THE HOPF INVARIANT

ALLEN YUAN

1. Introduction and Motivation The Hopf map S 3 S 2 is well known as a ber bundle with bers S 1 . It can be seen by viewing S 3 as the unit sphere in C2 = R4 and composing with the map C2 {0} CP1 = S 2 . Furthermore, in [2], Hopf proved that this map is not homotopic to the identity. In fact, we can do the same construction in quaternionic and the octonionic projective plane to get maps S 7 S 4 and S 15 S 8 , respectively, which have bers S 3 and S 7 , respectively. From these maps, there is reason to suspect that perhaps it is possible to get a handle on 2n1 (S n ), or homotopy classes of maps from S 2n1 S n . Suppose we are given a map f : S 2n1 S n of oriented spheres. From this, we can obtain a homotopical invariant as follows: regard f as the attaching map of a 2n-cell to the CW complex with exactly one 0-cell and one n-cell. In the resulting complex, there is an interesting cup product of the integral cohomology class corresponding to the oriented n-cell with itself (all other cup products are tautologically zero). Since the Z-cohomology in dimensions n and 2n are Z, this cup product is a multiple of the oriented 2n-cell. The integer constant of proportionality is an invariant of the homotopy class of the map f . We dene this constant as the Hopf invariant H (f ). The map H : 2n1 (S n )

Z turns out to be a group homomorphism (this may be hinted

at by the fact that precomposition of the attaching map with the antipodal map induces multiplication by 1 on the cohomology in dimension 2n). We attempt to study the properties of this homomorphism. It is not too dicult to show that when n is odd, the map is trivial, and when n is even, the image of the map contains the unique subgroup of index 2. In this paper, we present proofs of these and several related results. The proofs and notation given are based on proofs in [3]. One may ask, however, when this homomorphism is surjective.
1

This is equivalent to the question of when there exist maps of Hopf invariant 1. This question, however, is very deep in general. We shall show it is necessary that n must be a power of 2. Adams [1] showed that, in fact, the only n for which Hopf invariant 1 maps exist are n = 1, 2, 4 and 8. 2. Preliminaries 2.1. Denition of the Hopf invariant. Let f : S 2n1 S n be a map of oriented spheres. Assume that n > 1, because we have a good handle on 1 (S 1 ) - thus, all homotopy groups that follow are abelian. We denote by K the complex with one 0-cell, one n-cell, and one 2n-cell attached by f . We have that H (K ) =

for = 0, n, 2n and H (K ) = 0 otherwise. Let

H n (K ) denote the generator induced by the orientation of S n and H 2n (K ) denote the generator induced by the orientation of S 2n1 . The Hopf invariant is a map H : 2n1 (S n ) Z given by the formula 2 = H (f ) , where 2 denotes the cup product of with itself. One can rst observe the following. Observation 2.1. The Hopf invariant is 0 for n odd because by the anti-commutativity of the cup product, i.e., 2 = (1)deg 2 = 2 so H (f ) = H (f ) . 2.2. A combinatorial lemma. We will need the following technical result (sometimes known as Lucass Theorem): Lemma 2.2. Let p be a prime and a = 0 ai , bi < p. Then b a
m m i=0

ai pi and b =

m i i=0 bi p

be integers where

i=0

bi ai

(mod p).

Proof. There are many ways to attack this including induction and combinatorial arguments. For example, see Wikipedia for a combinatorial argument. We shall present a simple proof using the binomial theorem in but is also (1 + x)bi p =
i

Zp [x].
i

We have that the coecient of xa in (1 + x)b is

b a bi ai

(1 + xp )bi which has xa coecient

. Thus,

b a

(mod p) as desired.

3. Examples of the Hopf invariant To get a handle on the Hopf invariant, we may try computing the Hopf invariant of certain maps. For the Hopf bration S 3 S 2 described in the introduction, the Hopf invariant can be seen to be 1 because this map is the attaching map of the 4-cell in CP2 , in which the 2-cell has cup square equal to the cohomology class corresponding to the 4-cell. Another natural candidate for preliminary computations can come from consideration of the cell structure of S n S n when n is even. This consists of a 2n-cell attached to the wedge of two n-spheres. The attaching map is therefore a map g : S 2n1 S n S n . We can now compose this with the natural map F : S n S n S n given by identifying the two spheres via a pointed homeomorphism between them. We can now ask what the Hopf invariant of F g : S 2n1 S n is. Proposition 3.1. H (F g ) = 2 (up to a sign) when n is even. : Sn Proof. Let K be the complex formed by attaching a 2n-cell by F g . There is a map F S n g e2n K dened to be an extension of F : S n S n S n K where the nal map is the inclusion i. Such an extension to the 2n-cell exists, because the map iF g : S 2n1 = e2n K is nullhomotopic. Thus, we have the commutative diagram

Sn Sn n n n S S =S S n g e2n
i

> Sn
i

>K

In cohomology, we observe that the map F takes the generator H n (S n ) to 1 + 2 where 1 , 2 are the two generators of H n (S n S n ) corresponding to the two n-cells. Furthermore, by the denition of cup product (and using, say, simplicial cohomology), one can show that
2 2 the cup product structure on S n S n is 1 = 2 = 0, 1 2 = 2 1 = where is a generator

is a homeomorphism of the 2n-cell relative to its boundary of H 2n (S n S n ). Additionally, F to their image, and so it induces an isomorphism H 2n (K ) H 2n (S n S n ). Hence, we may let be in H 2n (K ) such that is the pullback of . Finally, putting this all together, we have

2 = (F )2 = (F )2 = (1 + 2 )2 = 1 2 + 2 1 = 2 = 2F F which implies that the Hopf invariant is 2.

We now get for free the following statement. Corollary 3.2. For n even, there exists a map of Hopf invariant 2. 4. Properties of the Hopf invariant We start by establishing a property of H that was mentioned in the introduction. Proposition 4.1. H : 2n1 (S n ) Z is an additive (abelian) group homomorphism. Proof. The idea is to relate the addition operation in the homotopy group to the denition of the Hopf invariant. To do this, we examine the following way of representing the sum of two elements f, g 2n1 (S n ). If Q : S 2n1 S 2n1 S 2n1 is the map that collapses the equator to a point, then we have f + g is the element represented by the composition S 2n1
Q

> S 2n1 S 2n1

f g

> Sn Sn

> Sn

where F is the same map as in the proof of Proposition 3.1. We then dene some spaces that will help us relate the complex K for f + g to some sort of sum of the complexes for f and g . Let K = S n F (f g)Q e2n , L = (S n S n ) (f g)Q e2n , and M = (S n f e2n ) (S n g e2n ). Dene the generators of the cohomologies to be H n (K ), H 2n (K ), H 2n (L), 1 , 2 H 2n (M ), 1 , 2 H n (L). Denote by the same names the pullbacks of these through inclusion maps of S n and S n S n . We have the commutative diagram

K< Sn <

>M

Sn Sn > Sn Sn

in Proposition 3.1 (i.e., extended from F to the where p is constructed analogously to F 2n-cells), q is induced by Q, and vertical maps are inclusions. We have that H (f + g ) = H (f + g )p = p (H (f + g ) ) = p 2 = (1 + 2 )2 in L, analogously to Proposition 3.1. However, observe that in M , we have
2 2 + 2 = H (f )1 + H (g )2 . (1 + 2 )2 = 1

Pulling back to L, we get (1 + 2 )2 = H (f )q 1 + H (g )q 2 = H (f ) + H (g ). Finally, plugging this into the original equation gives H (f + g ) = (H (f ) + H (g )) which proves the desired proposition.

It follows that by Observation 2.1 and Corollary 3.2, we deduce that for n odd, H is the zero map, and for n even, H has image containing 2Z (this implies, in particular, that 2n1 (S n ) contains an innite cyclic group). The obvious remaining question to ask is when H is surjective. This is visibly equivalent to the question of whether there exist maps of Hopf invariant 1. We saw one case in which this was true at the beginning of Section 3. In fact, the brations S 7 S 4 and S 15 S 8 also have Hopf invariant 1. It is a deep theorem due to Adams in [1] that these are in fact all examples. Instead of proving this theorem, we settle for a weaker result. Theorem 4.2. If there exists f : S 2n1 S n such that H (f ) = 1, then n is a power of two. Proof. Observe that (using the regular notation), we have that Hopf invariant 1 means = 2 = Sq n (here, we work with the

Z/2

Steenrod Algebra). In particular, we need that

Sq n = 0. A particular case where it is 0 is if Sq n =

ai Sq i where ai is in the Steenrod

Algebra. Call Sq n decomposable if such a decomposition exists. It suces to show that Sq n is decomposable if n is not a power of two.

Suppose n is not a power of two, and let n = 2k + m where 0 < m < 2k . Now, use the Adem relations to expand 2k 1 c Sq nc Sq c . m 2c Sq a+b = Sq a+b (mod 2) by Lemma 2.2 for p = 2.

Sq m Sq 2 =
c

We observe that the c = 0 term is

2k 1 m

It follows that Sq a+b is decomposable, as desired. This completes the proof.

We thus have a good picture of the image of the map H . We have proven that it is 0 when n is odd, 2Z when n is even but not a power of 2, case is always 2Z.

when n = 2, 4, 8, and either

or 2Z

otherwise. The theorem due to Adams [1] nishes the picture by showing that the otherwise

References
[1] J.F. Adams, On the non-existence of elements of Hopf invariant one, Ann. Math. 72 (1) 1960. [2] H. Hopf, ber die Abbildungen der dreidimensionalen Sphre auf die Kugelche, Mathematische Annalen 104: 637665 1931. [3] R.E. Mosher and M.C. Tangora, Cohomology Operations and Applications in Homotopy Theory, Dover Publications, 1968.