Annals of Mathematics

Commuting Vector Fields on S3

Author(s): Elon L. Lima
Source: Annals of Mathematics, Second Series, Vol. 81, No. 1 (Jan., 1965), pp. 70-81
Published by: Annals of Mathematics
Stable URL: http://www.jstor.org/stable/1970383 .
Accessed: 30/09/2014 16:12

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].

Annals of Mathematics is collaborating with JSTOR to digitize, preserve and extend access to Annals of
Mathematics.

http://www.jstor.org

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
 Commuting Vector Fields on S3

By ELON L. LIMA*

In this paperan answer is given to the question, firstposed to me by

S. Smale, of whetherthe sphereS3 admitsa pair of commutingvectorfields
that are linearlyindependentat each point. The answer is no. The same
problemwas consideredindependently by J. Milnorwho,in the listofproblems
of the Seattle TopologyConference(Summer 1963), definedthe rank of a
manifoldM as the maximalnumberof commutingvectorfieldson M that are
linearlyindependentat each pointand asked,amongotherthings,what is the
rankof S3.
In Milnor'sterminology, our main result(statedand provedin ? 4) is that
everycompact3-manifoldwithfinitefundamentalgrouphas rankone.
I am gratefulto S. Smale and A. Haefligerforstimulatingconversations
on this subject,and to J. Stallingsforusefulinformation.
1. Preliminaries
We shall adopt the C- pointof view, so that theworddifferentiableis to
meaninfinitelydifferentiable.In thissectionwe-reviewthe basic facts to be
used later.
A differentiable manifoldM
action of a Lie groupG on a differentiable
is a differentiablemap q: G x M-> M such that 'p(gh,x) = q(g, q(h, x)) and
q(e, x) = forany g, h e G, x e M, wheree denotesthe neutralelementof G.
x
For a given g e G, we writeq-: M-> M to denotethe map definedby q,(x) =
q(g, x). Each Ad is a diffeomorphism,whose inverse is q'-i. The action g
induces,in a well-knownway, a Lie algebra homomorphism g-%:B DX(M)of
theLie algebraofG intothe space of all differentiable vectorfieldson M, called
theLie homomorphism of q. Conversely,whenM is compactand G is simply-
connected,any Lie algebra homomorphism h: 9 DX(M)is of the formh =A
forsome action,q-:G x M a M [3, p. 82]. We shall use thisfact mainlywhen
G Ra, and then,to give q-*is the same as givingn differentiable vectorfields
Xi, ***, X. on M such that [Xi, XJ]= 0 i, j = 1, ... , n. In this case, we say
thatthegiven vectorfieldscommute. An actionof the additivegroupof the
reals is called a flow,so that a differentiable flowon a compactmanifoldM is
equivalent to a differentiablevector field on M. Given2 commutingvector
fieldsX, Y on a compact manifold M, the relation between the flows
* Fellow of the Guggenheim Foundation. Work partially supported by the National Science
Foundation under Grant #NSF GP1904.

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
 COMMUTINGVECTOR FIELDS ON SI 71

a 7: R x Mu M, generated respectively by X and Y, and the action

9: RK x M- Mdeterminedby themis that,forany r = (s, t) e R, x E M, one
has Pr(x) = M8Qt(x))= Yqt(MX)).
Givenan actionA: G x M-UM,theorbitofa pointx e M is thesetp(G x x)=
{9(g, x); g e G}. The isotropygroup of x is the set Gx= {g E G; q(g, x) = x}.
Gxis a closedsubgroupof G. Givenx e M, the map g pg(x) of G onto the
orbitof x induces,by passage to the quotient,a one to one continuousmap
G/GX M of the homogeneousspace G/Gxonto the orbit of x. When 9 is
-

differentiable, such map is regular (i.e., has an infectivederivativeat each

point)and,of course,whenG/Gx is compact,it is a diffeomorphism of G/Gxonto
the orbitof x.
We are interestedin the cases G = R and G = R'. Givena differentiable
flow#:R x M - M, the isotropygroupof a pointx e M may be {O}, a discrete
subgroupwitha generators, > 0, or thewholereal line. In the firstcase, the
orbitof x is a non-compact one to one regularimageof R. In thesecondcase,
the orbitof x is diffeomorphic to a circleand we say it is a periodic orbit,of
periods,, meaningthat28(x)= x if and onlyif s = n *s,, n integer. In the last
case, x is a fixedpointof the flow#,that is, &8(x)= x foreverys. If X is the
vectorfieldassociatedto the flowI, we referto the orbitsof the flowe as X-
orbits and a fixedpoint of e is a pointx e M such that X(x)- 0, that is, a
singularityof X.
WhenG = R', theisotropygroupGxof a pointx e M undera differentiable
actionA: R2 x M-y M may be one of the following. It may be the wholeR',
and then x is a fixedpointof p: its orbitreducesto {x}. Next, supposethat
dimGx= 1. Then Gxmay eitherbe a line throughthe originof RK,in which
case R2/Gx R and the orbitof x is a regularone to one image of R, or else
Gx consistsof a sequence of parallellinesL + l v, n = 0, ?] 1 +2, *** where.

L is a line throughthe originof RKand v e R is a vectornot containedin L.

In this case R2/Gx circle,so the orbitof x is a simpleclosedcurve. Finally,
theisotropygroupGxmayhave dimension0, that is, it may be a discretesub-
groupof the plane. This is thecase we shall considermostly.Here thereare 3
possibilities.The firstone is thatGx= {0}. Thentheorbitof x is a regularone
to one image of the plane RK. The second is that Gx = {nv; n = 0, +1, ?2, ..**
is a cyclicgroupwith one non-zerogeneratorv e R. ThenRK/Gx is an infinite
cylinderof which the orbit of x is a one to one continuousimage. The last
possibilityis thatGxbe a freeabeliangroupon 2 generators: Gx= {mu + ny;
m, n = 0 ? 1, + 2, *. } where u = (a, b) and v = (c, d) are linearly independent
vectorsin theplaneRK. ThenR2/Gxis a torusand theorbitof x is diffeomorphic
to this torus. Again we remarkthat the orbitof x is compactif and onlyif

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
72 ELON L. LIMA

R2/GX is compact. Let X, Y be commutingvectorfieldscorresponding to the

actionA: R2 x M - M. A pointx e M is fixedunderg if and onlyif X(x) =
Y(x) = 0. The orbitof x under p is 1-dimensionalif and only if X and Y
are linearlydependent(but not simultaneouslyzero) at x. The orbitof x is
2-dimensional if and onlyif X(x) and Y(x) are linearlyindependent.
Giventhe actionA: G x Ma M, if two pointsx, y e M belongto the same
orbitof A, thenthe isotropygroupsGxand G, are conjugate subgroupsof G.
So, whenG is abelian,2 pointsin the same orbithave thesameisotropygroups.
A set Xc M is said to be invariant under g if, for any x e X and g e G,
pg(x) e X. A subsetX c M is said to be a minimal setunderq' if it is compact,
invariant,non-empty, and containsno propersubset with these 3 properties.
By Zorn's lemma, any non-emptyinvariantsubset contains a minimalset.
Givena minimalset pAc M, theorbitof any pointx e pais dense in jte,and from
this it followsreadilythat any 2 pointsof pahave conjugate isotropygroups.
When G is abelian, the isotropygroupof all pointsin a given minimalset is
the same.
The o-limit set of a pointx e M under a flow0: R x M- AMis the set
&+,O(x)of all points y = limnE8,(x), where s$, ) Co. When M is compact, the
co-limit set _+,(x)of each point x e M is a non-empty,connected,invariant,
compactsubsetof M.
To concludethisintroductory section,we provea topologicallemmawhich
is well known,but we need everypart of its statement.
Let V= VI1 be a closed, connected (n - 1)-manifold, topologically
embeddedin a compact,connectedn-manifold M = M. The homologygroup
Hni( V; Z2) has onlyone non-zeroelement. Denote by i: V c M the inclusion
map. When i,: Hni-(V; Z2) Hn,-i(M;Z2) is the zero homomorphism, we say
that V boundsin M.
From now on we omitmentionof the coefficient groupsin homology,and
cohomology,assumingthemto be always Z2.
LEMMA 1. If Vn-l boundsin MB, then
(a) M - V has 2 connected components;
(b) V is thecompletepoint-setboundaryof each componentof M -V;
(c) Given a coordinatesystemx: She Rn such thatx(&?)is a ball around
the origin and x(&?n V) = {(x1, *, Xn) e x(&?);Xn = O}, let a, b e &?be points
such thatx(a) and x(b) lie in differentsides of the hyperplaneXn= 0. Then
a and b belongto distinctcomponentsof M - V.
PROOF. We prove(c) first.Replacing,ifnecessary,&2by a smallerdomain,
we may assume that x is also definedin the boundaryof &?. Clearly 72- V

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
 COMMUTINGVECTOR FIELDS ON S3 73

Fig. 1

has 2 connectedcomponents A, B, witha e A, b e B. Suppose,bycontradiction,

that a and b could be joined by an arc X,c M - V. Of course, X, is not
containedin &?. Let a' be the last pointof X, that lies in the boundaryof A.
Replacethe part of X, that goes froma to a' by an arc aa' such that x(aa') =
linesegmentjoiningx(a) to x(a'). Do thesame to b. Let X,be an arc connecting
a to b in &?such that x(X) = line segmentfromx(a) to x(b). Then aa', the
untouchedmiddlepartof -X,b'band X,forma closedcurveX in M, intersecting
V transversallyand onlyat onepointc. Hence theintersectionnumber(mod2)
of Vand X is Vim = 1. On the otherhand, V bounds in M, so V#X= O.
This contradictionproves(c).
The proofsof (a) and (b) may also be given by geometricarguments,
providedthat we assume V nicely embeddedin M (as in the case we shall
encounter). But it is neater-if less intuitive-to use the Lefschetz duality
theorem,accordingto which HO(M- V) H"(M, V). In orderto compute
H"(M, V), we considerthe cohomologyexact sequence:
H n-1
(M) H n-f
(V) ) Hn(M, V) Hn(M) Hn( V).

The firsthomomorphism is zerobecause it is the transposeof i,. The last group

is zero because dim V = n - 1. The secondand fourthgroupsare isomorphic
to Z2. Hence Hn(M, V) Z2 + Z2, So H(M- V) Z2 + Z2, and M- V has
2 connectedcomponents.This proves(a).
As to (b), it is clear that the point-setboundaryof each componentof
M - V is containedin V. To provethe converseinclusion,it suffices to show
that, forany open ball B c V, the subset V' = V - B does notdisconnectM.
Now V' is compactand Hn-l(V') = 0. Using dualityand exact sequence as
before,we see that HO(M - V') Hn(M, V') Z2, So M - V' is connected.

2. The main argument

THEOREM 1. Let X and Y be commutingvectorfields on a compact

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
74 ELON L. LIMA

simply-connected 3-manifoldM. Assume that X has a closedorbitand no

singularities on M. Then theaction of R2 on M determinedbyX and Y has
a compactorbit. (That is, a circle or a torus.)
PROOF. Denote by A,r2:R x M-a M the flowsgeneratedby X and Y
respectively,and let A: R2 x M-a M be the corresponding actionof R2 on M
so that,forr = (s, t) e R2 and x e M, wp(x)= ti7t(x)= itt(x). Considera point
xeE M whosea-orbitis closed,say of periodso > 0, so E80(x)= x0,but E8(xo)+
x0for0 < s < s0. Call K theclosureof theq-orbitofxo. By continuity, E8O(x)
x foreveryx e K. Since X has no singularities,the &-orbitof every x e K is
then closed, with a periodof the formso/n,n > 0 an integer. ClearlyK is a
compact,9-invariant,non-empty subsetof M, so we maychoosea minimalset
IAc K forthe actionA.
The X-orbitof everypointx e j is closed,withthe same periods, = so/n,.
In fact, by the minimality of pA,given any two pointsx, y e pA,each of them
belongsto theclosureof the9-orbitof theother. Supposethat,forsomes E R,
Es(x) =x. We maywritey = limepWrn(X),r- co in R2. Hence

Eg)=lim Eswrn(X) = lim (P, S~X) = lim P'P~) = y

Interchanging therolesofx and y,we see thatEs(x)= x if,andonlyif,Es(y)= Y.

Therefore,the X-orbitsof x and y have the same periods,.
Pick a pointxl e IA. If X(x,) and Y(x,) are colinear,thenthe q-orbitof xl
is thesame as its X-orbit,so it is a circle,and thetheoremis proved. We may,
therefore,assumethatX(x1)and Y(x1)are linearlyindependent.Let y, denote
the(closed)orbitof x1underX. For each x e "1, thevectorsX(x) and Y(x) are
linearlyindependent.The 9-orbitof x1is eithera one to one regularimageof
a cylinderor it is a torus. We excludethe lattercase, because it is what we
want to prove.
Let x be an arbitrarypointof -t1. Givenany neighborhoodU of v1in M,
thereare arbitrarilylarge positivevalues of t such that thepointiCt(x),in the
Y-orbitofx, returnsto U. Thismeansthatthe(0-limitsetY),+(x)of the Y-orbit
of x intersects'71. In orderto see this,let L -(R x 72,+?(x))
be the unionof
all the X-orbitsof pointsin i,+OO(x).The X-orbitof everypointof i,+O(x)has
periodsi because 72+,?(x) c IA. So L = &([0,s1] x iyo(x)), henceL is compact.
Since L is evidentlynon-empty, invariant,and containedin te,we must have
L = A. In particular,x E L. This meansX = Es(y)forsomes e R and some
y e 7?+?(x). Then y = p_(x), so y e ,1 and consequentlyy e -/1n 7+??(x),as we
wantedto show.
Let S be an open 2-dimensional cylindricalband, having-/1 as its equator
and transversalto all Y-orbitsthat intersectit. S will be constructedas a

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
 COMMUTINGVECTOR FIELDS ON S3 75

narrowruledsurfacespannedby geodesicsegmentsnormalto the cylindrical

q-orbitalong -1 (in some riemannianmetricof M). Replacing,if necessary,
S by a smallerband relativelycompactin it, we may assume that thereexists
s > 0 such that the segmentsof Y-orbitIx = {t7(x); It I < e}, of length2e and
originx e S, forma (trivial)fibration, withbase S, of the open set V = U Ix,
e
x S. Denote by7r: V S the projection map that assigns to each 7t(x)e V
its originx e S.
We chooseS so narrowthat the geodesic segmentsthat generate it are
transversalnotonlyto the p-orbitcontainingyi,but also to the P-orbitsof all
pointsx e S. Notice that, since the 9-orbitof everypointof , is dense in fA,
thereare pointsx e S n , arbitrarily closeto -1,so closethattheX-orbitofx lies
entirelyin V, henceit maybe projectedby wintoS. By thepreviousremark,
such projectionis a simpleclosed curve in S, transversalto the generatrices
of S, hence homotopicto '71 (and disjointfromit) in S. From this it follows
that we may narrowS downfurther,in sucha way thatits absoluteboundary
S - S = 8S consistsof 2 of thosecurves,thatis, twocircleseach of whichlies
entirelyin one cyclindricalp-orbitin pt,but not in the 9-orbitcontaining ,1.
Assume,fromnow on, that S has this property.
Definea real-valuedfunctionz: -i, R by lettingr(x) = smallestpositive
-

numberz such that iT(x) e S. Clearlyz is well defined,since for t # t' and
7t(x),i7t,(x)e S one musthave It-t' i > 2e. The precedingremarkshowsthat,
as x varies in 71, 2(x)(x) keeps away from&S. This implies that z-is a continuous
functionon 7,. Indeed, a simpleapplicationof the implicitfunctiontheorem
will show that z-is differentiable.
The map x -' r(x)(x) defines a diffeomorphismof 7, onto a simple closed
curve 72 C S. Unlike '1, 72 may or may not be an orbitof X. In any case 72
cuts each generatrixof S transversallyhenceexactlyonce, so '72 is homotopic
ifwe showthat71n
to.7 in S. Consequently, /2 = 0, therewillbe a ringA
in S, boundedby Y,and 72. This is provednext.
We have assumed,by contradiction, that the p-orbitcontaining'1 is not
a torus, so it is a continuousone to one image of a cylinder. Under these
circumstances,71and 72 mustbe disjoint,because the existenceof a common
pointy e 7, n 72 wouldgive y = 'TJx) forsomex e 71and z- z(x). On theother
hand, y =8(x), s e R. Hence
x = t-8.(x) = 9r(x), r = (-s, z) .

Since z- 0, thevectors(s1,0) and (-s, z-)are linearlyindependent.Theyboth

belongto the isotropygroupof x underthe action9, so this groupis free on
2 generators,and the9-orbitcontaining 71isa torus,contrarytotheassumption.

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
76 ELON L. LIMA

Noticethatwe are assumingactuallythat no pp-orbit in , is a torus,so if x e pe

and t # t', one musthave rt(x) # 08T (x) no matterwhichs e R is chosen.
Considernow the compactcylinderB = {'t(x); x e y1,0 _ t ? z-(x)}. To-
getherwith the ringA c S, boundedby v, and Y2, B formsa topologicaltorus
T = A U B, embeddedin the manifoldM. Now we applyLemma 1. Since M
is simply-connected, H2(M;Z2) = 0, by Poincareduality. Hence T bounds in
M. Lemma 1 then providesus with the following: M - T = C1 U C2 is the
disjointunionof 2 connectedcomponents,each of whichhas T as its point-set
boundary.
Of
Moreover,if s > 0 is the numberintroducedin the definition
0
of S,
and A = A - B (the absoluteinteriorofA), thentheset {'t(x); x e A, 0 < t ? s}
lies entirelyinonecomponent, say C1,whereasthe set {'t(x); x e Ay s < t < O}
is containedin C2.
We need, however,a little more than this, namely: for any y e y1,and
-s < t < O rt(y) e C2. (Similarly, if z e y2 and 0 < t _ e, then Y(t(Z) e C1.) To
provethis,let u x(u), 0 _ u _ 1, be a pathin A, goingfroma pointx(O)e A
to the given y, withouttouchingany otherpointof B besides y. Then ua
mt(x(u))is a pathin M, startingat rt(x(O))e C1and nevertouchingT foru < 1.
So, mt(y) = rt(x(l)) either belongs to T (that is, to B) or to C1. But it is clear
that rt(y) X B, so ht(y)e C2,

Fig. 2

Let y =jy1), so y is a closedorbitof X, entirelycontainedin C2. Choose

a pointw e v. As we have seen earlier, there are arbitrarilylarge positive
values of t forwhichrt(w) returnsto any pre-assignedneighborhood of 'v. In
particular,Yt(w)mustreturnto C2forsomevalues of t strictlygreater than
z-(w) + s. Now, for z-(w) + s < t < z-(w) + 2s, rt(w) e C1. Therefore, rt(w)
mayonlyleave C1and enterC2forvalues t > z(w) + 2s. But how? First of
all, rt(w)cannotcrossA fromC1to C2because all the streamlines of the flow

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
 COMMUTINGVECTOR FIELDS ON S3 77

r,in A lead to C1. Secondly,r7t(w)cannot touch B for values t > r(w) + 2e

becausethiswouldgive 7t(w) =8rt (w) witht' ? r(x) + e, hencet # t'. This,
however,contradictsa previousobservationand it is the finalcontradiction
that provesthe theorem.
3. Fields of 2-frames in 3-manifolds
Let V2(R3)denotetheStiefelmanifoldoforthonormal 2-framesineuclidean
space R3. The map{e1,e2} {e1,e2,e1 x e2},wheree1 x e2is thevectorproduct,
-

definesa diffeomorphism of V2(R3)ontothe set of all positivelyorientedortho-

normalbases of R3, that is, onto the groupS0(3) of all real orthogonal3 x 3
matriceswith determinant+ 1. Therefore,the fundamentalgroupof V2(R3)
has precisely2 elementsand its non-zerohomotopyclass is represented by the
=
closedpatht -. {e1(t),e2(t)},wheree1(t) (-sin 2wt,cos 2wt,0), e2(t)=(0, 0, 1),
0 ? t < 1. (See [4, p. 149].) Now considerV21(R3), the manifoldofall 2-frames
in R3. It is geometrically clear that V2(R3)is a deformationretractof V2(R3),
so these manifoldshave the same fundamentalgroup,and its generatoris still
representedby the same path as before.
(A) Let X c R3. A continuousfieldof 2-frameson X is a continuousmap
f: X - V2'(R3). Denote by D2 the unit disc of the plane z = 0 in R3 and by
S1 = OD2its boundarycircle. The continuousmap f: S1 V2'(R3),given by -

f(x, y, 0) = {e1,e2}, with e1= (-y, x, 0), e2=(0, 0, 1), when consideredas a
closed path, agrees withthe generatorof 71(V2(R3))above described,so it is
nothomotopicto a constantmap,thatis, it cannotbe extendedto a continuous
mapf: D2 - V2'(R3). In otherwords,thereis no continuousfieldof 2-frames
on D2 that agrees withf on S1.
We indicatenextsomeconsequencesof thiselementaryremark. The first
one is a simplificationof Reinhart'sargumentin [6, pp. 186-187].
(B) Considerthe solid torusN3 = N, in R3 generatedby rotatingthedisc
D2 = {(x, y, O)eR3; x2+ y2 1} around,say, theaxis y = 2, z = 0. Let T2=
T =Ndenote the2-torusgeneratedby the rotationof S1 = OD2. The various
positionsoccupiedby S1 duringthe rotationare the meridian circlesof T and
the circles describedby each given point of S' are called parallels. The
meridiansare homotopicto zero in N, the parallelsrepresenta generatorof
7w1(N) = Z. In T itself,parallels cannot be told apart frommeridiansand,
together,theyrepresentthe twofreegeneratorsof r1(T) = Z + Z. Let X and
Y denote orthonormalunit vector fieldson T such that X is tangentto the
meridiansand Y is tangentto theparallelsof T. There is no continuousfield
of 2-frameson N that agrees with {X, Y} on T. In fact, thereis not even a
fieldof 2-frameson D2 that agrees with X and Y on S1.

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
78 ELON L. LIMA

In orderto endowthe sphereS3 with a foliatedstructureof class Co and

dimension2, G. Reeb [5] has introduceda foliationof the solid3-torusN whose
leaves are T = &N and planar leaves fillingN - T. It is notdifficult to define
a continuousfieldof 2-frames{X, Y} on N, such that X and Y, at each point
of N, are tangentto the Reeb foliation. It is not possible,however,to choose
X and Y with the additionalpropertythat [X, Y] _ 0.
Moregenerally,the followingis true:
(C) Let N be a compact3-manifoldwhose boundaryis a 2-dimensional
torusT = ON,such that the homomorphism h: w1(T) r1(N), inducedby the -

inclusionT c N is not a monomorphism.Then thereexists no pair of com-

mutingvectorfieldsX, Y on N whichare linearly independentat each point
and tangentto theboundaryT.
The proofuses the loop theoremof C. Papakyriakopoulos[8], according
to whichthereexists a simpleclosed curve C c T whichis not homotopicto a
constantin T, but is theboundaryof a topologicaldisc D c N. We mayassume
that C and D are differentiably embedded. (This follows,forinstance,from
[2, Th. (6.3), p. 544].) We may also assume that D is transversalto T, hence
D n T = C, sincewe couldhave startedwiththeboundaryofa tubularneighbor-
hoodof T in N, insteadof T. Now assumethatthereexistlinearlyindependent
commutingvector fieldsX, Y on N whichare tangentto T. They definea
differentiable action q' of R2 on N such thateach orbithas dimension2, hence
the boundarytorusT is an orbitof q'. Therefore,if we fixat randoma point
x0e T, the isotropygroupof x0is a discrete(freeabelian) subgroupG, of R2,
with 2 generators. The map of R2 onto T, definedby r cr(x0) induces,by -

passage to the quotient,a diffeomorphism R2/Go T. This diffeomorphism

sends the quotient image of any line parallelto a non-zerovector(a, b) E R2
ontoan orbitof the vectorfieldU = aX + bY in T. Since we mayrepresent
anyelementof the fundamentalgroupof R2/GObytheimage of a straightline
in R2, it followsthenthat by choosinga and b conveniently in U = aX + bY,
we can make all the U-orbitsof pointsin T be homotopicto the given simple
closedcurveC. Oncechosena, b, thereis no difficulty in choosingc, d, so that
the vectorfieldV = cX + d Y is linearlyindependentof U. Clearlythereis a
diffeomorphism of T whichis isotopicto theidentityand carries,say,the U-orbit
of x0onto C. Such diffeomorphism may thenbe extendedto a diffeomorphism
of N, hencethe U-orbitofx0is, just as C, theboundaryof a differentiable 2-disc
E containedin N. A tubularneighborhood of E in N is diffeomorphicto theset
P = {(x,y, z) E R3; X2 + y2 1,-1 < Z < +1}.

carriesE ontoD2 and lets the vectorfieldsU, V definea

This diffeomorphism

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
 COMMUTINGVECTOR FIELDS ON S3 79

fieldof 2-frameson P which,when restrictedto D2, gives the one that was
precludedby the initialremark(A).

4. The main theorem

In the proofof Theorem3, an importanttool is the followingresult:
THEOREM 2 (Haefliger). A differentiable foliation ofcodimensiononeon
a compact,simply-connected manifoldhas at least one leaf thatis notsimply-
connected.
PROOF. This followsreadilyfrom[1, Prop. p. 390]. Indeed, everydiffer-
entiable foliationof codimensionone on a compactmanifoldadmitsa closed
transversal:just considera solutioncurve C of the fieldof line elements
orthogonalto the given foliation. By compactness,the c-limit set of the
solutioncurveC is non-empty, so it containsa pointx. Take a neighborhoodV
of x, triviallyfiberedby arcs oftheorthogonaltrajectoriesto the foliation.The
curveC willcut the neighborhoodV alongcountablemanyintervals. Onemay
easilyconnecttwoconsecutiveintervalsof C n V so as toobtaina differentiable
simpleclosed curve y, transversalto the leaves of the foliation. Since the
manifoldis simply-connected Yis homotopicto a constant. By the mentioned
Propositionin [1], there exists a leaf F of the foliationand a loopin F such
thatthe germof homeomorphism of R corresponding to thatloopvia holonomy
is not the identity. Since the holonomyis invariantunder homotopy(see [1,
p. 378]), such loopis not homotopicto a constantin F, hencethe leaf F is not
simply-connected.
The main theoremmay now be proved.
THEOREM 3. Everycompactdifferentiable 3-manifoldwithfinitefunda-
mentalgroup has rank one.
PROOF. Since everyodd-dimensional manifoldhas rank at least one, the
above statementmeansthat if M is a compactdifferentiable 3-manifoldwith
finitefundamentalgroup,any 2 commutingvectorfieldsX, Y on M must be
linearlydependentat somepoint. TheuniversalcoveringspaceofMis a compact
simply-connected 3-manifold to whichthegiven vectorfieldsmay be lifted,so
we do not lose any generalityby assuming,as we do fromnow on, that M is
simply-connected.
Let thenX, Y be commutingdifferentiable vectorfieldson M. Suppose
(as if it were possible)that X, Y are linearlyindependentat each pointof M.
For each x e M, let D, denotethe vectorsubspaceof Mxspannedby X and Y.
The mapD: x Dx definesa 2-dimensional
- sub-bundleof the tangentbundle.
This tangent sub-bundleis integrablebecause of the relation [X, Y] _ 0

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
80 ELON L. LIMA

(cf. S. Lang. Introductionto Differentiable Manifolds,Ch. VI). The maximal

integralsubmanifoldsof D are 2-dimensionalleaves of a foliationof M. By
Theorem2, thisfoliationhas a non-simply-connected leaf. Nowtheleavesofour
foliationare also theorbitsofthe actionq: R2 x M- Minducedby the commu-
tingvectorfieldsX, Y. Therefore,a non-simply-connected leaf mustbe a (one
to one continuousimage of) cylinderor a torus. Let x0e M be a pointwhose
p-orbitis a cyclinderor a torus. The isotropygroupGoof x0is discreteand
not zero. Let (a, b) e R2 be a free generatorof G0. ReplacingX, Y by X' =
aX + bY, Y' = cX + d Y, so thattheconstantsa, b,c, d satisfyad - bc # 0,
we have X', Y' as linearlyindependent commuting vectorfieldson M, suchthat
the X'-orbitof x0is closed. Change notation,going back to X, Y, instead of
X', Y'.
By Theorem1, thereis a pp-orbit
in M whichis diffeomorphic
to a 2-torus
T. Let M1and M2be the two 3-manifoldswith boundary T such that MN
M1U M2 and M1n M2 = T. At least one of the inclusionhomomorphisms
r1(T) - rw(Ml),w1r(T) ) w1(M2)has a non-zero kernel for otherwise, by Van
Kampen's theorem[7, p. 177], w1(IM)would be the freeproductof w1(Ml)and
w1(M2), with amalgamationof thesubgroup(isomorphic to) r1(T). Thenw1(M)
would containa subgroupisomorphicto r1(T) (Schreier's theorem)which is
contradictory,since r1(M) = 0. Let N be one of the manifoldsM1,M, suchthat
the injectionr1(T) w1r(N)
- has a non-zerokernel. We are in the situationof
? 3, paragraph(C), hencethe commutingvectorfieldsX, Y must be linearly
dependentsomewherein N. This concludesthe proofof the theorem.
REMARK.It may be that any simply-connected compact 3-manifoldis
horneomorphic with thesphereS3. If thisis thecase (or if one wishesto prove
the abovetheoremonlyformanifoldsthatare coveredby S3), then,insteadof
the Loop Theorem,onymay use the theoremof Alexander(Proc. Nat. Acad.
Sci. (1924),pp.6-8) accordingtowhichany2-torusTpiecewiselinearlyembedded
in S3 is such that at least one of the components of S3 - T is a solid torus.

COLUMBIA UNIVERSITY AND

I. M. P. A., RIO DE JANEIRO

BIBLIOGRAPHY

1. A. HAEFLIGER, Varietes feuilletees, Annali Scuola Normale Sup. di Pisa, Serie III, vol. XVI,
fasc. IV (1962), 367-397.
2. J. MUNKRES, Obstructionsto the smoothingof piecewise-differentiable
homeomorphisms.
Ann. of Math., 72 (1960), 521-554.
3. R. S. PALAIS, A global formulation of the Lie theory of transformationgroups, Memoirs of
the AMS, no. 22, 1957.
4. L. S. PONTRJAGIN, Topologische Gruppen, Teil 2, Leipzig, 1958.

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions
 COMMUTINGVECTOR FIELDS ON SI 81

5. G. REEB, Sur certainesproprietestopologiquesdes varietesfeuilletes,Paris, 1952.

6. B. REINHART, Periodic orbitson two manifolds,Bol. Soc. Mat. Mex., 5 (1960), 184-187.
7. H. SEIFERT-W. THRELFALL, Lehrbuchder Topologie, Leipzig und Berlin, 1934.
8. J. STALLINGS, On the loop theorem,Ann. of Math., 72 (1960), 12-19.

(Received December16, 1963)

This content downloaded from 66.180.162.103 on Tue, 30 Sep 2014 16:12:55 PM

All use subject to JSTOR Terms and Conditions