CONJECTURE

HOW NOT TO PROVE THE POINCARE

John R. Stallings

Introdu tion
I have ommitted | the sin of falsely proving Poin are's Conje ture. But that
was in another ountry; and besides, until now no one has known about it.
Now, in hope of deterring others from making similar mistakes, I shall des ribe
my mistaken proof. Who knows but that somehow a small hange, a new interpre-
tation, and this line of proof may be re tied!
In the ba k of my mind when I on eived my proof was this theorem.
Theorem 0. (For n 6= 2). Let f : M ! K be a map of a onne ted orientable
n-manifold into an n- omplex, and let C1 , : : : , Ck be some of the n-simplexes of K
su h that the degree of f on ea h Ci is zero (that is, the homology map indu ed by
f , Hn (M ) ! Hn (K; K int Ci ), is zero. Suppose f indu es a homomorphism of
1 (M ) onto 1 (K ). Then f is homotopi to a map into K (int C1 [


[ int Ck ).
A spe ial argument establishes this theorem for n = 1. For n  3 we may argue
as follows. We shall make a number of hanges on f whi h will be independent
of ea h other, sin e n  3; hen e we need only onsider the ase that k = 1 and
suppose that C1 is overed twi e with opposite orientations by f (M ). The inverse
image of a small ell in C1 is the union of two ells A and B in M . Let P be a path
in M from A to B ; fP represents an element of 1 (K ). Sin e 1 (M ) ! 1 (K ) is
onto, we an modify P by adding on a loop whose image represents the inverse of
fP ; thus we an suppose fP is a null-homotopi loop in K .
Sin e the dimension of M is at least 3, we an hoose P to be a non-singular
path and hange f by a homotopy in the neighborhood of P so that f (P )  C1 . If
T is a tube around P , then A [ T [ B will be an n- ell mapped into C1 with degree
zero. A further homotopy within A [ T [ B will un over a point of C1 ; by pushing
away from that point, we un over all of int C1 .
But, in my proof of Poin are's Conje ture, I need this theorem for n = 2. The
argument above fails in this ase for several reasons. We annot un over 2- ells
independently of ea h other; we annot make the path P non-singular; if P were
non-singular, the homotopy bringing f (P ) into C1 might ause us to over up ells
whi h we want to un over.
The reader may be able to pat h up some of these points. If he pat hes up all
these points, he will have proved the Poin are Conje ture (for we shall show how
Typeset by AMS-TEX
1
2 JOHN R. STALLINGS

Theorem 0 for n = 2 implies the Poin are Conje ture) in orre tly. For, Theorem 0
is false for n = 2: Consider a torus with two 2- ells C1 and C2 atta hed to kill the
fundamental group; there is a map of the 2-sphere into this omplex; by a homotopy
we an un over either C1 or C2 , but not both simultaneously.
1. A onje ture about the 3-sphere
In the 3-sphere S 3 let T be a tame 2-manifold su h that both omponents of
S 3 T have free fundamental groups. Let U and V denote the losures of the
omponents of S 3 T .
A ording to theorems of Papakyriakopoulos, both U and V are handlebodies.
The only on eivable su h embedding of T is shown in Fig. 1.
[Figure 1 is a pi ture of a standard 2-manifold embedded in R3 , bounding a han-
dlebody, with a urve on it separating it into two pie es, su h that in ea h half of
the omplement in R3 , the urve bounds a disk.℄
However, if the genus of T is greater than 1, this standard embedding has a
property whi h does not seem to follow immediately from the fa t that U and V
are handlebodies. Namely, there is a simple losed urve C on T , not ontra tible
on T , yet bounding 2- ells both in U and in V .
Conje ture A. The existen e of su h a urve C an be proved only from the
hypothesis that both U and V are handlebodies.
If we hope that Conje ture A is true, a reasonable dire tion to attempt a proof
of the Poin are Conje ture an be made as follows.
Poin are's Conje ture is that any simply- onne ted 3-manifold M is a 3-sphere.
It is known that any orientable 3-manifold su h as M , has a Heegaard representation
as U [ V , where U and V are handlebodies and U \ V is their ommon boundary,
a 2-manifold T .
If the genus of T should happen to be one, then M is a lens spa e, and so, if
simply- onne ted, is a 3-sphere.
Assume that we ould prove Conje ture A for M , rather than for the 3-sphere:
That is, if the genus of T is greater than one, then on T there is a simple losed
urve C , not ontra tible on T , yet bounding 2- ells in both U and V . Then we
ould write M as the onne ted sum M1 # M2 of two manifolds whose Heegaard
representations would have less genus. And so by indu tion on the genus, we would
know that M is indeed a 3-sphere.
2. Redu tion to group theory
Let M = U [ V , T = U \ V be a Heegaard representation of a 3-manifold.
We obtain a diagram of fundamental groups, with homomorphisms indu ed from
in lusions:
1 (U ) ' 1 (T ) ! 1 (V )
& # .
1 (M )
  CONJECTURE
HOW NOT TO PROVE THE POINCARE 3

Sin e ' and are homomorphisms onto, it follows from van Kampen's Theorem
that 1 (M ) is isomorphi to the quotient of 1 (T ) by (ker ')
(ker ). Hen e M is
simply onne ted exa tly when

1 (T ) = (ker ')
(ker ):

We have an obvious homomorphism to investigate,

'  : 1 (T ) ! 1 (U )  1 (V ):
The kernel of this homomorphism is learly,

ker '  = (ker ') \ (ker ):

Therefore our geometri problem has been \redu ed", by virtue of Dehn's Lemma,
to the more algebrai problem: Does ker '  ontain an element whi h an be
represented by a simple losed urve on T ?
Now this annot be true for arbitrary 3-manifolds, for if it were we would have
proved, \Every 3-manifold is a onne ted sum of lens spa es", whi h is absurd.
Theorem 1. '  is a homomorphism onto, if and only if M is simply onne ted.
First, if '  is onto, sin e 1 (U )  1 (V ) is the produ t of the kernels of the
proje tions onto its fa tors, it follows that 1 (T ) is the produ t of ker ' and ker ,
and hen e M is simply onne ted.
Conversely, if M is simply onne ted, then 1 (T ) = (ker ')
(ker ). Let (;  )
be an arbitrary element of 1 (U )  1 (V ). Sin e ' and are onto, there are 1 ,
1 in 1 (T ) su h that '(1 ) =  and (1 ) =  . We an de ompose 1 and 1
thus: 1 = x2 , 1 = 2 y , where x and 2 belong to ker ' , and 2 and y belong
to ker . Then '  (2
2 ) = (;  ). Hen e '  is onto.
Now our \redu tion" an be stated as the following onje ture:
Conje ture B. Let T be an orientable 2-manifold of genus n > 1. Let F1 and F2
be free groups of rank n. Let  : 1 (T ) ! F1  F2 be a homomorphism onto. Then
there is a non-trivial element of ker  whi h is represented by a simple losed urve
on T .
We have shown that Conje ture B implies both the Poin are Conje ture and
Conje ture A. It is likely that from the data of Conje ture B we an re onstru t a 3-
manifold; in whi h ase, then, onversely, the Poin are Conje ture and Conje ture A
together would imply Conje ture B.

3. Finding simple losed urves.

Papakyriakopoulos and Maskit have dis overed an interesting hara terization
of simple losed urves in terms of planar overing spa es. Their results show that
Conje ture B is equivalent to the following:
4 JOHN R. STALLINGS

Conje ture C. In the situation of Conje ture B, there is a non-trivial normal

subgroup N of 1(T ), su h that N  ker  and su h that the overing spa e of T
whi h orresponds to N is a planar surfa e.
However, in our dis ussion we are interested in a dierent hara terization of
simple losed urves.
Let us say that a homomorphism ' : G ! A  B of a group G into a free produ t
of groups is essential if there is no element x 2 A  B su h that x
'(G)
x 1 is
ontained in one of the fa tors A or B .
Theorem 2. If G = 1 (T ), where T is a losed 2-manifold, and if ' : G ! A  B is
an essential homomorphism, then there is some non-trivial element of ker ' whi h
is represented by a simple losed urve on T .
Proof. : Represent A  B as the fundamental group of X = XA [ XB where XA and
XB are open sets in X with fundamental groups A and B respe tively, and where
XA \ XB is simply onne ted. Represent ' by a ontinuous fun tion f : T ! X .
We an divide T into submanifolds TA and TB , whose interse tion is their om-
mon boundary, su h that f (TA )  XA and f (TB )  XB . The omponents of
TA \ TB are simple losed urves, whose images by f lie in XA \ XB and hen e are
ontra tible in X .
If, ontrary to the on lusion we wish to draw, every su h simple losed urve
were trivial on T , then any one, say C , would bound a 2- ell D. Redene f on
D so as to map D into XA \ XB , and then redene TA and TB ; this will redu e
the number of omponents of TA \ TB . Finally, we obtain a map f 0 and a division
of T so that either TA or TB is empty. f 0 indu es the same homomorphism on
fundamental groups as f does, modulo an inner automorphism (we may have moved
the basepoint around), so that ' is inessential, ontradi ting hypothesis.
Reinterpreting Conje ture B in the light of this theorem, we have:
Conje ture D. In the situation of Conje ture B, the map  : 1(T ) ! F1  F2
an be fa tored through an essential map of 1 (T ) into some free produ t A  B.
Thus have we repla ed the purely geometri Poin are Conje ture by the purely
algebrai Conje ture D.
4. Geometri \proof" of Conje ture D
Let us onsider the map  : 1 (T ) ! F1  F2 with omponents ' and , so that
 (x) = ('(x); (x)). It is possible given that  is onto, after a moderate amount of
algebrai sli kness, to nd a presentation of 1 (T ) as
Y
n
fa ; b ; : : : ; an ; bn : [ai ; bi ℄ = 1g
1 1
i=1
and bases f1 ; : : : ; n g and f1 ; : : : ; n g of F1 and F2 su h that, modulo the om-
mutator subgroups , (ai )  (i ; 1) and (bi )  (1; i ).
  CONJECTURE
HOW NOT TO PROVE THE POINCARE 5

Now interpret F1 as the fundamental group of a bouquet of ir les X1 _ X2 _

_ Xn, where Xi orresponds to the basis element i . Similarly interpret F2
as the fundamental group of Y1 _


_ Yn , with Yi orresponding to i . Dene
W = (X1 _


_ Xn )  (Y1 _


_ Yn ).
We may interpret  as the homomorphism indu ed from a fun tion f : T ! W .
Be ause of our lever hoi e of bases, f will have degree zero on the tori Xi  Yj
for i 6= j and degree one on the tori Xi  Yi .
Let W  denote (X1  Y1 ) [ (X2  Y2 ) [


[ (Xn  Yn ). This is the union of n
tori with a single point in ommon.
If only, alas, Theorem 0 were valid for dimension two, we ould on lude that the
map f : T ! W ould, up to homotopy, be fa tored through a map g : T ! W  .
The map g would have to indu e an isomorphism on 1-dimensional homology, and
hen e, writing 1 (W  ) as the obvious free produ t, the map of fundamental groups
would be essential.
And so Conje ture D would be proved. Poin are's Conje ture would follow.
Fame and Fortune would be ours.
5. Con lusion
There are two points about this in orre t proof worthy of note.
The rst is that when we try to prove Theorem 0 in dimension two, we always
run up against the problem of trying to simplify, by some geometri tri k, the
situation. But any little homotopy that would simplify the pi ture always in fa t,
greatly ompli ates it. This phenomenon has hara terized every attempt that I
have made or heard of to prove Poin are's Conje ture. This is the pla e to look for
aws in any asserted \proof".
The se ond point is that I was unable to nd aws in my \proof" for quite
a while, even though the error is very obvious. It was a psy hologi al problem,
a blindness, an ex itement, an inhibition of reasoning by an underlying fear of
being wrong. Te hniques leading to the abandonment of su h inhibitions should be
ultivated by every honest mathemati ian.
6. Afterword
This is a TeXed version (November 2000) of the original [3℄, whi h was written
when the author was at Prin eton University.
Referen es
[1℄ B. Maskit, A theorem on planar overing surfa es with appli ations to 3-manifolds, Ann. of
Math. 81 (1965), 341{355.
[2℄ C. D. Papakyriakopoulos, A redu tion of the Poin ar e Conje ture to group-theoreti onje -
tures, Ann. of Math. 77 (1963), 250{305.
[3℄ John R. Stallings, How not to prove the Poin ar e Conje ture, Topology Seminar Wis onsin,
1965, Ann. of Math. Studies, vol. 60, 1966, pp. 83{88.

Mathemati s Dept., University of California, Berkeley, CA 94720

E-mail address : stallmath.berkeley.edu