Annals of Mathematics

Classification of the Mappings of a 3-Complex Into a Simply Connected Space

Author(s): Hassler Whitney
Source: Annals of Mathematics, Second Series, Vol. 50, No. 2 (Apr., 1949), pp. 270-284
Published by: Annals of Mathematics
Stable URL: [Link] .
Accessed: 18/11/2014 00:26

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
[Link]

.
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@[Link].

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

[Link]

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
ANNALS OF MATHEMATICS
Vol. 60, No. 2, April,1949

CLASSIFICATIONOF THE MAPPINGS OF A 3-COMPLEX INTO

A SIMPLY CONNECTED SPACE
HASSLER WHITNEY
(ReceivedMarch 29, 1948)
1. Introduction
Let R be an arcwiseconnected and simplyconnected Hausdorff space,whose
secondhomotopy groupT2 = w2(R)has a finiteset of generators.(This is
alwaysthe case if R is a finitecomplex,since72 is thenisomorphic withthe
secondhomology groupof R, by a well knownTheoremof Hurewicz.) We
shallclassifythemappings ofa 3-complex K intoR. We takeK as a simplicial
complex, withordered vertices. For therelationofthispaperto theliterature,
and references in brackets,see thefollowing paper.
It wouldbe naturalto deducethe classification theorem fromtheextension
theorem of thefollowing paper. We shallnot do thishowever;forthe proof
oftheclassification theorem is simpler, andfurthermore, it is notevidenthowto
obtaintherelatively simpleformula ofthepresent paperfromthemorecompli-
catedoneoftheextension theorem.
The case that7r2has no elements ofevenorderis especially simple;theclassi-
ficationforthiscase is obtainedin ?8. The restofthepaperis devotedto the
specialproblemarisingwhenelements of evenorderare present;thestudyof
"junctions"in a preceding paper' is thenneeded. It will be notedthat in
applyingtheformulas of thepapermentioned, whichwe referto as RH, there
are no "crossings oftubes"whenprojectedintothegivenplane;thusa simpli-
ficationofRH couldbe usedin thepresent paper(butnotin thefollowing one).
The results of??2,3 and4 arelargely wellknown;theyarefundamental to the
discussionsofthepresent paperand thefollowing one.
2. Preliminaries
Let al, * ,a, be independentgeneratorsof 722. Let a,, be of order n,,
ifn, = 0). Then ti
and generatethecyclicsubgroupr2,of T2 (7r2 is infinite
is the directsum
(2.1) T=T
2 7 i .*.. ED 2,

to this decomposition,
Corresponding X may be written
any r-r2-cochain
as a sumofr-Tru-cochains:
(2.2) X = XI + *-+ Xs,
Let CT(K, G), t(K, G), B?(K, G), If(K, G) denote the groupsof r-G-
and r-G-cohomology
r-G-coboundaries,
cochains,r-G-cocycles, classes respec-
1H. WHITNE4Y, Relationsbetween groupsofa simplyconnected
thesecondand thirdhomotopy
space,theseAnnals,50 (1949),180-202.
270

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
 MAPPINGS OF A 3-COMPLEX 271

tively. Let G be a finitegroup, expressedas the image of the group of in-

tegersunder a homomorphismk; then G is isomorphicto the integersmod n
forsome n. (We shall use G = 7r2, , n = n,,, r = 2). There is a well known
operation fromI'(K, G) to Hr+'(K, J) (J = integers),definedas follows.
(This will not be used until ?11; forthe definitionwith d in place of 6, see for
instanceAlexandroff-Hopf, TopologieI, Berlin,1935,pp. 222-223.)
Given an r-G-cocycle X, let w,.X= X be the corresponding integralcochain,
witheach coefficient g (mod n) replacedby a, where?(a) = g and 0 ;5 a < n.
(We could omit these inequalitiesif desired.) Since 3X = 0, the coefficients
of AXare multiplesof n. Set
(2.3) aOn'(X) = (1 /n) bcnX = (1/n) SX.
The cohomologyclass of O"+'X depends on that of X only. (We do not need
this fact.)
If E' is orientedEuclidean 3-space, and j,3 is E3 plus the point at infinity,
the elementsof 7r3are representedby the mappingsf of V3 into R, with the
point at infinitygoing into the fixedpoint P of R. Note that, aftera small
deformation, we may assume that only a bounded part of E3 goes into R - p.
With a fixedcoordinatesystemxi , X2 X3 in E3, we shall let Et denote the plane
x3= t, and E+ and ET, the half spaces X3 > t, X3 _ t respectively. Addinga
bar to any of thesesymbolsdenotesadding the point at infinity.
Let ao X I4 denoteorientedr-cellshomeomorphic in a definitefashion.
Identifycorresponding pointsof theirboundaries. Then
Sr r r r
S 12 r r (or r r
D01 = al1- 0,0 } = 72 - al1 0 =
02 0'2 - 0o0

are orientedr-spheres. Let fi be a mapping of a, into R such that all three

mappingsagree on the boundaryof the cells. Then mappingsof the Sk are
determined,whichdetermineelementshikof 7r7, and
(2.4) h01+ h,2= ho,
a well knownlemmain homotopytheory.
The CartesianproductI X K of the unit intervalI and K is subdividedinto
cells I X oa , orientedso that
0(1 X a) = 1 X a - 0 X a - I X do.
A deformation ft(O < t < 1) offointof, (both definedin K) is equivalent to a
mappingF of I X K.
Let -r be a convex Euclidean cell, and let rm1be a boundarycell; let T' be
07r minusthe interiorof tr1-. If f and g are mappingsof Tr into a space which
agree in r', then they are homotopic,keeping the points of r' fixed. (This
follows from the fact that T' is retract of r'.) In particular, if f is defined in
Ir and f(p) = p in tr l thenf is homotopicto f' withf'(p) = 3 in rr.
Applyingthesefactsto I X K showsthat ifa mappingfo is definedin K, and
ft (O < t < 1) is definedover a (closed) subcomplexof K, ft may be extended
over the restof K.

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
272 HASSLER WHITNEY

to certainmappings
3. Cocyclescorresponding
Let L be a complex,formedof Euclidean cells. Let fo be a mapping of L
into R such that fo(LT-') = p. The correspondingr-7r,-cocycle WfOin L is
definedas follows. For each oV,fo(aou) = P; hencefodefinesan elementWfO;(c)
of 7r,in or. That 8Wf, = 0 followsfrom the fact that fo is definedover all
(r + 1)-cells of L. Any Wr is obtainable from a mapping of L', which is
extendablethroughLr+l if aJWr = 0.
Let ftbe a deformationof the above fosuch thatft(Lr-2) f(L') =p.
Let this defineF in I X L. Then foreach o-lX F(a(I X a = P; hence F
maps I XTr-1intoR, defining an elementXr-1(r-l) of7rr* By definition of WF,
(3.1) WF.(I X A) = -A
forall chains AT'. Since F in 0 X L and 1 X L definefoand fi respectively,
and AWr = 0, we have, foreach _r,
0 = aWr . (I X Or) = W;.o(I X air)
= Wfl. Or _- Wfr. Or _ Xr-1. ar;

consequently
(3.2) Wl- W =AX=

We may expressthis by sayingthat the changein the cocyclecorresponding to

the mappingis the coboundaryof the deformationcochain.
Conversely,if F is definedin I X Kr except interiorto the cells I X o, ) if
F in 0 X K and 1 X K definesfoand fi in K respectively, ifF(a(I X Hi')) 0,
and if Xr-1 exists satisfying(3.2), then F is extendablethroughI X Kr, i.e.,
fois homotopicto f, in Kr. For F = 0.
Let f and g be mappingsof L intoR whichagree on Lr1. We cannotat once
define Wf and We; but we can define the difference cochain A', an r-7rr-
cochain,as follows. For each or, choose a rT ; identifyart with aoct ; mapping
at accordingto g and r! accordingto f gives a mapping h of St - 7X Ar -

definingArg(o_) e 7rr. (This cochain appears both in Eilenberg [2] and in

Pontrjagin[7].) Clearly 'Ar' + Ar = A;,h(see ?2), and
(3.3) Af,g= - Wj, if the latter are defined.

Let ftbe definedso thatf = foin LTand ft(Lr-1)= ; we definethecorrespond-

ing deformationcochainAf , an r-rr-+-cochain, as follows(compare Pontrjagin,
[7]). Take any at~; let F, in I X Si, correspond to ft in 4rt. Define G in I XS,
correspondingto gt in ot ,with gt = fo. Then G = F on (I X or)r = 0 X or U 1 X
or u I X adfr, so that we may set
(3.4) Art(O-)= ra+,t(IX St).
Let ftbe a deformationof L such that fi = fo in Lr and ft(LT1)= 3. Then
(3.5) i =+fI =

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
 MAPPINGS OF A 3-COMPLEX 273

this is a generalizationof (3.2). To prove this, take any oria; let ft in osr+'
defineF in I X ari'. Let gt = foin c$a'; let gt defineG in I X 0a$r', and let
G = F in O X i+l u 1 X ai +. Since gt is constant,G definesthe same element
of 7rr+as woulda mappingG' ofthe set obtainedfroma (I X 4+') by contracting
each I X p (p e ca{+l)to a point,and usingG = F in the two faces left. Since
G' definesA"+fl/-+1), so does G in 0(I X r+t).
Say 0aCR = E eio; then 0(1 X 4+') = -E ei(I X ') +- * . Choose
'4, and change G to F in I X ar ; using (2.4) (applied to two mappings on
I X 'rand one on 0(I X ao+") - I X A'), this alters the elementof 7r7+1by
-eiA.+%(I X a') = -ejA't(o4). ChangingG to F in each face in turn shows
that the total alterationis
x;fjA e(a)
E = ta = -at(r+l)

Thus F, in 0(1 X or'), definesA;+f)Qt1)

-r6Af4a'). But F is definedin
I X as'; hence this elementis 0, giving(3.5).
4. The 2-cocycleof a mapping
Let foand fi be homotopicmappingsofK (or ofK2) into R, such thatf6(Kl)=
f1(K') = p. Then, as is well known,W2, _ W2. For let the deformation
ft offo intofi defineF in L = I X K. Since R is simplyconnected,we may
deformF overthe 1-cellsI X c? so that it maps themintop5;extendthe deforma-
tion over L (keepingF fixedin 0 X K and 1 X K), obtainingF'. This gives a
'
newhomotopyf offointofi withf'(K0) = p. The resultnowfollowsfrom(3.2).
Conversely, iffo(K1) = f1(Kl) = f and WfI- Wfo, then fo and fi are homo-
topicoverK2; see the conversestatementfollowing(3.2). The2-cohomology class
WI
W2 corresponding tof is a homotopy
invariant.
Choose a fixedcocycle W2 fromeach cohomologyclass W2; then each f is
homotopicto somef' with Wf,the chosencocycleof Wf.
5. Standard mappings
Let aobe a fixedoriented2-cell,in the shape of a circulardisc. A fixedpoint
of 0o will be called its top. For each A = 1, * , v,let 4, be a fixedmappingof
aOinto R with ,(2-0 =-p, definingthe elementatsof 72
Given any 2-simplexa2 with orderedverticespo, P', P2, and any element
a of 7r2, we definea corresponding standardmappingof a2 into R, as follows.
Write
(5.1) a = alai +** + aa, O < a. < no if n, > O;
thisrepresentation of a is unique. Let pi and p2 be the mid pointsof pop,and
POP2 respectively. Let Q be a rectangleinteriorto 02, bisectedby the segment
PlP2 Cut Q into rectanglesQi, * , Q^, each bisected by plp2 , ordered
in the directionfromp* to p2 . In each QA, choose k, = Ja, I disjoint discs
Q *, Q,,, each bisectedby plP2 . The top of Qmiis the point of the disc
.

furthestfromp1p2 .

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
274 HASSLER WHITNEY

Take first,for a given p, the case that n. > 0; then k, = a,. There is a
unique linearmappingof uo into each Qi whichpreservesorientationand sends
the top of so into the top of Qi. Using4)., this gives a mappingof Q,siinto R,
defininga,. Map Q. - Ei Q,.iinto b. The resultingmappingof Q, intoR de-
finesa,,a,..
Suppose now that no = 0. If a,, 0, we proceedas before. Suppose a, < 0.
Then thereis a unique linearmappingof 0 into each Qi whichreversesorienta-
tion and sends the top of u' into the top of Qgi. Using X,,gives a mappingof
Qi defining- a. Hence the mapping in Q, defines- ka = a,,a.
Map a2 _ E Q,ai into p. This definesa standardmappingf of a2 into R
whichdefines? auau = a.
A mappingf of K into R is standardif it is standardin each 2-simplexof K.
It is completelystandardif,further,the corresponding cocycle WJis the chosen
cocycleof WJ. Using ?4, we see that any mappingis homotopicto a completely
standardone.
6. A particulardeformation
Let fobe a standardmappingof K into R, and let X1 be a 1-7r2-cocycle in K.
Corresponding X1to we shall a
define deformation of
ft fointof' such thatfi = fo
in K2, and Af0,f bears a certainrelationto X1.
To defineft. we defineF in I X K. First,set F(I X K0) = p. Next, take
any r2= I X al, O* = pop,. Set p! = 1/2 X pi; definerectanglesQ,, and
discs Q,.iin rs exactlyas in ?5. We may then definea standardmappingF of
-r into R, definingthe elementX'(ar), as in ?5.
We remarkforfuturereferencethat if a mappingF' of I X K' is given such
thatF'(a(I X oa)) = 0 foreach i, thenlettingX'(al) be the elementof72 defined
by F' in I X al, and definingF in I X K' as described,we may deformF' into
F, keeping F' = F fixed in each a(I X al).
We now choosea standardextension ofF througheach T3 = I X cr2,
22 = POPIP2 .
Choose a small circulardisc C in c_2,near the center of the edge P1P2. Set
B = I X C. Take first sl = 1. To each disc Qi in I X pop,we attach a tube;
let it go into r3,then run down close to the face I X pop,towardsI X pi, then
run close to the face I X P1P2 towardsB, and run into B; keep it always at the
heightt = 1/2,and keep the top filamenton top. (See the figurein ?7.)
Run tubes fromthe discs (with jA= 1) in the face I X POP2into B in the
same manner,moving close by I X P2. Run short tubes fromthe discs in
I X Plp2 intoB. We may do thisso that,ifprojectedinto0 X _2, no tubescross.
We must now join the tubes in B. If firstno = 0 (,u = 1), the algebraic
numberof tubes (with an obvious definition)enteringB is 0, since 5X' = 0
and henceSX' = 0 (all pu); we may therefore insertshorttubes in B, joiningthe
above tubes in pairs. We do thisso that top filamentsare on top, and no tubes
crosswhenprojectedintoO X c_2. Let F map each tube intoR accordingto the
mapping4l defininga, ; thenF is continuousso far.
Suppose now that no > 0 (jA= 1). If the algebraicnumberof tubes entering

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
 MAPPINGS OF A 3-COMPLEX 275

B is 0, join the tubes as before. Otherwise,this numberis in, (,u- 1), as is

easily seen (see ?11). We then join any pairs of opposite sign, and inserta
junction to which the rest are attached; we may clearlydo this so that, when
projectedinto 0 X Cr2,no tubes cross,and so that top filamentsand the top of
the junction are on top. The exact mannerof insertingthe junction will be
givenin ?10.
We carryout the same process with ,u = 2, with the followingchange. As
soon as each tube has lefta face I X pipi, it is carriedto a slightlyhigherlevel
t > 1/2, so that it will intersectno tubes with ,u = 1. Insert the tubes with
= 3, bringingthemto a higherlevel yet,etc.
If Q is the rectanglein a2 as definedin ?5, then,sincefo = f, in K2, we may
defineF in I X Q so thatft = foin Q. SettingF(p) = p in the restof I X 2
makes F continuous.
So far, F is definedin 0 X K u I X K2. Extend F arbitrarilythrough
I X K (see ?2).
7. The corresponding special case
A%0,f3,
In the special case that I2 has no elementsof even order,we now determine
Aa for the mappings consideredabove. Take any u3 = poplp2p3; choose
rT3and identifyar3 with Oa'; map a3 accordingto fi and r" accordingto fo ; the
resultingmappingf* of S3 = a3- T determines = A A~ ,ll(f3).
Definethe oriented3-sphere
S' = 0 X a3 + I X au3- 1 X a = -e3(I X f)
Using the mappingF of ?6, definethe mappingF' of S' into R by setting
F' =F in OX3 ul X ao3, P(l X p) = F(O X p);

we shall show that this determinest.

First,fi maps a3 the way F maps 1 X a. Since F is definedthroughout
I X a3, we may replace F in 1 X r3by F = F'in0 X ac3u I X Oa3. Next,
foin r3 is the same as foin a-,i.e., the same as F' in 1 X a3. Hence f* in S is
equivalentto F' in S'.
Let us representt X Oa3by the subset Et of R3, 0 < t < 1, lettingthe pointat
infinitycorrespondto a point p' in PoP2P3nearpo, for which F(t X p') =
0 ? t < 1. We may now considerF in I X ao3 as a mappingofEo,1 = E+nEn .
The tube systems T1, ***, Tl, which we have insertedin each I X ac, for
ar C Oa3, are now tube systemsin Eo, , approximatelyat the level X3 = 1/2,
and withtop filamentson top.
Suppose the tube systemTo is pulled away fromthe othertubes as described
below. We then lose an amount A,t say fromt. The tube systemT, taken
by itself,definesan elementg of 7r3. Then if i' is the mappingdefinedby F',
but with T1 , **-, T' omitted,

t = v' + E At + E .

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
276 HASSLER WHITNEY

If T1, * T, are omitted,then F(t X p) = F(O X p), O < t < I, p e do3.

Now I' is definedby F' in S', but using this new F. With the new F, we may
omit the interiorof I X do3and identify0 X au3 and 1 X a^3. The new F' is
now a mappingof the cells 1 X a' and 0 X a3, whose boundariesare identified.
But in each of these,F' is the same as F in 0 X a3; hence t' = 0, and
(7.1) l
At01(o 3) = Z 'At + E

Firstwe determinethe Aut Each Tx lies whollybelow T, , if X < ju. Hence

+ 1).
we may bringthe tube systemsto separatelevels, To at the level t = JAI/(v
Keeping To at this level, pull it towardsthe 2-cell I X P2P3. In this process,

FIG. 1

the tubes throughI X PoPi must cut throughthe vertical tubes in I X Q,

Q C P1P2P3, but no othercuttingsare necessary. (In the figure, Q appears
as a shaded rectanglein PlP2P3.) We may now shrinkT, into a small region
separatedby a spherefromany othertube system. Hence zXyis definedby the
above cuttings. Since T;, cuts throughno Tx, the order in which the To,are
pulled away is immaterial.
Each tube of T, throughI X pop,has a cross sectiondefininga, ; the set of
thesetubesdefinesXl(pop1). The set of tubesin I X Q definesWf0p p2p3).
Pullingeach tube ofthe firstset in turnthroughall the tubes of the secondset, a
totalchangeofXp1(pop1)WfO(plp2p3) is causedin t (see RH, ?6). Hence
(7.2) Act = EXp1(pOp1)WfO(p1p2p3)
= d(X# v, Wf~o)(a')2

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
 MAPPINGS OF A 3-COMPLEX 277

using the usual formulafor the , product. The correctsign is determined

in (12.1); we do not need it here.
The tube systemT, in its new positiondefines ,. If thereare no junctions,
RH Theorem 13A shows that g = 0. If there are, then n, is odd, and RH
Theorem19A gives the same result. Therefore(7.1) and (7.2) give
(7.3) AB +-l= ,X W2 in the special case.
8. The classificationtheorem,special case
We give firsta condition for the homotopy of two completelystandard
mappings,thenan enumerationof the homotopyclasses.
THEOREM 8A. Let 7r2 have no elementsof even order. Then two completely
standardmappingsf, g of K into R are homotopic if and onlyif
(a) W2 = W2
(b) Thereis a 1-7r2-cocycle
X1 and a 2-7r3-cochainy2 suchthat
(8.1) Af = X1, Wf2+ y2.

Suppose firstthat (a) and (b) hold; we mustshow thatf and g are homotopic.
ftoff = fo intofi,
Because of (a), Af gis defined. We firstdefinea deformation
by the constructionof ?6, using X' or -X1 so that (7.3) becomes
A Bfl =X1 tWf2

Next definea deformation f offo = fi intofA, as follows. UsingF' in I X K,

set F'(I X K1) = 1. Define F' in each I X a2 so that fl = fo there, and
A2'(as) = Y(or2) (see ?3). Extend F' arbitrarilyover the restof I X K. Then
(3.5) gives
Alf"f;= 6A2 = aY.

Therefore
A1,11= ,+ =
W + e Y2 = Aa

and it followsthat A;,- = 0.

Nowf' and g agreeon each Oof8, and A",g(af) = 0; thereforefl may be deformed
into g in 3i (keepingit fixedin M). Thus fl and g, and hence f and g, are
homotopic.
Suppose nextthatf and g are homotopic. Let ftbe the deformation, usingthe
intervalI': 0 < t _ 2, and let F be the correspondingmapping of I' X K.
Let I" be the interval1 _ t ? 2. We shall carryout certaindeformationson
F, keeping it fixed at t = 0 and t = 2.
First, deformF so that F(I' X K0) -p. Next, deformthe new F so that
F(I" X K') = p. We now deform it so that it is standardin I X K1 (see ?6).
Finally,deformthe new F in each I X a2 , keepingit fixedin 0 X a2 U I X af,
it
so that is the standard extension definedin ?6. (For the existenceof the
as
firsttwo deformations, see the end of ?2.)

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
278 HASSLER WHITNEY

Since f and g are homotopic,and f = g in K', W 2 Wi (see ?4). Since f

and g are completelystandard,Wg2= Wf, proving(a).
Because of the last deformationof F, if we let jX' denote the deformation
cocycle correspondingto F over the I X al, using the sign which occurs in
(7.3), (7.3) gives
A3 f = X 1 Wf2.

Since -W2 = X' = O, W2 = WJ= Wf2,,andg =f2 =fi in K2. Also

fj = f,(1 _ t ? 2) in K'. Hence y2 = A2(t E I") is defined. Now (3.5) gives
= Y
'iaf,

These two relationsgive (8.1), proving(b).

THEOREM 8B. Let 7r2 haveno elements of evenorder. For each 2-7r2-cohomol-
ogyclass W2,let WT2be thechosencocylein it, and letT(T2) be theset of all 3-73-
cocycles
XI e
(8.2) /(X' W2) = XI "y W2, Z'(K, 7r2).
The set ( of homotopy classes of K intoR falls intodisjointsubsets2(W2). The
classes in S(W2) are in one-onecorrespondence with theelementsof thequotient
group
(8.3) Z'(K, 7r3) mod [B'(K, 7r3) + T(W2)].
Of courseZ3 = C3,since dim(K) = 3. If W2 = 0, (8.3) is simplyH3(K, 7r3);
in thiscase thecorrespondence may be set up in a naturalfashion,but apparently
in
not soi the generalcase.
The partitionof 25 into subsets T(W2)was noted in ?4. Let us show that
each 2(W2) is non void. Definef as follows. Set f(K ) = j5. For each a.,
choose the standard mappingf of o2 into R definingW2(u,~). Since 6W2 = 0,
we may extendf throughK = K3. Then the homotopyclass off is in S(W2).
In each (2(W2), choose a fixedhomotopyclass ho = h(W2),and let fo = f(W2)
be a completelystandardmappingin this class. We shall set up a correspond-
enceu betweenhomotopyclassesin 25(W2)and elementsof (8.3). Set u(h(W2)) =
0. Now take any otherhomotopyclass h' e ( TV2); let f' be a corresponding
completelystandardmapping. Since Wfw= W2 = Wfo, f' = foin K2. Hence
we may define
Afojf e Z3(K, 73),

thus determining an elementu(h') of (8.3). Suppose we had chosenf" insteadof

3 3
f'. Thenf' and f" are homotopic;hence,by Theorem8A, A,, = -
lies in B3 + 1, and u(h') is independentof the choice off'.
If h' and h" are distincthomotopyclassesin S(W2), andf',f" are corresponding
completelystandardmappings,then by Theorem8A, Af p,p is not in B3 + a;
hence u is one-one into.
To show that u is onto, take any 3-7r3-cocycleX*. Set f = fo in K2. In

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
 MAPPINGS OF A 3-COMPLEX 279

eachac, definef so that Af,()do X*(o3). Then ifh is thehomotopy

classof
f,u(h) X* mod (B3 + I). This completesthe proof.
9. The case R = S
S2, then7r2 is infinite
If R is a 2-sphere cyclic. By RH, ?9,we maydefinea
new pairinga . 0 of 7r2withitselfinto 7r3such that 2(a . A) = acd. Let X c Y
denoteproducts,usingthispairing;then
2(X ,Y) =X ,Y.
Making this substitutionin Theorems 8A and 8B, we find the Theorem of
Pontrjagin,[7]; see also Steenrod,[10].
10. Insertionof the junctions
In ?6, we were not explicitabout how junctions were to be inserted. For
any ,Awithno > 0 and even, we must make the insertionmoreprecise.
Take any 2 pOP1P2. Let the point p start at po, and move around &a2,
goingto pi, thento P2, thenback to po. The segmentI X p then moves
around I X &S2, crossing all discs (withthisA) to whichtubeswereattached.
The firstsuch disc (which is in I X pop, if there are any discs there) will be
named even forG2; the discs (withthis A) are named even and odd alternately.
The tubes enteringthe cylinder B in I X a2 are named accordingto the discs
theyend on, and if thereis a junctionJ in B, the tubes endingat J are named
even and odd correspondingly (compare RH, ?20). We insert J so that its
even discs attach to even tubes. (Note that the boundaryof B correspondsto
the curve Ci in RH, ?20.)
In thefigurein ?7, let each pair ofcurvesfroman edge pipito a circlerepresent
any numberof tubes fromI X pip to the corresponding cylinderB. The first
tube, attached to the firstdisc, named even, is marked by a heavy dot.

11. Whatjunctionsare possible

Givenany 1-7r2-cocycle
X1' = X1, recallthatX,1is the integralcochain
ai satisfy0 ? a, < no (we are taking
to X1, whosecoefficients
corresponding
no > 0 even).
Take any 2 = POpP2. Since3X1 = 0,
(11.1) Xl(pop1) + Xl(PlP2) - Xl(pop2) 0 (mod no).

Call the threetermsa2, a0, a, respectively. Since 0 ? a2 + ao < 2 n, , we see

thatthe lefthand side of (11.1) equals0 orn, accordingas a2 + a0 < noor Z
no ; this is the conditionthat 02 X1(&_2)be 0 or 1 respectively.
Let the mappingF correspondto W2 and X'. Then for any a2 and ju, the
algebraicnumberof tubes enteringI X oc_2equals the lefthand side of (11.1).
Hence thenumberofjunctionsin I X c2equals On.X (o2), and is 0 or 1 accordingas
X, (pop1) + X. (plp2) is < noor 2 no.

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
280 HASSLER WHITNEY

We now considerorientations. Since

d(1 X _2) = -I X 2+ =- I Xpop, +*,
the discs in I X PoP,and I X PlP2 are orientedoppositelyto 0(I X a2); hence
tubes in I X 02 attached to these discs are orientedinto I X a0, towards the
cylinderB. Thereforeif thereis a junctionin I X a2, it is negativerelativeto the
naturalorientationofI X 0.2.
12. The corresponding
A'O, general case
Given a3= PoP1P2P3, we determinet = AOf,(a3) by (7.1). As to Abe,it was
found before,except for the proper sign. To determinethe sign, we shall
determinesome orientations.
Since
I X a3 =IX pOplP2 + *(-IX pOplp2)=I X POP1
+ ,
I X poP,lies positivelyon the boundaryof the part of S' in I X PoP1P2. There-
forepositivetubes T, cuttingI X PoP1go out of I X PoP1P2, into I X pOpipS.
(See the arrowsin the figure.)
The verticaltubes T* in I X P1P2P3 have cross sectionswhose boundariesare
orientedin the directionfrompi to P2 to p3to pI. If the tubes T' are continued
around the tubes T*, they would go in the opposite direction. Therefore
(see RH, ??2, 6)
=
-X,(pOpI)Wf20(pp2p3),

(12.1) E = -X1(pop)Wf2(plp2p3).
As before,g = 0 unless n. > 0 is even. Suppose this is the case. Let 0.2
be thefaceofu3opposite pi. Since S' = I X aa3+ *=(-)l Xi+ **
ifthereis a junctionJi in I X 0.i, it musthave the sign - (- iY; see the end of
?11. Hence the possiblejunctionsin I X 003 are:

Jo : -, J1: -, J2:-, J3: +.

If thereare no junctions,of course = 0. If thereare, then thereare the

same numberof negativeas of positivejunctions.
Let Bi be the verticalcylinderin I X ai. in whichthe tubes are joined,contain-
ing Ji if it exists. Then each tube througha face I X 0.r runsfromone Bi to
anotherBk, corresponding to cells of I X K with I X al as a face. The tubes
enteringa givenB2 are named even and odd alternately,as in ?10.
Let Case (23) denotethe case that just the junctionsJ2and J3are present,etc.
CASE (23). Since J2 is present,X,'(pop,) > 0 (see ?11), so that there are
tubes throughI X pop,, which run fromB3 to B2. Let T be the firstone.
Since pop,is the firstedge of both 02 and 0_3, T is the firsttube enteringboth
B2 and B3, and hence is named even at both ends.

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
 MAPPINGS OF A 3-COMPLEX 281

We now apply RH Theorem20A; the formula(19.2) reads in the presentcase

(12.2) , = (Z, *D)nor(oe).
Here, the Bi whichactually containjunctions,i.e., B2 and B3, are thoughtof as
verticesof a linear graph G*, and the parts of the tubes outside B2 and B3 are
thoughtof as 1-cellsof G*. Each vertexBi is orientedaccordingto the sign of
JL. By D,, is meant any 1-chainmod 2 whose boundary is the sum of the
junctions;we may let D, correspondto the tube T. Here and in the later cases,
the tubes employedwill always join cylindersBi, B, of opposite signs; hence
ZA".D, = 0 (see RH, (20.1)). Since thereare no crossings,whenG* is projected
into 0 X Ou3(see the figurein ?7), Z, = 0; hence we need consideronly Z7.'
Now Z., *D, countsthenumberof 1-cells(mod 2) ofDMwhoseendsare ofopposite
parity. Since T is even at both ends,Zlf-D, = 0 in this case, and g = 0.
CASE (12). First suppose XA1(pop3)$ 0. Let T be the firsttube through
I X pop ; it runsfromB. to B1. In the given orderingof tubes enteringB2,
T is the last tube; hence, cyclically,the next tube is the first,named even;
hence T is named odd at B2. For the same reason, T is odd at B1. Hence,
again using T forD,, we findg = 0.
Now suppose X,(Pop3) = 0; then thereare no correspondingtubes through
I X W then inserta closed tube T* with cross section definingaM,
We
whichrunsfromB2 throughI X popiinto B1, then out again, throughI X pops
nearerI X p3than before,into B2 again and joiningits startingpoint. We do
thisso that whenprojectedinto0 X aOo3,thereis no crossing,and so that the top
filamentis on top. Let T be the part of T* runningfromB2 to B1, and let
T' be the part runningback.
If we name the ends of T odd and the ends of T' even,the ends oftubesaround
B2 and B3 alternatein parity. Hence again we may apply RH Theorem 20A,
with T forD. ; again -0.
CASE (03). Since Jois present,Xl(plp2) # 0. Hence thereis a tube through
I X plp2; call the firstone T. At Bo, T is the firsttube, and hence is named
even. At B3, all tubes throughI X pop,come beforeT; hence T is named even
or odd accordingas the numberof these,whichis XA(pop1)(see ?11), is even or
odd. Thus
Z,,,*D,= 'X,(pop,) (mod2).
Puttingin (12.2) gives
(12.3) =

CASE (01). There is a tube throughI X P2P3in thiscase; call the firstone T.
At Bo, the end of T is named even or odd accordingto whetherX1,(pIp2)is even
or odd; at B1, accordingto whetherX1(pOp2)is even or odd. Hence the ends of
T are named in the same or oppositemanneraccordingto whether(using (11.1))
X!(plp2) + X!(pOp2)_X1(pOp0) 0 (mod 2)
or not. Hence the formulafor , is the same as in Case (03).

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
282 HASSLER WHITNEY

CASE (0123). An example of this case, with n, = 4, is given by

(X0o= X'(pop,) etc.).
Xo, = 3, X0 = 2, X03 = 1, X12 = 3, X13 = 2, X23 = 3.
Therearetubesthrough tubesofeach
bothI X pop,andI X P2P3 ; callthefirst
kindT and T' [Link] letD, correspond the
to T + T'. As before,
endsofT arebothnamedeven,andtheendsofT' arenamedthesameoropposite
accordingto whetherXl(pop,)is even or odd. Consequently the formulafor
g is thesameas in cases (03) and (01).
We maysummarize the cases as follows. If Jois absent,g = 0. If Jois
present,t is givenby (12.3). Hence,referring to ?11,
(12.4) 0= , Xp
in all cases.
Puttingthisrelationand (12.1) into(7.1) gives
(12.5) AX = W0 + XXn>[Link](X 0X)nlv(ai).
Notethatthelast termmaybe written
(12.6) Enp>O, even (X;', 5X),1(a).
We shallfindanotherway of writting this. WriteXl in the form
(12.7) Xl = E(aia,,)a!4, 0 a i < n,.
ThenXI, = E aial, and if [a4 o2]denotesincidence
numbers,
1 = hi b bi = Ei [o-: o
We defineO*by
(12.8) 0* X I = E, (bi/no)ar,,ao
notethat
(12.9) On = (0nX )a.
Now
X1 0 X1 = E (aja,,)a! Ej (b1/n,,)a4,,a
= (L aivc4<, >2 b/n)fy~y^
(12.10) on =, (X n
Thus (12.5) maybe written
(12.11) 4eoii = - i WX + Enp>O, evenn,/2(X, v ,X,

theorem,
13. The classification generalcase
The homotopyand classification theoremsare the obviousgeneralizations
ofTheorems 8A and 8B; theproofsofthepresenttheorems are likethoseofthe
earlierones.

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
 MAPPINGS OF A 3-COMPLEX 283

THEOREM 13A. Two completely standardmappingsf, g of K intoR are homo-

topicif and onlyif:
(a) WJ= W2
X1 and a 2-7rrcochain
(b) thereis l-7r2-cocycle Y2 suchthat
(13.1) A,0 = b- X Wf + En,,>O, even (n;&/2)(Xa O^,oX ).
THEOREM 13B. For each W2,withchosencocycleW2, letT(W2) be thesetofall
3-7r3-cocycles
(13.2) {(X1, W) = -X W + En; >O.,even(n;&/2)(X, ,v On
classes of K intoR falls intodisjoint
X1. The set e of homotopy
for1-7w2-cocycles
subsets 5(We). The classes in 5(W2) are in one-onecorrespondence with the
elementsof thequotientgroup
(13.3) Z(K, 7r3)mod [B3(K, 7r3)+ I(W2)I.

We must show that the subsetZ' ofZ3 enclosedin bracketsin (13.3) is a sub-
group of Z3; in the formercase it was clear, since / was linear in X'. It is
to show that
sufficient
(13.4) #(X1,
I2) + 4/(Y1,W2) -,(X1 + Y1,W2),
(13.5) 46(-X1,W2)--,4(Xl) W*).
Use (12.6) in (13.2), and writex, y v1forX1, Y,', 71(a,). Then
U= V(X1,W2) + ,t(Y1,W2) - {t(Xl + y1 W2)
= {I(x <, bx) + D(y A by) ->z[(x+y) 5(x + y)in
= -E(x ,~by+ yx.
Now (see RH, (7.7))
2(x <, by)71= (X1, 02 Y,,)2n,,n(aJ = 0;
hence we may replace x bvby by -x b- by. For the same reason, we may,
in the expression
(y 6y X)71=( Oy02X1)n,,n(a,,),

replace Y, and n,.Xlby the correspondingcochains y' and x* reduced mod 2.

But
a= ,0noYes = 2(n,/2)nOY ;

hence by' = 0. Since o02X: = 0, we may apply the commutativelaw, giving

(y' x*)n,,q -(x* <, y')n,,71
= (8x -, y)7.

Therefore
-u JZ(5x y- y - X =- by)71 = [J( W- ,

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions
284 HASSLER WHITNEY

proving(13.4). The proofof (13.5) is similar;we must note that

(13.6) O~n ;- ,-)
np(Is

REMARK. We could prove that Z' is a subgroupofZ3 as follows. Let fobe a

completelystandardmappingdefiningW2. Choose
V, = Yt + 1(X,, E)eZ' i = 1,2;
we wish to show that V3 + VJ and -V are in Z'. To V3 and V3 correspond
deformationsft(O < t < 1) and ft(1 ?t I< 2); then
AfJ = V1,
A6sfoJof 1l Vf,,
l
Vf .f 2 = 2i
Now fo and f2 are homotopic. Hence, by Theorem 13A, there exist cochains
Y2 and X3 such that

A10,1 = bY3 + {(X8, W2).

ft(O? t < 1) as before,

This gives V3 + V3 = Afl0j2 CZ'. Next,define and set
ft = f2-t (1 < t < 2). The latter deformationcorrespondsto some V4 in Z'.
Usingft, 0 < t < 2, showsthat V3 + V1 = 0; hence -V3 = V3 e Z.
HARVARD UNIVERSITY

This content downloaded from [Link] on Tue, 18 Nov 2014 00:26:55 AM

All use subject to JSTOR Terms and Conditions