
analyticaldynamics[18], giving an indication of its central importance in rigidbody
dynamics. Whittaker considers a rotation of an arbitrary "rigidbody" for which"the
mutualdistancesof everypair of specifiedpointsis invariable"about an arbitrary
point,0, whose positionin space is unchanged.He reformulates Euler's theorem
in
usingplanes space rather than great circles on a sphere, and providesa proofin
thespiritof Euler's secondconstruction. OthershavenotedthatWhittaker also makes
no explicitreferenceto orientation preservation, his
though proof concludes withan
argument confirming that the line of intersection of two planes he constructs is indeed
fixed.For Whittaker, the conditionthata rigidmotionis propermay be implicitin
the factthatit is pathconnectedto the identity. Since it is nowhereindicatedhow
thatfactentersintotheproof,by ourcurrent standardstheproofwouldbe considered
incomplete.

4. A GEOMETRIC EULER'S THEOREM FOR DISTANCE-PRESERVING

TRANSFORMATIONS. We now aim to give anotherproofof theresultwe proved
above usinglinearalgebra,usingonlythekindof classical sphericalgeometryargu-
mentsthatEuler himselfused, translatedintothemodernidiom.We will show that
length-preserving transformationsof a sphereabout its center,bothproperand im-
proper,have a fixedor reverseddiameter.We will addressthegenericand nongeneric
rigorouslythatthediameterwe construct
cases, and in each case confirm is fixedor
reversed.

898 ASSOCIATIONOF AMERICA [Monthly116

© THE MATHEMATICAL

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
 Butfirst, sincethetheorem ofthefirst section as wellas Whittaker's formulation are
in
presented the framework of orthogonal linear transformations of R3, while Euler's
originaltreatment refers to distance-preserving motions ofa sphereaboutitscenter,
we briefly formalize therelationship between thesetwosettings, so thatwe canfocus
onceand for all on one setting, and so that results abouteither one can be madeto
correspond systematically with one another.
We denotebyO(n) theorthogonal lineartransformations ofR", i.e.,thegroupof
lineartransformations of R" that preserve the standardinner product ofeverypairof
points.Wewilldenote byS""1 theunit in
sphere R", and forx,y e S""1 wewilldenote
byZ(x, y) theanglebetween them, measured in radians.We note thatbydefinition of
radianmeasure, Z(x, y) is theusualspherical distance between x and y; that is, is
it
thelength oftheshorter two
ofthe geodesic(i.e.,greatcircle)segments joiningthem
In
(orTCiftheyareantipodal). particular, this makes S""1 into a metric space.Ifx is
anynonzero in
vector R", we will write x := -^ foritsnormalization, a pointofSn~'
andwe notethatifx andy arebothnonzero thenZ(x, y),theanglebetween them, is
bydefinition Z(x, y). Finally, we willdenotebyRot"thegroupofisometries ofS""1.
IfT is anymapofS""1toitself, there is a canonical waytoextend T toa self-map of
R" thatweshalldenotebyf ; namely, wedefine T(0) = 0, andifx is a nonzero vector,
thenf (x) := ||x|| T(x), wherex = x/||x|| € S""1 is thenormalization ofx. We call
f theconicalextension of T, andwe notetheobviousfactthatitis normpreserving.
Notethatif T is therestriction of an orthogonal mapR of R" to thesphere,then
f (x) = | |x|| T(x) = | |x|| R(x) = R(| |x|| x) = R(x), thatis,conicalextension recovers
an orthogonal map from itsrestriction to the sphere. Next note that ifT € Rot" then
foranynonzero f = =
x andy,Z(f x, y) Z(Tx, Ty) Z(x, y) Z(x, y). = It is now easy
toseethatrestriction tothesphereis actually anisomorphism ofO(n) withRot".

Lemma.A transformationT : R" - > Rnthatpreserves

inner is necessarily
products
linearandhencean orthogonal
transformation.

Proof Lete,-,i = 1,2, . . . , n denotethestandard basisforR", so forx e R", *,-:=

(jc,e,->arethecomponents of x. We willshow that Tx dependslinearlyon thext.If
e¡ := T(e,-),thenbyassumption =
(ej,e'y) (ef,e,), so theej arean orthonormalbasis
=
andhenceTx = £-=i (Tx,e;>e; = £?=1(x,e,-)e{ £Li *,-e{ . ■

Proposition.//Te Rot"thenT € O(n). ThusR '-+ RIS""1andT h+ f are mutu-

allyinverse between
groupisomorphisms O(n) andRot".

Proof Wehaveseenthatf alwayspreserves normsand,sinceT € Rot",f alsopre-

servesangles.Since (x,y) = ||x|| ||y||cosZ(x, y), it followsthatT preserves
inner
and
products, so bythe above Lemma it is an orthogonaltransformation. ■

Thustheorthogonal ofR3(ortheorthogonal
transformations matrices thatrepresent
themwithrespect tosomebasis) andthe isometriesofthe are
sphere just different
ways
ofseeingthesamething.
We denotebyRot+ theconnected component oftheidentityin Rot".Thiscorre-
spondsto thesubgroup of
SO(n) O(n) of elements havingdeterminant +1. Wewill
call elements of thiscomponent rotations or rigidmotionsof S""1 (aboutitscen-
ter).Bothterms andstillusedwidely
arehistorical in bothphysics mathematics.
and
It was clearlythisconceptthatEulerhadin mindwhen hisoriginalstatement
in of

December
2009] the axis of a rotation 899

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
his theoremhe said: "Quomodoqunquesphaeracircacentrumsuumconvertatur" ("In
whateverway a sphereis rotatedarounditsown center").
A thirdequivalentnotiondeservespassingmentionas well,forbothhistoricaland
mathematical context.Ifwe considerthegroupofEuclideanmotionsofR3,i.e.,isome-
trieswithrespectto thestandardmetric,p(jc, y) := ||jc - y''9 theisotropysubgroup
fixingtheoriginis also isomorphicto O(3) andRot3.We notethatthesubspacesiden-
tifiedin the linearalgebraicsetting,e.g., planes and lines throughthe originin E3,
correspond to greatcirclesand antipodalpointson S2, respectively, andraysin E3 cor-
to
respond pointson thesphere.Henceforth we will implicitlyidentify thesedifferent
frameworks while,withoutloss of generality, we explicitlyremainin Euler's context
of transformations of thesphere.
Nextwe listthefourways a pointmaybehaveundera distance-preserving map-
pingT of a sphereto itself.A pointp is fixedby T if T(p) = p, whichimpliesthat
T~l(p) = p. A pointp is reversedby T if T(p) = -p, the antipodalpointof /?,
whichimpliesthatT~l(p) = -p. (Here,thenotation-p indicatestheuniquefarthest
pointon the spherefromp, and does not implya linearstructure.) We collectively
call the fixedand reversedpointsof T its characteristic directions,consistentwith
terminology from linear algebra. pointp swappedby T if p is notfixedunder
A is
T and T(p) ^ -/?, butT(p) = T~l(p). A pointp is generic(withrespectto T) if
T(p) ^ T"1 (/?)- whichimpliesthatp is notfixed.In thelanguageof dynamicalsys-
tems,a genericpointhas period> 2, swappedand reversedpointshaveperiod2, and
fixedpointshaveperiod1, and thisis clearlyexhaustive.
Even the behaviorof two pairs of antipodalpointsdoes not uniquelydetermine
a rigid motionof the sphere.For example,if we findtwo such pairs a quarter-
circumference apart,one pair fixedand the otherreversed,theycould be the axis
of a nongenericrotationand an antipodalpair on its equator,or alternatively, they
could be an antipodalpairon theequatorof a reflection and its axis. The situationis
settledby theuniquenesstheorembelow.A betterknownprototype comesfromplane
geometry, where it is an but
elementary important factthat a Euclidean motionis de-
termined by how it maps any three
noncollinearpoints. The following theorem states
thatthesame principleholds forisometriesof thesphereif we call threepointson a
spherenoncollineariftheyare notall containedin anyone greatcircle.In particular,
no twoofthreenoncollinear pointscan be antipodalpoints.In theframework oflinear
transformations, this condition guaranteeslinear independence of the three associ-
ated vectors,and in thatsettingit is well knownthattheirimagesdetermine a linear
transformation uniquely;but here we derivetheresultfromthekindof elementary
geometric considerations Eulerused.

Uniqueness Theorem. An orthogonaltransformation of E3 is uniquelydetermined

by how it maps any threenoncollinear
points on a sphere.Thatis, ifT' and T2 are
and
orthogonaltransformations P', P2, P3 any threenoncollinearpointson a sphere
centeredat theoriginthenTx(Pj) = T2(P/), 7 = 1,2,3 impliesTj = T2.

Proof We will show thatif T is orthogonaland T(P7) = Pj for7 = 1,2,3, thenT

mustbe the identity.The theoremfollowsimmediately by consideringT = T¡~*Ti.
Let Pi and P2 be nonantipodal pointsfixedby T and q anypointon thesphere.If the
forj = 1 or 2, thenpreservation
distancefromPj to q is zero or a half-circumference
of distanceimmediatelyshows thatq is fixedby T. Otherwise,takingPt as center
and thedistanceof P¡ to q as radiusgives two circlesC, withdistinctnonantipodal
centers.Unlessq is on thegreatcircle/containingPi and P2 (in whichcase itis again
immediatethatq is fixed)thesecirclesintersect in two distinctpointsqx and q2, and

900 © THE MATHEMATICAL

ASSOCIATIONOF AMERICA [Monthly116

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
 Figure2.

eitherq = q' or q = q2. Since Pi and P2 are fixedand distancesare preserved,<?i

and #2musteitherbe fixedor swappedby T. Since anypointthatis equidistantfrom
q' and q2 lies on /,and P3 is fixedand does notlie on /,preservation
of theunequal
distancesfromqx and q2 to P3 forcesbothq' and q2 to be fixed. ■

Our nextstepis to convertEuler's constructionscarefullyintoproofsof existence

of a fixedor reversedaxis forthecase of genericrotations(properor improper)when
a genericpointcan be found,and forthecase of nongenericreflections (in an equator
or axis) when a swapped pointcan be found.Afterthat,it becomes a simpletask
to combinethemwithour uniquenesstheoremfora completeEuleriananalysisof
orthogonal transformations of R3.
We beginouranalysisof distance-preserving transformationsT of R3 abouta fixed
centerC by assumingthatthereis a pointwhose image and preimageare distinct.
We recall thatsuch a pointis called genericwithrespectto T and thatT is called
genericif thereexistssuch a point.If A is a genericpointforT, let a := T(/4), and
a := T~](A). We can modifythe constructions of Euler and Whittaker usingthese
threedistinctpointsA, 0, and a to definea diameterwhosepointsare eitherfixedifT
is properor reversedif T is improper.We will thendevelopalternative constructions
to handleexceptionalcases.

GeometricProofofEuler's Theorem: GenericCase. LetT bea distance-preserving

transformationofa sphereto itselfand let A be a genericpointon S. Thenthereex-
ist antipodalpoints px and p2 such that eitherT(p') = p' and T(p2) = p2, or
T(/?i) = p2 and T(p2) = p'. Furthermore, anypointon S thatis notp' or p2 is also
generic.

Remark. Althoughwe cannotrelyon itfortheproof,itmayhelpto keepin mindthe

followingpictureit implies:eithera, A, and a are on a commonlatitudecircleabout
theaxis of a rotationby 0, where0 < 0 < n, or a and a are on a commonlatitude
circleand A is on theoppositelatitudecircle,and in eithercase, thelongitudeof A is
halfwaybetweenthelongitudesofa and a.

Proof (See Figures3 and 4.) Let /be theperpendicular bisectorof a and a, and T(/)
its image underT. Since T preservesdistances,T(/) is theperpendicular bisectorof
T(a) = A and T(a), and it
therefore is also a greatcircle.Since d(a, A) = d(T(a),

December2009] the axis of a rotation 90 1

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
 pi pi

- A« '
/ <*L

' ' I
JT{1)

P2 pi
Figure3. Figure4.

T(A)) = d(A,a), we have A el. This means a = T(A) e T(/), and since a & /,
T(/) / /and thereare twodistinctantipodalintersection pointsof/ withT(/), p' and
T
p2. maps the antipodalpointsp', p2 € / to antipodalpointsT(/?i),T(p2) £ T(l).
Because px and p2 are theintersection pointsof/ withT(/), theyalso are antipodal
points on T(/).
We will now showthateitherT(p') = px and T(p2) = p2 and T is proper(Figure
3), or else T(p') = p2 and T(p2) = p' and T is improper(Figure4). To do so, we
will showthatforj = 1 and j = 2, trianglespjCiT(pj) and T(pj)Apj are congruent,
as represented symbolicallyin Figure5(a).

p T(p) p T(/?)

A '^^^ ^**'/ a A V-^^^ ^"^^V a

(a) (b)
Figure5.

Since p' and p2 are bothon the bisectorof a and a, d(pj,a) = d(pj,a), and
applyingT to the leftpair,d(T(pj), A) = d(pj, a). Since T preservesdistanceand
a = T(A), we findd(T(pj), a) = d{ph A). By simple symmetry, d(T(pj), pj) =
d(pj,T(pj)), which is the thirdside congruence needed to shows thattriangles
PjCiT(pj) and T(pj)Apj are congruent.But since a,
/?7, and T(/?y)all lie on thegreat
circleT(/), trianglepjCiT(pj) is degenerate,and it is nothardto checkthiscan only
be thecase if triangleT(p/)Apj also is degenerate.In otherwords,T(pj) lies on /
as well as T(/), i.e., T(p') = px and by antipodality T(/?2)= p2, or T(p{) = p2 and
T(p2) = Pi.
We mayalso use d(a, A) = d(A,a) to invokea different setof congruent triangles,
PjAa and T(pj)aA (Figure5(b)), and show thatour construction agreeswithEuler's
in thepropercase. WhenT(p') = px thiscongruenceshowsthat(Figure3) triangle
PxAa is isosceles and thegreatcircleT(l) containinga and p' does make"thesame
angleat a withthearc Aa as /makeswiththearc Aa at A" as in theoriginalconstruc-

902 ASSOCIATIONOF AMERICA [Monthly116

© THE MATHEMATICAL

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
tion.Euler's and Whittaker's constructions mustbe modifiedto workin theimproper
case, sincetheaxis will notbe on the bisector of A and a.
To showthatanypointon 5 other than px, p2 is generic(and hencethatifa generic
pointexistswe areguaranteed to findone amonganytwononantipodal points),we use
ouruniquenesstheoremfororthogonal ofE3
transformations proved earlier toconfirm
thatthetransformation T is therotation about the axis an
pxp2by angle otherthan0
or Ti,perhapsfollowedby a reflection in itsequator.
When T(/?i) = p' and T(p2) = Pi, we can constructa properrotationR about
P'Pi by an angle 0 / 0, n so thatR(/?i) = pu R(«) = A, and R(A) = a / a, and
since p' , a, A satisfytheconditionsof theuniquenesstheorem, T = R and all points
on the sphereotherthanp' and p2 are generic. When T(p') = p2 and T(p2) = P'
we make the analogous argument with the improperorthogonaltransformation R'
an
thatrotatesabout pxp2by angle / 0, n, 0 and then in
reflects its equator,so that
R'(p,) = p2, R'(a) = A, and R'(A) = a / a. T = R' and again everypointon the
sphereotherthanp' and p2 is generic. ■

A 1
' h i'q
' h J
Figure6.

Nextwe treatthecase whenswappedpointsexist.

Geometric Proof of Euler's Theorem: Nongeneric Case. Let T be a distance-

preservingtransformationof a sphere to itself and let A be a swapped point of
T. Then thereexistantipodalpoints p' and p2 such thateitherT(/?i) = p' and
T(p2) = p2 and theequatore bisectingp' and p2 consistsofreversedpoints,or else
T(p') = p2 and T(p2) = p' and theequator€ bisectingp' and p2 consistsoffixed
anypointon S otherthanp' and p2 or thebisectingequatoris also a
points.Further,
swappedpoint.

Remark. If the pointsof € are reversedpoints,thenT is a (proper)rotationby n

in € (improper).
radiansabout p'p2. If the pointsor € are fixed,T is the reflection
Althoughwe cannotrelyon it fortheproof,it mayhelp to keep in mindthepicture
thatitimplies(Figure6): eitherA anda havethesamelatitudeandoppositelongitudes
withrespectto the axis of a rotationby n radians(the axis fromM to - M in the
figure),or theyhave the same longitudesand oppositelatitudeswithrespectto the
equatorof some reflection (theequatorbisectingQ and - Q in thefigure).

December2009] the axis of a rotation 903

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
Proof.SinceA anda arenonantipodal, thereis a uniquegreatcircle,l', containing
A anda, andbecauseT takesgreatcirclesto greatcircles,it takesl' to thegreat
circlecontaining T(A) = a andT(a) = A, i.e.,to itself.Nowa distance-preserving
transformation mapstheperpendicular bisector ofanytwodistinct pointstotheper-
pendicular bisector of their images. Since A and a are swapped, also mapsZ2,the
T
perpendicular bisector of A and a, to itself. Thus T takes theintersection ofthetwo
perpendicular great circles l' and Z2 to itself. This consists of two points,themid-
point,M, of A and a, and itsantipodal point-Af,distinguished bythefactthatM
is closerthan-M to bothA anda. Thesepointsmustbe individually fixedsince
= =
d(T(M), a) d(T(M), T(A)) d(M, A) d(-M, a). < Then since M and -M are
fixed,T also the
mapsZ3, equatorbisecting M and -M, to itself.This greatcircleis
perpendicular toboth l' and Z2since it bisects of
pairs points on both.We denotethe
of intersection of Z3 with lxby Q and - Q, and of Z3 withZ2byR and
antipodal points
=
-#. SinceT(Z,) /,, 7 1,2, 3,= we know that the set {ß, -ß} is mappedtoitself.
Ifbothpoints were fixed, the circle h containing M, g, and -Q could onlycontain
fixedpoints, but since it also contains A and a which are swapped, deducethat
we
Q and-g must be reversed. For the same reason, we know thattheset{R, -R] is
mappedtoitself.
If R is a reversed point,therotation byn abouttheaxis containing M and - M
fixesM andswapsQ andR withtheir antipodal points, so bytheuniqueness theorem,
T is equivalent to this rotation, and all points otherthan the fixed pointsM and -M,
andpointsonZ3, which consists entirely of reversed points, will be swappedpoints.
If R is a fixedpoint,thereflection inZ2,theequatorbisecting Q and-Q andcon-
taining M, -M, R, and-Ä, fixesM and/?andswapsQ withitsantipodal point,so
bytheuniqueness T is
theorem, equivalent to thisreflection, and all pointsother than
thefixedgreatcircleZ2 and the reversed pointsQ and - Q will be swappedpoints.■

theorem.
theproofofthefollowing
Wearenowreadytocomplete

Theorem.LetTbea distance-preserving ofa spheretoitselfThenT

transformation
maybe classified directions
and itscharacteristic according
identified to one ofthe
usingno
sixpossibilities,
following more thanthree points,and their
noncollinear
imagesandpreimages underT.
1. T is theidentity,
2. T is theantipodal map,
3. T isa rotation by aboutan axis,
n
4. T is a reflectionacrossan equator,
5. T is a genericrotationaboutan axis,
6. T is a genericrotationaboutan axisfollowed acrossitsequator.
bya reflection
Cases 1, 3, and 5 areproper, If T is proper,
and cases 2, 4, and 6 are improper.
thereis at leastonepair ofantipodal fixedpoints,and if T is improper,thereis at
leastonepairofantipodal reversedpoints.

Proof.Ifnogeneric orswapped pointsexist,theneverypointonthesphere mustbe ei-

therfixedorreversed.Sincedistancesarepreserved,a reversedpointcanonlybefound
a quarter-circumferencefroma fixedpoint,andfrom thisifeverypointis eitherfixed
orreversed, andT is theidentity
infactall pointsmustbe fixed, (proper),orall points
mustbe reversed,andT is theantipodal map(improper). Otherwise,genericpoints
existor swappedpointsexist,andtheabovetheorems guaranteethatthesecasesare

904 ASSOCIATION
© THEMATHEMATICAL 116
OFAMERICA[Monthly

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
mutuallyexclusive.If a genericpointexists,thetheoremforthegenericcase tellsus
how to findthefixedor reversedaxis, and classifyT as a properor impropergeneric
rotationaboutthataxis. If insteada swappedpointexists,the theoremforthe non-
genericcase tellsus how to findthefixedaxis and reversedequatorclassifyingT as a
properrotationby it aboutthataxis,or thefixedequatorand reversedaxis classifying
T as an improper acrossthatequator.
reflection ■

Remark. Algorithmically, we know that most orthogonaltransformations T are

generic. In thatcase, unless we chose one of the two fixed or reversed points,the
very firstpoint whose image and preimage we examine will immediately classifyT
and findits axis. If thefirstpointwe examineis notgenericbutswapped,it stillsuf-
ficesforclassification and findingthecharacteristic
directions.If it is neither,
we can
examinea secondpointone-quarter circumference fromthefirst.If it is notgeneric,
thenT cannotbe generic,sincetheonlynongenericpointsof a genericT are antipo-
dal. If itis notswappedeither,we can sampleone morepointa quartercircumference
fromeach of thefirsttwo.Each of thesethreewill be eitherfixedor reversed,and the
particular combinationtellsus thatT is theidentityor theantipodalmap,or thatT is
a nongenericrotationor reflection, forwhichtwoof thepointsare on itsequatorand
theotheron itsaxis.

5. OTHER PROOFS OF EULER'S THEOREM. Finally,we surveybrieflysome

classic approachesto proofsof Euler's Theorembased on severalareas of modern
mathematics: linearalgebra,topology,and differentialgeometryand Lie theory.To
anyone familiarwithone of these the
fields, correspondingproofwill seem succinct,
and so "superior"to Euler's proofbased on sphericalgeome-
natural,and intuitive,
try,givenabove. But thisis analogousto comparingtheoftencomplicatedgeometric
proofsthatIsaac Newtongives in his Principiawiththeslickdifferential-equations-
based proofsof the same factsin a moderntreatiseon classical mechanics.In both
cases, one shouldkeep some historicalperspectiveand rememberthatthe succinct-
ness and eleganceof themodernapproachesis onlypossiblebecause theyare based
on a quitemassivemathematical thatwas builtup overmanyyearsby
infrastructure
generations of mathematicians.

1. Approach Based on Linear Algebra. Perhapsthe best-knownproofof Euler's

ofa matrixand so, via
Theoremis based on thetheoryofeigenvaluesand eigenvectors
on thecharacteristic
thedeterminant, polynomialof thematrix.Once one has under-
stoodtheseadmittedly somewhatadvancedlinearalgebraconcepts,theproofis short
Here in briefis one version.
and transparent.
• If A is a matrix,thennontrivial
solutionsof Av = k' correspondto eigenvaluesof
A, namelyrootsof thecharacteristicpolynomialof A: Xa(^) •= det(A - AI). Note
=
thattheconstanttermx (0) det(A) is theproductof theeigenvalues.
• The characteristic
polynomialof a real 3x3 matrixA is a cubic withreal coeffi-
cients, so there
are twopossiblecases; either:
(R) xa has threerealroots,rÌ9r2,and r3,or
(C) xa has a realroot,r1?and distinctcomplexconjugateroots,r2and r3 = r%.
•
Anyeigenvalueof anyorthogonalmatrixhas complexmodulus1. In case (R), this
saysTj = ±1, j = 1, 2, 3, and in case (C) it says r2r3= 1.
• Sincethedeterminant ofa properorthogonal matrixA is + 1, theproductofitseigen-
valuesis +1, andin case (R) thissaystheyareeither{+1, +1, +1} or {+1, -1,-1}.

December2009] the axis of a rotation 905

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
 In case (C) it says thatrx- 1. Since in bothcases +1 is an eigenvalue,thereis al-
waysa nontrivial solutionto Av = v.
•
Similarly,if A is an improperorthogonaltransformation, thenthe productof its
eigenvalues is -1, and in case (R) this says that they are either {+1, +1, -1} or
{-1,-1,-1}. In case (C) thereal rootis rx= - 1. Thus - 1 is alwaysan eigenvalue,
so thereis a nontrivial solutionto Av = -v. ■

Remark. This proofcan be generalizedto a rotationA of R" by observingthatits

realeigenvaluesare ±1 and thecomplexones come in complexconjugatepairs.Then
W1can be written as a directsumof orthogonaleigenspaces,and theeigenspacesare
even dimensionalforeach complex-conjugate pair.The - 1 eigenspacemustalso be
evendimensional(and mayalso be viewedas thesumoftwo-dimensional eigenspaces
on whichA acts as a half-turn rotation).Therefore,ifn is odd,the 1-eigenspacemust
be nontrivial.In [12], thisproofis used to findtheaxis in thenongenericcase. In the
genericcase, theaxis is derivedfromtheEuler-Rodrigues formula[14, 15], givinga
rotationmatrixin termsof an axis and anglethatare assumedto exist,so itcannotbe
used as an existenceproof.

A different formulaofRodrigues[16], whichis equivalentto Hamilton'squaternion

multiplication and appearedearlier,can be used to proveEuler's theorem, as notedin
[2, 3]. Let Rj denotetherotationabouta unitvectoru7 by angleOj, and let (c;, Sj) =
(cos -Tf,sin-j-). The formulastatesthatR2R1 = R3 wherec3 = c'c2 - £1^2(1*1 • u2)
and 5"3u3 = C's2u2+ c2s'U' + S's2(u2 x Ui). If R is a 3 x 3 rotationmatrix,and v sat-
isfiesw = v x Rv ^ 0, thenR is a rotationaboutw takingv to Rv, followedbya rota-
tionaboutRv. We can also writeR as theproductofthree"Eulerangles,"i.e.,rotations
aboutcoordinateaxes. Withproperconsideration of nongenericcases, thesedecom-
positions, combined withtwo or threeapplicationsof Rodrigues'formula,amountto
a constructive proof of Euler's theorem.

2. Approach Based on Topology. A simplecorollaryof theLefschetzFixed Point

Theoremis thatifa compactmanifoldX has nonzeroEulercharacteristic, x CK)>then
any continuous self-mapping of X thatis homotopic to the must
identity have a fixed
point.Recall thatbi(X), the ith Betti number of X, is the rankof its ith homology
groupand theEulercharacteristic is thealternating
sum £,•(- l^ft/ÍX).Now forthe
sphereS", b0 = bn = 1, and all otherBettinumbersare zero,so x(S") is twoor zero
dependingon whethern is even or odd. So we see thatin particularany continuous
self-mapof S2kthatis homotopicto the identitymap will have a fixedpoint.Since
Euler's conceptof a rigidmotionof a spherecertainlyincludedthatthemotionwas
theendpointof a continuousfamilythatstartedat theidentity, Eulerwouldhave seen
thisas a generalization of his result.The Lefschetztheoremuses quite sophisticated
topologicalarguments, butforthespherethereis an elegantand simplerapproachthat
as
getsnearly strong a result.

Theorem (Borsuk-Hirsch). Let f : S" -> Sn be a continuousself-mapof the n-

dimensionalsphere.If f does nothave anyfixedpoints,thenit is homotopicto the
antipodalmap -I. Similarly,ifthereis no pointp ofSn such thatf(p) = -p, then
f is homotopicto theidentity
map I.

Proof For 0 < t < 1, define0, : Sn -► Rn+l by 0,00 = (1 - 0/00 - tx. If for
some t and x, (/>t(x)= 0, thentx = (1 - t)f(x), so takingthe normof bothsides
and using ||*|| = ||/(jc)|| = 1, t = 1 - t so t = '. Then 'x = 'f(x) so f(x) = x.

906 © THE MATHEMATICAL

ASSOCIATIONOF AMERICA [Monthly116

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
Hence,if / has no fixedpoints,thenO,(jc) := 0*00/110*0011definesa homotopy
between/ and-/. Changing thedefinitionof0, to 0,00 := (1 - t)f{x) + tx,the
sameargument showsthatifthereis nox e Snwithf(x) = -jc, then0,00 defines a
homotopy between / and/. ■

Supposen = 2k is evenand writeR2kas theorthogonal directsumof k two-

dimensional subspaces.If we simultaneouslyrotateeachof themthrough an angle
t,thenas t variesfrom0 ton we geta homotopy of/ with- /. Thustheidentity map
andtheantipodal mapof S2k~larehomotopic. However, thedegreeoftheidentity
mapofS2kis +1 whilethedegreeoftheantipodal mapofS2kis -1. Sincehomotopic
mapsofann-dimensional manifold the
have samedegree, itfollowsthatforaneven-
dimensional spheretheidentitymapandtheantipodal maparenothomotopic, andso
bytheabovetheorem anycontinuous : ->
mapf S2k S2keither hasa fixedpointp or
a pointp thatis mappedto itsantipodal point-p. Ofcourse,if/ is an orthogonal
transformation, andhencehasa fixedpoint,andithas
ithasdegree+1 ifitis proper,
-
degree 1 ifitis improper,andhencethereis a pointp thatis mappedtoitsantipodal
point.See [10]formoredetailsonthisapproach.

3. ApproachBased on Differential Geometry and Lie Theory.Euler'sTheorem

is an easyconsequenceofthefollowing threewell-knownresultsfromRiemannian
geometry (See thediscussion
andLie grouptheory. below.)

Hopf-Rinow-DeRham Theorem.A Riemannianmanifold M is geodesically

com-
pleteifandonlyifitis metrically Hence,if
complete. M is then
compact, every pairof
pointsofM canbejoinedbya geodesicparameterizedbyarclength.

A compact
Proposition. Lie groupalwaysadmitsa Riemannian metricthatis invari-
antunderbothleftandright
translation.
Moreover, forany suchmetric,thegeodesies
toarclength
proportionally
parametrized at
and starting theidentity exactlythe
are
one-parameter Hence,everyelement
subgroups. a
of compact Lie groupliesonsome
one-parametersubgroup.

Lemma. Theone-parameter subgroupsofSO(n) are of theformg(t) = exp(rA),

x
whereA is somen n skew-symmetric Thatis,if'€ M",thenx(t) = g(t)y
matrix.
equationx'(t) = Ax(i) withinitial
is theuniquesolutionof thelineardifferential
x(0) v. Hence,ifA v 0, theng(t)' = yforall t.
condition = =

Forproofssee [17],pages342,401,and378.
Now supposeg e SO(3). By theaboveproposition andlemmathereis a 3 x 3
skew-symmetric matrixA anda realt suchthat =
g exp(íA), andas we sawinSection
2, if

an -031 '
-ai2 0 a23 I ,
(0 a3l -¿*23 0 )

andweletv := (a23«3i an)1' thenAv= 0, so gv = v,i.e.,v is anaxisforg. (Ifv = 0,

thenA is thezeromatrix ThisprovesEuler'sTheorem. ■
andso g is theidentity.)

December
2009] theaxis of a rotation 907

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
DEDICATION AND ACKNOWLEDGMENTS. Dedicatedto thememoryof LeonhardEuler,"themaster
ofhisbirth.
ofus all,"on theoccasionofthe3OOthanniversary
The authorsgratefully suggestionsfroma reviewerthatresultedin significant
acknowledgeimportant
improvements in thefinalmanuscript.

REFERENCES

1. R. Alperin,The matrixofa rotation, CollegeMath.J.20 (1989) 230. doi : 10 . 2307/2686771

2. S. Altmann, Hamilton,Rodriguesandthequaternion scandal,Math.Mag. 62 (1989) 291-308.
3. , Rotations,Quaternions and Double Groups,Dover,New York,2005.
4. S. Axler,Down withdeterminants! thisMONTHLY102 (1995) 139-154. doi : 10 . 2307/2975348
5. , LinearAlgebraDone Right,Springer- Verlag,New York,1997.
6. B. Casselman,A Manual of Geometry and PostScript, CambridgeUniversity Press,Cambridge,2005;
also availableat http : //www. math.ubc . ca/'cass/graphics/manual/pdf /chl2 .pdf.
7. L. Euler,Découverted'un nouveauprincipede mécanique[El 77], Mémoiresde l'académiedes sciences
de Berlin6 (1752) 185-217; reprinted in Opera Omnia,series2 5, 81-108.
8. , Formulaegeneralesprotranslatione quacunquecorporum rigidorum [E478],presentedtotheSt.
Petersburg Academyon October9, 1775.NoviComm.Acad. Sci. Petropol.20 (1776) 189-207;reprinted
in Opera Omnia:series2 9, 84-98, andin TheoriaMotusCorporum SolidorumSeu Rigidorum [E478a],
ed. nova,1790,449-460.
9. J.P. Fillmore,A noteon rotation matrices, IEEE JournalofComputer Graphics& Applications 4 (1984)
30-33. doi : 10 . 1109/MCG . 1984 . 275935
10. A. GranasandJ.Dugundji,FixedPointTheory,Springer, New York,2003.
11. B. K. P. Horn,Closed-form solutionof absoluteorientation usingunitquaternions, J. Opt.Soc. Amer.A
4 (1987) 629-642. doi : 10 . 1364/JOSAA . 4 . 000629
12. W. Kahan,Cross-products and rotations in 2- and 3-dimensional Euclideanspaces (2003), availableat
http : //www . cs . berkeley . edu/~wkahan/MathHl 10/Cross . pdf.
13. D. Kaiman,The axis ofa rotation: Analysis,algebra,geometry, Math.Mag. 62 (1989) 248-252.
14. B. Palais,A newformulafora rotation matrix(to appear).
15. B. Palais andR. Palais,Euler'sfixedpointtheorem: J.FixedPointTheoryAppi 2
The axis ofa rotation,
(2007) 215-220. doi:10.1007/sll784-007-0042-5
16. O. Rodrigues,Des lois géométriques qui régissent les déplacements d'un systèmesolidedans l'espace,
J.Math.PuresAppl.5 (1840) 380-440.
17. M. Spivak,A Comprehensive Introduction to DifferentialGeometry, vol. I, 3rded., Publishor Perish,
Houston,TX, 1999.
18. E. T. Whittaker, A Treatiseon theAnalyticalDynamicsofParticlesand RigidBodies , Dover,New York,
1944; reprint ofthe4thed., CambridgeUniversity Press,Cambridge,1937.

BOB PALAIS receivedhis B.A. fromHarvardCollegein 1980 andhis Ph.D. fromUC Berkeleyin 1986.He
is currently
a ResearchProfessorat theUniversity of Utahin Salt Lake City.He splitshis timebetweenthe
Mathdepartment (teachingand doingresearchon scientificcomputation and mathematical visualization)and
thePathologyDepartment, wherehe workson DNA meltinganalysisand bioinformatics. His article"Pi is
wrong!"(http: //www. math,utah. edu/~palais/pi .html)stillgeneratesthemostspiriteddiscussions.In
his sparetime,he enjoysmountaineering,oftenwithothermathematicians.
MathDepartment, of Utah,Salt Lake City,UT 84112
University
palais @ math.
Utah,edu

RICHARD S. PALAIS receivedhis B.A. fromHarvardCollege in 1952 and his Ph.D., also fromHarvard,
in 1956. Aftera longcareerin teachingand theoretical researchat Brandeis,he partiallyretiredin 1997 to
workon developinga mathematical visualizationprogram(see http://3D-XplorMath.org).He movedto
UC Irvinein 2004, wherehe teachesparttimewhilecontinuing to prog^m and writeaboutnovelalgorithms
forvisualizingcomplexmathematics. He is proudofhismanymathematical descendants(70 accordingto the
Mathematical GenealogyProject)and also foran entry(together withLuc Bénard)in the2006 NSF/Science
Magazine ScientificVisualizationChallengethatwon FirstPrize and was on the Sept. 22, 2006 coverof
Science(http : //www . sciencemag . org/content/vol313/issue5794/cover . dtl).
Department ofMathematics, UniversityofCalifornia,Irvine,Irvine,CA 92697
palais @ math.uci.edu

908 © THE MATHEMATICALASSOCIATION OF AMERICA [Monthly 1 16

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions
STEPHEN B. RODI had seven yearsof Jesuitseminarytrainingafterhighschool includinga three-year
philosophy curriculumtaughtin Latin.This explainshis B.A in 1965 fromSpringHill College witha triple
majorinmathematics, philosophy,andLatin.He holdsa master'sdegreefromMarquette University(1967) and
a Ph.D. fromThe UniversityofTexasat Austin(1974), bothin mathematics. Since 1976he variouslyhas been
a departmentaladministratorand mathematics facultymemberat AustinCommunity College. He looksback
fondlyon his involvementin manymathematics duringthe 1980sand 1990s,includingthegrowth
initiatives
of theAmericanMathematical Associationof Two-Year Colleges,theconsortium calculusprojectbased at
Harvard, andthefirst
incarnationoftheMathematical SciencesEducationBoard(MSEB). As retirement nears,
morethaneverhe enjoysreadinghistory urgeto helpstudents
ofall sortsandgivingin to a grandfatherly with
theirmathematicalquestions.
Department ofMathematics,Rio GrandeCampus,AustinCommunity College,1212 Rio GrandeSt.,
Austin,TX 78701
srodi@ austincc.
edu

Yet Another Calculus Proof of the Fundamental Theorem of Algebra

polynomialwithcomplexcoefficients.
Theorem. Suppose p is a nonconstant
Thenp has a rootin C.

Proof. We thinkof p as a functionfromE2 to R2 and applytechniquesof ad-

vanced calculus. Of course,p is continuouslydifferentiate, and the set Z =
:
{z p'z) = 0} is finite. At each point a e C ' Z, p'(a) is invertible;thus,by
theinversefunctiontheorem,a has a neighborhood containedin C ' Z whose
imageis a neighborhood of p(a) on whichp is invertible. As a result,theimage
is
p[C ' Z] open.
We now claim thatp[C] is closed. To see this,considerany convergent se-
quence {p(Zk)} in p[C]. It is easily verified that 'p(z)' - > oc as 'z' -> oo,
so the sequence {zki mustbe bounded.Thereforeit has a subsequence{zkm}
thatconvergesto some b e C. Thus, by the continuity of /?,lim^^ pdk) -
=
lim,,,.^.p(Zkm) p(b)- This shows that each convergent sequencein p[C] con-
verges to a in
point p[C': thus p[C] is closed, as claimed.
Now considerp[C] n (C ' p[Z]) and p[C ' Z] n (C ' p[Z])' the firstis
closed in C ' p[Z] and thesecond is open in it. But it is easy to see thatthese
two sets are equal, and C ' p[Z], like thecomplementof any finiteset in C, is
clearlyconnected.Thereforep[C] n (C ' p[Z]) = p[C ' Z] H (C ' p[Z]) must
be either0 or C'p[Z'.
Since C ' p[Z] is denseand p[C ' Z] is open,p[C ' Z] n (C ' p[Z') cannot
be empty,so p[C] n (C ' p[Z]) = p[C ' Z] n (C ' p[Z]) = C ' p'Z' Thus
p[C] 3C' p[Zl and since p[C] is closed and C ' p[Z] is dense,p[C] = C.
In otherwords,p(z) takeson each value in C, and,in particular, /; musthave a
root. ■

- Submittedby RalphKopperman,CityCollege of New York

December2009] the axis of a rotation 909

This content downloaded from 134.68.189.135 on Thu, 21 Mar 2013 23:54:57 PM

All use subject to JSTOR Terms and Conditions