if-THEORY AND THE HOPF INVARIANT

By J. F. ADAMS {Manchester) and M. F. ATIYAH {Oxford)

[Received 29 December 1964]
Introduction
THE non-existence of elements of Hopf invariant one in ir%n-^(Sn), for
n ^ 1, 2, 4, or 8, was established in (1) by the use of secondary cohomo-
logy operations. The main purpose of this paper is to show how the
use of primary operations in .K"-theory provides an extremely simple
alternative proof of this result. In fact ^-theory proofs have already
been given in (8) and (4) but neither of these proofs is elementary:
(8) uses results on complex cobordism, while (4) uses the connexion
between the Chern character and the Steenrod squares established in
(3) [see however (6) for a more elementary treatment of the results
of (3)]. The simplicity and novelty of our present approach is that,
unlike all previous attacks on the Hopf invariant problem, we consider
not the stable but the unstable version of the problem: that is to say
we shall prove directly
THBOKEM A. Let X be a 2-ceU complex formed by attaching a in-ceU to
an n-sphere, where n ^ 1, 2, 4, or 8. Then the cup-square
H*(X;Zs) + H**(X;Zt)
is zero.
For other versions of this theorem and for the historical background
of the problem we refer to (1).
Like the proofs of (4) and (8) our proof also extends to show the non-
existence of elements of Hopf invariant one modp, a result first
proved by the use of secondary operations in (9), (10). In fact our
methods yield a good deal more. In particular we shall establish the
following new result suggested to us by James:
THEOREM B. Let p be an odd prime and m a positive integer not dividing
p—\. Let X be a finite complex such that
(i) H*{X; Z) has no p-torsion,
(ii) H^{X; Q) = Oifk^k Omodm.
Then the cup-p-th-power
H**[X\ Zv) + H*"P(X; Zv)
is zero.
Quart. J. Math. Oxford (2), 17 (1966), 31-38.
32 J. P. ADAMS AND M. F. ATIYAH
Remark. Taking m = 2 and X to be the quaternionic protective space
of dimension p we see that the condition j ) 0 1 modwi cannot be
dispensed with.
We begin in § 1 by presenting our proof of Theorem A. Our aim being
to emphasize the simplicity of the proof, we refrain from any generaliza-
tions at this stage. The remainder of the paper is then devoted to
extending the methods of § 1 to a more general context. In § 2 we make
a short algebraic study of the operators t/jk. Then in § 3 we prove
Theorem B andfinallyin § 4 we show how Theorem B implies the non-
existence of elements of Hopf invariant one mod p.
1. Non-existence of elements of Hopf invariant one
We assume here the basic results of JT-theory for which we refer to
(7). We shall also need the operations i/r* introduced in (2). These are
defined in terms of the exterior power operations Afc, by the formula
0*(z) = Qk{XHx),...,X"(x)) {xeK(X)),
where Qk is the polynomial which expresses the Arth-power sum in terms
of elementary symmetric functions. Their basic properties are:
</r*: K(X) ->• K(X) is a ring homomorphism, (1.1)
^f* and ift commute, (1.2)
if p is a prime, ^(x) = xp modp, (1.3)
in n
if u e S(S ), then </r*(u) = k u. (1.4)
The proofs of (1.1), (1.2), and (1.4) are all elementary and can be found
in (2) [§ 5], while (1.3) is an immediate consequence of the congruence
(2 <HY = 2 of modp.
If we apply (1.4) to the wedge of spheres X^/X271-1 (where X" denotes
the ^-skeleton of X), we deduce at once that,
if ueK2n(X), then <p*(u) = kHt, modK2n+1(X). (1.5)
Here Kg(X) denotes the qth filtration group of K{X), i.e. it is the kernel
of K(X) -+ KiX"-1).
We are now ready to give the proof of Theorem A. The result is
trivial for n odd; in fact 2a;2 = 0 for x e Hn(X; Z), while Hin{X; Z) is
free (since n ^ 1). Thus we may suppose that n = 2m. Then H*(X; Z)
is the associated graded ring of K(X) (7) [§ 2], and so S.{X) is free on
two generators a e K^X) and beK^X).
To prove the theorem we have to show that, if m =/: 1, 2, or 4, then
a2 = 0 mod 2,
 ON JT-THEORY AND THE HOPF INVARIANT 33
or equivalently by (1.3) that
4?{a) = 0 mod 2.
Let us compute i/^(a) and ^ ( a ) . By (1.5) these must be of the form

for some integers /x, v. Since, by (1.2), tfihfj3 = ^V 2 , we deduce, using

(1.1) and (1.5), that
Zm{2ma-\-fib)+v2imb = 2m(3ma-\-vb)+fiZbnb,
and so fr1^"1— l)/x = 2m(2™— l)v. (1.6)
But, by elementary number theory [cf. (2) Lemma 8.1], we have
if m =£ i j 2, or 4, then 2™ does not divide 3 m —1. (1.7)
Thus (1.6) implies that /x is even and hence
^*(a) = 0 mod 2
as required. This completes the proof of Theorem A.
2. Eigenspaces of i/A
We recall that the Chem character induces a ring homomorphism
7 L1O
[( ) 1 ch: K*{X) ® Q -> H*(X; Q)
and that, if x e K(X) with

then [(2) Theorem 5.1 (vi)]

Thus, if we use the Chem character to identify K(X) ® Q with

2 Him(X; Q), the subspace Him(X; Q) becomes (for k > 1) just the
eigenspace Vm of ^ corresponding to the eigenvalue km, which shows
in particular that this eigenspace is independent of k. The dimension
of Vm is just the 2mth Betti number B^X). The following lemma is
then a consequence of (2.1):
LEMMA. 2.2. Lei X be a finite connected complex and assume that the
Betti numbers B^X) are zero for m =£ 0, m^, m,,..,, mr. Then, for any
sequence of integers kv..., kT
J I (</*—(kt)™*) = 0 in K(X) ® Q-
We now observe that this result is stated purely in terms of X-theory
and makes no reference to oohomology or the Chern character, provided
that we define the Betti numbers (as we may) by

3*96.2.17 D
34 J. F. ADAMS AND M. F. A T I Y A H
Moreover we can prove (2.2) purely in K-theoTy simply by using (1.5)
and induction on the filtration. We propose therefore to take (2.2) as
our starting point and to use only X-theory. Rational cohomology was
mentioned only for motivation.
Let k, I denote integers greater than 1. From (2.2), taking kt = k
for all t, we see that </r* is semi-simple and has eigenvalues &"*< on
£ <g> Q. Let Vifk denote the eigenspace corresponding tofc"1*.Then

Applying (2.2) with kt = k for j # t and kt = I we see that

This being true for all k, I > 1 it follows that Vi)e = V^ and so Viik is
independent of k (as was shown earlier by use of cohomology). We
denote it therefore by V^ Thus we have a decomposition

R{X) ®Q = ®Vi (2.3)

invariant under all the ^*. Let TT{ denote the projection operator corre-
sponding to T^. Then for any sequence of r—1 integers k^,..., k(_iy ki+1,
..., kr (all kt > 1) we have the following expression for 77^:

In fact 774 annihilates Vt for j ^ 1 and is the identity on Vt.

3. The pth power mod p

So far we have only considered the vector space S.{X) ® Q. Now we
turn our attention to the image of S.{X) in S(X) igi Q. An element
of this image will be called an integral dement of R(X) ® Q. If
x e R{X) ® Q, then there is a least positive integer d such that dx is
integral. We call d the denominator of x. For convenience we shall
now make the following definition. Given a sequence % , . . . , mT of
distinct positive integers and an integer t with 1 ^ i ^ r we define
dt(mv..., mr) to be the highest common factor of all the products
JJ (k^-kp),
where {kj} (j =fc t) runs over all sequences of r—1 integers > 1. With
this notation (2.4) gives the following result.
PROPOSITION (3.1). Let X be as in (2.2) and let x e R(X) <g> Q be
integral. Then the denominator of TT{X divides di(m1)...>mr).
 ON X - T H E O R Y AND T H E H O P F I N V A R I A N T 35
p
Now let p be a prime and let us compute ip (x) using the decomposi-
tion (2.3). Thus _ v
x — 2.7ri*>
*) = 2 *>""*<*• (3-2)
m
Suppose now that x is integral and that, for each i, p < does not divide
di(m1,...,TOr). Then (3.1) and (3.2) show that in R(X) ® Q we have

**<*> = f .
where y is integral and q is prime to p. Transferring this result from
R{X) ® # to R(X) and using (1.3) we obtain
THEOREM C. Let X be as in (2.2), let p be a prime and suppose for each
i that p™* does not divide d^rn^,..., mr). Then for any x e R(X) toe have
XP e pR(X)+ToTs R(X)
(where Tors denotes the torsion subgroup). In particular, if K(X) has
no p-lorsion, then xp = 0 modp.
This theorem, stated entirely in K-theory, is the most general result
concerning the triviality of thepth power modp given by our method.
From it we shall now deduce a corollary about the pth power map in
H*{X; Zp):
COBOLLABY. Let X, p be as in Theorem C and assume further that
H*(X; Z) has no p-torsion. Then, for any m > 0, the p-th power map
H^(X; Zp) -* H*»»(X; Zp)
is zero.
Proof. Let Ap denote Z localized at p, i.e. the ring of fractions m/n
with n prime to p. Since the differentials of the spectral sequence
H*(X; Z) => K*(X) are all torsion operators [(7) 2.4] and since X has
no p-torsion, it follows that the localized spectral sequence (i.e. the
spectral sequence obtained by applying ® Ap) is trivial and that K*(X)
has no p-torsion. Thus we have
H^(X; Ap) as H*»{X; Z) ® Ap s K^X^K^^X) ® Ap. (3.3)
Also, since X has no p-torsion,
H*»(X; Ap) -> H^(X; Zp) (3.4)
is surjective. Hence, if a e H^X; Zp), we can find x e Kim(X), a.eAp
so that x <g> a represents a via (3.3) and (3.4). Then x" ® aP represents
ap. But, by Theorem C, xp = 0 (modp). Hence a" = 0 as required.
In order to apply Theorem C and its corollary in any given case it is
necessary to verify the arithmetical hypothesis, namely that p"*< does
36 J. F. ADAMS AND M. F. ATIYAH
not divide dt. In general this can be quite complicated. In the special
case required for Theorem B however, where we have mi = mi, the
arithmetic can be dealt with as we shall now show. The following lemma
for odd primes is the appropriate generalization of the lemma for p = 2
already used in § 1:
LEMMA (3.5). Let p be an odd prime, m a positive integer not dividing
p— 1 and 1 ^ i ^ p. Then, for a suitable integer k, the p-primary factor
of Y[ (&"**—&m^) is at most p™'1, where the product is taken over all in-
tegers j with 1 < j <. p and j ^ t.
Proof. We begin by recalling that the multiplicative group of units
of the ring Zp/((?/8ay) is cyclic of order p^^ip—l). Let k be an integer
whose residue class modp2 generates 02. Then it follows that the
residue class of k modp^ generates Of. Thus we have
kn = 1 mod^ <=> n = 0
and this holds for all /. With this choice of k we shall compute the
^-primary factor p* of JJ (k™*—kmi). Suppose that (m,p—1) = h, so
that
m = ah, p-l = bh, a > 1,
and let the jj-primary factor of m (or equivalently of a) be p1. Then we
find that
f
where, as usual, [x] denotes the integral part of x. Thus
1
)= h{f+1)

Moreover, equality cannot hold in all places since / + 1 = p1 implies

/ = 0 and therefore hpf = h = m/a < m. Hence e < m as required.
We are now ready to prove Theorem B. First we observe that
replacing X by Xipm+1 aflFects neither the hypotheses nor the conclusion
of the theorem. Thus we may suppose the Betti numbers B^X) are
zero except for ft = m ^
Moreover we may assume that X is connected. But (3.5) implies that
pm does not divide d^m, 2m,...,pm) for 1 ^ i ^ p and so the hypotheses
of Theorem C are certainly fulfilled. Theorem B then follows from the
corollary to Theorem C.
Remark. Of course the corollary to Theorem C applied with p = 2
leads to a suitable strengthening of Theorem A along the lines of
Theorem B. We leave this to the reader.
 ON X-THEORY AND THE HOPF INVARIANT 37
4. The mod p Hopf invariant
In this section p denotes an odd prime. Let / : S2"1" -*• S2™*1 be any
map and let ^ = ^ ^ ^ eimp+1

be the associated 2-cell complex. The mod^j Hopf invariant is denned

to be the Steenrod operation
i*»: H**"(Xf; Zp) -> H»»P^(X,; Zp). (4.1)
We propose to prove
THEOBEM D. The modp Hopf invariant w zero for m > 1.
Proof. We begin by recalling that the loop space QiS*™^ has the
following cohomology:
HT(D.Sim+1; Z)^Z if r = 0 mod 2m,
= 0 otherwise.
This is an elementary consequence of the Serre spectral sequence.
Suppose now that / : S2™? -»• S2"1*1 is any map and let

be its 'adjoint'. We form the space

Yg = Q.S2m+1 ff

Ifm does not divide #—1, this satisfies all the conditions of Theorem B
except that it is not a finite complex. However, we can approximate
Yg by a finite complex up to any dimension and so the conclusion of
Theorem B holds for Yg: that is the pth power map
H**(YB; Zp) + H*">(YB; Zp)
is zero. Hence suspending once
pm. H^+^SYg; Zp) -> H^p+^SYg; Zp) (4.2)
is also zero. But, by definition of g, f is homotopic to the composition
Sg
L
where e is the 'evaluation' map
e(t, w) = w(t),
and so e extends to a map
e': 8Yg + Xf.
Since e induces an isomorphism on H*"1*1, it follows that e' induces a
monomorphism in cohomology, and so the vanishing of (4.2) implies
the vanishing of (4.1). This proves the theorem for values of m not
38 ON ^ - T H E O R Y AND T H E H O P F I N V A R I A N T
dividing p—1. However, these exceptional values are easily dealt with
by a method due to Adem. I n fact, for 1 < m < p— 1, we have

pm — _ pipm-l
m
[(5) § 24], and so P™ is zero on a 2-cell complex.

REFERENCES
1. J. F. Adams, 'On the non-existence of elements of Hopf invariant one', Ann.
of Math. 72 (1960) 20-104.
2. 'Vector fields on spheres', ibid. 75 (1962) 603-32.
3. 'On Chem characters and the structure of the unitary group', Proc.
Cambridge Phil. Soc. 57 (1961) 189-99.
4. 'On the groups J(X) TV', to appear in Topology.
5. J. Adem, 'The relations on Steenrod powers of cohomology classes', Algebraic
Geometry and Topology, Princeton (1957).
6. M. F. Atiyah, 'Power operations in iT-theory', to appear.
7. and F. Hirzebruch, 'Vector bundles and homogeneous spaces', Proc. of
Symposia on Pure Mathematics 3, Differential Geometry, American Math.
Soc. 1961, 7-38.
8. E. Dyer, 'Chem characters of certain complexes', Math. Z. 80 (1963) 363-73.
9. A. Liulevicius, 'The factorisation of cyclic reduced powers by secondary
cohomology operations', Memoirs ofOie American Math. Soc. No. 42 (1962).
10. N. Shimada and T. Yamanoshita, ' On triviality of th&mod p Hopf invariant',
Japanese J. of Math. 31 (1961) 1-25.