13.

Adams operations
For many of the applications of topological K-theory, we do not need the entire structure of
K ∗ (X); the group K 0 (X) suffices. One of the main reason for this is the Adams operations, which
create a very rigid structure on the groups K 0 (X).
Definition 13.1. A commutative ring R has a pre-λ-ring i if there exist operations λn : R R
satisfying the following relations for all r, s ∈ R:
(L1) λ0 r = 1.
(L2) λ1 = 1R .

(L3) λn (r + s) = nk=0 λk (r)λn−k (s).
We can construct a pre-λ-ring structure
n on K 0 (X) by using the exterior product of bundles.
n
More concretely, we define λ [E] = E. This also satisfies the extra nice property that
λn (E) = 0 if n > dim E.
Suppose that R is a pre-λ-ring. We can define Λ(R) to be the abelian group of power series with
coefficients in R and constant term equal to 1 under multiplication.

Then we can put a
pre-λ-ring
structure on Λ(R) in the following manner. Let α(t) = 1 + n≥1 an tn and β(t) = 1 + n≥1 bn tn .
Write formally
 
α(t) = (1 + ξn t) and β(t) = (1 + ηn t).
n≥1 n≥1
We then have  
(1 + ξm ηn t) = 1 + Pn t n .
m,n n≥1
Note that Pn is a symmetric polynomial in the ξi , ηi , and thus it can be written as a polynomial in
terms of ai ’s and bi ’s (whose coefficients are the elementary symmetric polynomials). In fact, by
degree considerations, Pn will only depend on a1 , . . . , an , b1 , . . . , bn . For example,
 
a1 = ξn b1 = ηn
n≥1 n≥1

and 
P1 = ξ n η m = a1 b 1 .
n,m≥1
For P2 we have

P2 = ξm1 ξm 2 ηn1 ηn2
m1 ,m2
n1 ,n2
(m1 ,n1 )=(m2 ,n2 )
  
2
= ξm1 ξm 2 ηn1 ηn2 + ξm ηn1 ηn2 + ξm1 ξm2 ηn2
m1 <m2 m,n1 <n2 m1 <m2 ,n
n1 <n2
 
2
= a2 b2 + b2 ξm + a2 ηn2 = a2 b2 + b2 (a21 − 2a2 ) + a2 (b21 − 2b2 )
m n
= b2 a21 + a2 b21 − 3a2 b2 .
By the general theory of symmetric polynomials, it will always be possible to express Pn in terms
of a1 , . . . , bn . Note that in order to compute Pn we do not need to assume that there are infinitely
iSometimes these are called λ-rings and what we call “λ-rings” are called special λ-rings. This does indeed
occasionally lead to confusion.
44
many variables ηi and ξj ; if there are at least n of each then the calculation of Pn in terms of the
ai and bj will be correct. We can then define a multiplication on Λ(R) by
def

α(t) ∗ β(t) = 1 + Pn (a1 , . . . , bn )tn .
n≥1

We also define  
(1 + ξi1 · · · ξin ) = 1 + Ln,m tm .
i1 <···<in m≥1

By a similar argument to the above, these will only depend on the ai and bj . For example,

L2,1 = ξ n ξ m = a2 .
n<m

We then define
def

λn α(t) = 1 + Ln,m tm .
m≥1

Note that a pre-λ-ring structure on R defines a homomorphism of abelian groups λ: R Λ(R)

given by

r λn (r)tn .
n≥0

Definition 13.2. R is a λ-ring if λ: R Λ(R) is a morphism of pre-λ-rings.

If we analyze this definition we get the following alternate formulation:
Definition 13.3. A λ-ring R is a pre-λ-ring satisfying the following extra conditions:
(L4) λn (1) = 0 for n > 1.
(L5) λn (rs) = Pn (λ1 (r), . . . , λn (s)).
(L6) λn (λm (r)) = Ln,m (λ1 (r), . . . , λmn (r)).
Example 13.4. Λ(R) is always a λ-ring.
Example 13.5. K 0 (X) is a λ-ring. To check that this is the case, it suffices to check that exterior
products of vector spaces satisfy (L1)-(L6). (L1) and (L2) are satisfied by definition. (L3) is
satisfied because
(V1 ⊕ V2 ) ∧ (V1′ ⊕ V2′ ) = V1 ∧ V1′ ⊕ V2 ∧ V1′ ⊕ V1 ∧ V2′ ⊕ V2 ∧ V2′ .

(L4) holds because for any vector space V of dimension n, m V = 0 for m > n. For any vector
space V , we can write V ∼ = V1 ⊕ · · · ⊕ Vm for linear subspaces Vi ; for any other space W write
W ∼ W
= 1 ⊕ · · · ⊕ W n . Then

V ⊗W ∼ = Vi ⊗ Wj .
1≤i≤m
1≤j≤n

In other words, the tensor product V ⊗ W is a “symmetric polynomial” in the variables Vi and
Wj . The polynomial Pℓ tells us exactly how to get this polynomial in terms of the symmetric
polynomials in Vi and the symmetric polynomials in Wj ; thus λℓ (V ⊗ W ) = Pℓ (λ1 V, . . . , λℓ W ).
A similar explanation works for λm λn (V ).
Thus a λ-ring is one where symmetric polynomials “work correctly.”
Once we have a theory of symmetric polynomials we can use Newton’s identities to define good
“power operations.”
45
Definition 13.6. Let sk be the polynomial such that when it is applied to the first k elementary
symmetric polynomials it produces the sum of the k-th powers of the variables. (By the general
theory of symmetric polynomials this polynomial is independent of the number of variables.) We
define the k-th Adams operation ψ k : K 0 (X) K 0 (X) to be
ψ k ([E]) = sk (λ1 ([E]), . . . , λk ([E])).
Since the Adams operations are defined using the λ-ring structure we expect that the operations
will be nicely natural. In fact, we have the following theorem:
Theorem 13.7. The ψ k are ring homomorphisms K 0 (X) K 0 (X) satisfying the following rela-
tions for all compact Hausdorff X:
(1) ψ k f ∗ = f ∗ ψ k ; in other words, the ψ k are natural transformations.
(2) ψ k (L) = Lk for a line bundle L.
(3) ψ k ◦ ψ ℓ = ψ kℓ .
(4) ψ p (α) = αp (mod p), in the sense that for all α there exists a β ∈ K 0 (X) such that
ψ p (α) = αp + pβ.
(5) When X ∼ = S 2n and we restrict to K  0 (X), ψ k (α) = k n α.
We will prove this theorem later. For now we use it to prove a geometric theorem.
We begin with a definition:
Definition 13.8. The attaching map of a 2n-cell is a map f : S 2n−1 S n . We define the Hopf
invariant of f to be the integer h such that α ⌣ α = hβ (for α the cohomology class represented
by S n and β the cohomology class represented by the 2n-cell attached by f .
Let X = S n ∪f D2n . Then we have a cofiber sequence
Sn X S 2n .
Since K  is an evenly graded cohomology theory on spheres, we get the following short exact
sequence:
0  0 (S 2n )
K  0 (X)
K  0 (S n )
K 0.
′
Let α be the generator of K  (S ) and let β be the generator of K
0 2n ′  (S ). (We know that both of
0 n

these groups are Z.) Write α for the image of α in K (X) and β for any preimage of β ′ in K 0 (X).
0

Since multiplication is trivial in K 0 (S n ), β ′ 2 = 0; this means that β 2 = kα for some integer k.

Fact: k is the Hopf invariant of f .j
The Hopf invariant is partially interesting because it controls how “polynomial” a simple coho-
mology ring can be. Indeed, suppose that we want to construct a cell complex X with 3 cells such
that the cohomology is Z[x]/x3 , where |x| = n for some n? When n = 2 we can do this by setting
X = CP 2 . When n = 4 we can do this by setting H = HP 2 , and when n = 8 we can do this by
setting X = OP 2 .
We now assume n > 1.
Theorem 13.9. h can equal ±1 only if n = 2, 4, 8.
Proof. When n is odd, β 2 = 0, so k
= ±1. Thus we can assume that n is even; write n = 2m.
It suffices to check that for m
= 1, 2, 4, β 2 ≡ 0 (mod 2). Using property (4) of the Adams
operations, we note that it is equivalently sufficient to show that ψ 2 (β) = 0 (mod 2).
Write n = 2m. By Property (5),
ψ 2 (β) = 2m β + µα and ψ 3 (β) = 3m β + λα
jThis is a consequence of the Atiyah–Hirzebruch spectral sequence for K-theory. Since we do not intend to take
a side foray into spectral sequences here, we leave this fact as a black box. For a reference on spectral sequences, see
for example [HatA]. We will also prove this later (assuming we have time) using the Chern character.
46
for some integers µ and λ. Our goal is therefore to show that µ is even. By Property (3) ψ 2 ψ 3 =
ψ 3 ψ 2 , and we can deduce that
2m (3m β + λα) + µ32m α = 3m (2m β + µα) + λ22m α.
Rearranging this we get
3m (3m − 1)µ = 2m (2m − 1)λ.
Note that 2m
|3m . Thus our goal is to show that when m
= 1, 2, 4, 2m
|3m − 1. Let ν(m) be
the largest power of 2 dividing 3m − 1. Note that when m is odd,
3m − 1 ≡ 3 − 1 = 2 (mod 8) and 3m + 1 ≡ 3 + 1 = 4 (mod 4).
Thus for m odd, ν(m) = 1. In addition, when m is even,
3m + 1 ≡ 1 + 1 = 2 (mod 8).
Now suppose that m = 2ℓ j, with j odd. Then we can write
ℓ ℓ−1 j ℓ−1 j ℓ−2 j ℓ−2 j ℓ−1 j
32 j − 1 = (32 − 1)(32 + 1) = (32 − 1)(32 + 1)(32 + 1) = · · ·
ℓ
 Lj
= (3j − 1) (32 + 1).
L=0
We get one power of 2 from each of the terms in the product except for the one when L = 0, which
gives us 2; we also get a single power of 2 from the first term. Thus ν(2ℓ j) = ℓ + 2. Thus 2m |3m − 1
exactly when 2ℓ j ≤ ℓ + 2. 2ℓ ≤ ℓ + 2 when ℓ = 0, 1, 2; in all of these cases we cannot have j > 1, so
these ar the only solutions. Thus the only possible values for m are 1, 2, 4, as desired.

47