19,20,21 &22

The next paper, by Atiyah and Hirzebruch, is the foundation

paper on K-theory. It requires no more prerequisites than may be
expected at this stage, but the writing may be found condensed in
places. The following exposition, by Hirzebruch, can be recom-
mended for its clarity. The next note, by myself, is merely a
summary of results which were later published in full. For all
these three papers (but especially the last) the reader can consult
Husemoller for further details; or see Atiyah.
The fourth extract in this group is intended to give an idea
of one particular application of K-theory. The prerequisite is a
certain sympathy for homotopy-theory and homological algebra.

196

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 19

VECTOR BUNDLES AND HOMOGENEOUS SPACES

BY

M. F. ATIYAH AND F. HIRZEBRUCH

Dedicated to Professor Marston Morse

Introduction. In [1] we introduced for a space X the "ring of complex
vector bundles'* K{X). The Bott periodicity of the infinite unitary group
[8; 9; 10] implied that K satisfied the "Kiinneth formula"
K(X X S9) c* K(X) ® K(S*)
which was fundamental for the proof of the differentiable Riemann-Roch theo-
rems [1; 16].
Using the Bott periodicity we construct in §1 a "periodic cohomology theory":
For every integer n, the abelian group Kn(X) is defined, K°(X) is K(X) and
/T +2 (X) is isomorphic with K*(X)t the group K\X) is the kernel of the homo-
morphism K\X X S1) -> K\X) induced from the embedding X-+X X S\
This cohomology theory satisfies all the axioms of Eilenberg-Steenrod [14]
except the "dimension axiom." For the space consisting of a single point,
K* is infinite cyclic for even n and vanishes for odd n. The axioms without
the dimension axiom do not characterize the theory, even if the values of Km
are given for a point. There is a spectral sequence relating the ordinary co-
homology theory with our periodic theory (§2).
In §§3-5 we try to get information on K° and K1 for classifying spaces and
certain homogeneous spaces. An important tool is the differentiable Riemann-
Roch theorem which we recall in the beginning of §3. The final goal would
be to answer all those questions for the if-theory on homogeneous spaces which
for the ordinary cohomology theory have been treated so successfully by A.
Borel (see for example [3]). We can give only partial results in this direction.
The new cohomology theory can be applied to various topological questions
and may give better results than the ordinary cohomology theory, even if
the latter one is enriched by cohomology operations (see [2] and M. F. Atiyah
and J. A. Todd, On complex Stiefel manifolds, to appear in Proc. Cambridge
Philos. Soc). This justifies the new theory.
In spite of its length, the present paper is by no means a final exposition.
The proofs are often sketchy and the definitions and results could be generalized
in certain cases. For example, using real vector bundles and the Bott periodicity
of the infinite orthogonal group, we can define a periodic cohomology theory
with period 8. This is not more difficult than in the unitary case. Furthermore,
the definition of K(X) in 1.1 can be given for any topological space. For con-
venience, we have restricted the theory to the special class 21 (see 1.1). We
197

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)

have then the homotopy classification theorem (1.3)

(1) K(X) S*[X,ZX Bu], (X e « ) .

For this actually 21 could be chosen much larger. But in general (1) would
be wrong. The restriction to 21 simplifies the presentation of certain conse-
quences drawn from the spectral sequence. For any topological space, we
can take the right side of (1) as a definition of a functor k(X). If Z X Bv
is endowed with a natural structure of a commutative ring (up to homotopy),
then k{X) has a natural (commutative) ring structure for any space X and the
rings K(X) and k(X) are isomorphic if X t $. Such a "ring" structure on
Z X Bu has been defined by Milnor (not published). In view of Milnor's
construction it would perhaps be more natural to study the functor k(X),
but since Milnor's result is not yet at our disposal we have studied K(X) where
sum and product structure is automatically given by the Whitney sum and
the tensor product of vector bundles.
For the classifying spaces Bo we have defined 3Z(B0) as an inverse limit
indicating by the curly letter that we mean neither K{BQ) nor k(Bo). We
conjecture that X(Ba) is isomorphic to k(Ba) for any compact l i e group G.
But we shall deal with this question elsewhere. We prove for a compact con-
nected Lie group G that X(Ba) is isomorphic with the completed representation
ring R(G) (see 4.8).

1. A cohomology theory derived from the unitary groups.

1.1. Let H be the class of those spaces which can carry the structure of a
finite CW-complex. For I t S w e have defined in [1] an abelian group K(X).
There we gave the definition only for a connected X, but we may define K(X)
in general as the direct sum of the groups 2£(X.) where the «X\ are the con-
nectedness components of X. For the sake of completeness we recall the defi-
nition of K(X) and give it directly for a space X t % not necessarily connected.
We adopt the usual definition of a complex vector bundle over X except
that we allow the bundle to have fibres of different dimensions over the various
connectedness components of X. We can now verbally repeat the definition
of II]:
Let F(X) be the free abelian group generated by the set of all isomorphism
classes of complex vector bundles over X. To every triple / = (£, £', {") of
vector bundles with £ ^ £' © £" we assign the element [t] = [£] - [$'] - [£"]
of F(X), where [£] denotes the isomorphism class of £. The group K(X) is
defined as the quotient of F(X) by the subgroup generated by all the elements
of the form [/].
The tensor product of vector bundles defines a commutative ring structure
in K(X); the unit 1 is given by the trivial bundle of dimension 1.
K is a contravariant functor: for a continuous map f : Y —* X (Y, X c S )
we have the natural ring homomorphism f : K(X) —> K(Y) induced by the

198

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
lifting of bundles under /. We denote it by f to distinguish it from the analogous
homomorphism /* in ordinary cohomology theory.
1.2. Let 8 be the class wnose objects are the pairs (Ar, x0) with X t 21 and
xQ e X. Usually we shall write an object of 3 simply by indicating the space X.
Very often the base point x0 of X is naturally given by the context. For X t 8
we define the reduced group R(X) as follows: the ring K({xo\) is canonically
isomorphic with Z (the ring of integers). The imbedding i : {xo\ —> X induces
the ring homomorphism

We define i£(X) to be the kernel of i\ It is an ideal of K(X). Whenever a

symbol like R(X) occurs it is to be understood that Ar is a space with base
point, i.e., an object of 8.
We now consider the class SB consisting of pairs (X, Y) where X can be given
the structure of a finite CW-complex in such a way that Y becomes a subcomplex.
For (X, Y) t 23 we define
K(X, Y) = R{X/Y).
Here X/Y is obtained from X by collapsing Y to a point which becomes then
the base point of X/Y. By [19], X/Y t 8. Note that R{X) = K(X, x0) for
X t l If Y is empty (Y = 0), then X/0 = X* (where X + is the topological
sum of X with an extra point which becomes base point of X*) and K(Xt 0) =
R(X+) = K(X).
For X, Y t S the objects X V Y and X A Y of 3 are defined. (In the liter-
ature, X A Y is also denoted by X # Y). X V I7 is obtained from the topo-
logical sum of X and F by identifying the base point of X with the base point
of Y to one point which becomes the base point of A" V Y. The space X A Y
is X X Y with the union of the axis x0 X Y and X X t/0 collapsed to a point
which becomes the base point of X A Y. We have the natural maps
X V Y->X X Y-*X A 7
and may write
(i) x A y- x x Y/X v y.
The operations V and A are associative and commutative and A is distributive
over V. This means, for example, that there is a canonical homeomorphism
between X A Y and Y A X.
If Sn E 3 is the standard n-sphere with base point, we write
(2) ff(X)-S"Al, (XeS).
This is the nth suspension of X. Since
Sn = Sl A S1 A • • • A Sl (n times)
n
it follows that S (X) is the n times iterated suspension of X.

199

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
DEFINITION. For any integer n ^ 0 we put 2T"(X, Y) = R{ST{X/Y))}
((X, Y) t ffl). ForXt% we put K'n(X) = R-n{Xt 0) = R(S*(X*)). For
X 11 with base point x0 we put R~n(X) = iT n (X, x0) = R(ST(X)).
For n = Owe have the groups already defined:
K°(X, Y) = K(X, Y), K°(X) = K(X), R°(X) = R(X).
Of course, the X~B are also contra variant functors.
1.3. We write [A, B] for the set of homotopy classes of maps of the space
A into the space B and correspondingly [A, 17; B, V] for the homotopy classes
of maps of the pair (A, U) into the pair (J5, V). If the spaces A and B have
base points, then we write [A, B]o for the set of homotopy classes of maps
preserving base points.
Let Bv be the classifying space of the infinite unitary group [10] and Z X Ba
the cartesian product of it with the group of integers (Z having the discrete
topology). In Z X Bvwe choose a base point lying in 0 X Bv, The classification
theorem for unitary bundles [18, §19.3] gives rise to the following natural bi-
jective maps (compare also [16, §1.7, 2.1]):

-(X, Y) s [S\X/Y), Z X Berk ((X, F) « » ) .

&[X, Y; V(Z X Bv), point],

S* [X, 7 ; Qn"lC7, point], n > 0.
We recall that Z X Bc^ is weakly homotopy equivalent to an liT-space, namely
to Q.U (Bott, see [8]). Thus all the above sets of homotopy classes are endowed
with a natural group structure. The above bisections are in fact all group
isomorphisms. Since 17 is weakly homotopy equivalent to 12(Z X Bu), the
space tf(Z X Bv) is weakly homotopy equivalent to Z X Ba and we have
an isomorphism
(3) /T ( "* a) (X, Y) & K~n(X, Y), n % 0.
We shall give later an explicit description of an isomorphism between these
two groups.
If x0 denotes the space consisting of a single point, then

and thus [9]

K'n(x0) ^ Z for n even and K~*(x0) = 0 for n odd.
1.4. PROPOSITION. // (X, Y) t 53 we have exact sequences

200

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
(B+1) n
(i) > X- (y) A K~ (X, Y) -» X""(X)
-> K - ( y ) -* > K°(X, Y)
u+1)
(ii) > £- (F) -U X""(X, 7) -> £"(X)
-* Jf-"(y) -> > K\X, Y)
For (ii) u?e assume X, Y t 8 tw7A £0 = 2/0 c 7 .
PROOF. We use the paper of Puppe [17]. If Y and X are arbitrary spaces
with base point and / : Y —> X a map preserving base points, then there is
a sequence of spaces and maps (with base points)
y -u x £4 cf *4 S1Y -> slx -»s'c, — s 2 y -* s 2 x ->. • •
such that the following is true: if V is any space with base point, then the
functor [ , V]o gives an exact sequence of sets. Here we note that exactness
is a property of sets with preferred elements—the group structure is irrelevant.
The preferred element is always given by the constant map onto the base point.
We recall the construction of C/. First we take the cone
CY = F X / / 7 X l U t / 0 X / .
Then we take the topological sum CF + X in which we identify (y, 0) t CY
with j(y) for each y t Y. The space Cf contains X as subspace. Cf/X is (canoni-
cally homeomorphic with) the first suspension of Y. This gives rise to the
maps Y —>f X -+Pf Cf —>0/ SlY. All the other maps in Puppe's sequence are
suspensions of these. If Y is a subspace of X and / the injection, then we have
a natural homeomorphism X/Y S Cf/CY. If (X, Y) belongs to 93 then it
satisfies the homotopy extension condition and according to Puppe the map
Cf —> Cf/CY followed by the above mentioned homeomorphism is a homotopy
equivalence h. The composition h o Pj is the natural projection X —» X/Y.
Taking this into account Puppe's theorem applied to V = Z X Bv gives the
exact sequence (ii) and all homomorphisms in this sequence are canonically
defined by Puppe's maps. K'n(Xt Y) -* R'n(X) is induced from X -> X/Y.
The sequence (i) is obtained by replacing in (ii) Y and X by Y+ and X* re-
spectively.
REMARK. If Y — {x0) then the sequence (i) breaks off in split exact sequences

0 — R-n(X) -> K'H(X) -> X-(xo) -* 0.

Hence
K"\X) & R'n(X) © *n(Z X Bv), (see 1.3).
n
The exact sequence (i) is obtained from (ii) by adding to R~ (X) and also to
R'n(Y) the direct summand *n(Z X Bv).
1.5. We have mentioned in 1.1 that K(X) = K°(X) is a commutative ring.
We wish to define more generally products also involving the groups K~* (n ^ 0).

201

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
Suppose X, Y t 8. Then X V F = X X y0 U z0 X F is a subspace o f l X K .
We apply 1.4 (ii) to the pair (X X Y, X V Y). The exact sequence breaks
off in this case into split exact sequences.
(4) 0 -> R'\X A F) -» R-\X X Y) -» R'n(X V F) -> 0, (n & 0),
and we have a canonical decomposition
(5) R'\X X F) s £~"(X A F) © £-(X) ® £-(Y).
For the proof of (4) we observe that R'n{X X F) - • £' W (X V F) is surjective
and that this homomorphism may be regarded as the projection onto a direct
summand. For this we make use of
R~n(X V F) = R(S\X v F)) = R(SnX v 5 T ) = R(SnX) ® R(SHY)

We have the following natural group homomorphisms which are all induced
by the tensor product of vector bundles
(6) K(X) (x) K(Y) K(XX Y), (X, Y eSD,
(7) R(X A Y), (X, Y eS),
(8) K(X, Xo) ® JST(rf Fo) r,x o x : rvJX X Yo) i
for (X, X,) and (Y, Yo) £ 93«
It is clear how (6) is denned. If a t R(X) and b t R(Y), then the product is
in the kernel of R(X X F ) - > R(X Xy0VxQXY). By (4) and (5) the product
is well defined as element of R(X A F), (n = 0). If we replace in (7) X by
X/X o and F by Y/Yo we get the definition of (8). More generally we have
a group homomorphism
(9) K~m(X, Xo) ® K-n(Y, Fo) -> 2T(w*n)(X X F, Xo X F U X X F o ),
for (X, Xo), (F, Fo) c 93 and m > 0, n ^ 0.
We get this from (7) and the fact that
SM(X/X0) A S"(F/F 0 ) = £m+B(X/Xo A
- sm+\xx F/XO
The equality sign means that there is a natural homeomorphism between these
spaces. If one uses the natural identification of X X F with F X X, one gets
from (9) a product
(90 K-\Y, Fo) ® K'm(X, Xo) -> K-(m+H)(X X F, X o X Y \J X X F o ).
LEMMA. / / a t K'm(Xt X o ) and b t K'n(Y, F o ), then ab « (-l) m n 6a where
db is the image of a (x) b under (9) and ba the image of b (x) a wnder (9').

202

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 09)
PROOF. The sign comes from the use of the various "natural identifications"
between different spaces. Sm A (X/Xo) A Sn A (Y/Yo) and S* A (Y/Yo) A
Sm A (X/Xo) are identified just by the permutation. However, for the defi-
nition of (9) we employ the identification
am.m: Sm A ST—S***
which comes from a map Sm X S* —• Sm*n of degree + 1 (all spheres and also
the cartesian product in this order have the standard orientations). If ft is
the permutation Sm X Sn -* Sn X Sm, then an,m o fi o a-ln has degree ( - l ) m \
This shows that ab and ba correspond to elements of
G - [S-+"(X X Y/Xo X Y\JXXYO),ZX B*]9
m+n
which are related with each other by a map of S onto itself of degree (—l) m \
Since the group structure of G can also be defined by the suspension coordinate
like a homotopy group, the lemma follows.
1.6. Using the diagonal map as in the definition of the cup product we get:
PROPOSITION. Let X t 21. Then ]T«*o K~n(X) is a graded anti-commutative
ring. Let (X, Y) t 33. Then there is a ^graded homomorphism"
K~(X, Y))-+T, K-'(X, Y),

making 2Z mi0 K~m(X, Y) a graded module over X«*o K~n(X).

The products have functorial properties. For example, if (X, Xo) —** (X't XJ)
and (F, Yo) —*f {Y\ Y!>) are maps with the pairs all belonging to SB, then we
have the commutative diagram
2r"(X', Xi) (x) K-\Y', YS) ~> JST-^^-^X' X Y', X'QX Y'\J X' X Y'«)
l !
(10) J/ ®(7 [
n ( +B)
jr-(z, Xo) ® K' (Yf Yo) — /r - (x x Y, X 0 X r u x x r 0 ).
Furthermore, for / : Y —* X, the induced homomorphism /' : ^2mi0 K~m(X) —>
52*io K~m(Y) is a ring homomorphism, etc.
1.7. The Bolt isomorphism. The existence of the Bott isomorphism (see 1.3
(3)) is the central and deep point of the cohomology theory we are developing.
We give now the explicit description of this isomorphism.
Let x0 be the space consisting of a single point. Then (1.3) K~2(x0) is infinite
cyclic. By definition K~2(x0) = R(S2). Let rj be the complex line bundle
over S2 whose first Chern class equals the canonical generator of H2(S2t Z).
Then n represents an element [rj\ t K(S2) and [rj] — 1 is a generator of R(S2) =
K-*(x0) which we denote by g. Hat 7Tm(X, Xo), then ag t K~lm+2)(X, Xo).
Here we use 1.5 (9) with Y = x0 and Yo empty.
THEOREM. The map a —» ag is an isomorphism o] K~m(Xf Xo) onto
(m+2)
/JL" (X, Xo). In particular, ^n^o.K'n{xo) is the polynomial ring Z[g].

203

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
For a proof of this central theorem we refer to [10]. For any (X, Xo) t 33
the graded group £«so K~n(X, Xo) is a module over Z\g\. Multiplication
with gk gives an isomorphism of K~n(Xf Xo) onto K~in+2k)(X, Xo). This holds
in particular if Xo is empty or reduces to the base point, i.e., 2«*o K~n(X)
and Yl*±o R~n(X) are both modules over Z[g], Let 0 denote the multiplication
by g. The next lemma follows from 1.6 (10).
LEMMA. / / (X, F) and (X', Y') belong to 58 and if f : (X, Y) -* (X', Yf)
is a continuous map, then f(3 = j3f where f : K~n(Xft F') -> K~n(X, F), (n ^ 0),
is the induced homomorphism, in other words: f is a homomorphism of Z[g]-modules.
LEMMA. / / (X, Y) z 53, then (3 gives a homomorphism of exact sequences (1.4 (ii)),
i.e.y we have the commutative diagram (n ^ 0)
(F)-!>/r"(X, Y) -> R~\X) -> R~\Y)
[ \ \
gri%+*\Y) -*+ ir ( B + 2 ) (x, Y) -+ i?"(B+2)(X) -* £- ( n + 2 ) (y).
TAe corresponding statement holds for the exact sequence (1.4 (i)).
PROOF. This follows from the preceding lemma. We take into account that
the homomorphism 8 is also induced by a map, namely by Cf —> SlY.
1.8. The group K"2n(X, Y) can be identified with K°(X, F ) a n d i r (2n+1)(X, Y)
with K~*(X) by the Bott isomorphisms:
F:K°(X, Y)-*K-2n(X, Y),
/3n : K'\Xt Y) - • K"(2B+l) (X, 7).
This allows us to define Kn(Xt Y) for any integer n by
KB(X, 7) = K°(X, F) if n is even,
iT(X, F) = K " 1 ^ , 7) if n is odd.
The groups 2£B(X, Y) satisfy the usual axioms of a cohomoiogy theory [14]
(in the category S3 with all continuous maps of one pair into another one being
admissable) except that Kn(x0) does not vanish for n ?* 0 (1.3). The existence
of an exact sequence
(11) • K*(Y) -*> 2T+1(X, Y) ~> 1T+1(X) -> iT +1 (I0 - * . . • ( - « < n < • )
follows from 1.4 and the second lemma of 1.7.
Let (X, F, Z) be a triple with X D F D Z and ail the pairs (X, F), (X, Z),
(F, Z) belonging to 53. Then we have an exact sequence
(11*) > K\Y, Z) -4 K n+I (X, F) ~* 1T+1(X, Z) -> iT + 1 (F, Z) - • • • • ,
(-co <n < oe>),
1 +1
where the 8 of (11*) is the composition 2T(F, Z) - • K*(Y) -* 1T (X, F).

204

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
The exactness of (11*) would follow from 1.4(ii) applied to the pair (X/Z, Y/Z)
if this belonged to S&. But (11*) is also a consequence of the cohomology axioms.
(Excision-, homotopy-, and dimension axioms are not needed for this formal
deduction of (11*); compare [14, Chapter I, §10].)
1.9. In 1.8 we have completed the construction of a cohomology theory sat-
isfying all axioms except the "dimension axiom." Since these "cohomology
groups" are periodic (Kn(Xt Y) = Kn+2(X, Y)) it is convenient to define
K*(X, Y) - K\Xy Y) 0 K\X, 7), (X, Y) t 93,
and similarly for K*(X) and R*(X). K*(X) is then an anti-commutative
ring, graded by Z2, i.e., K°(X) is a subring and
K°(X)-K\X) C K\X), Kl(X)-K\X) C K\X).
Moreover K*(X, Y) is a Z2-graded module over K*(X). Since 5 respects the
periodicity, we have the exact triangle
K*(Y)±K*(X, Y)
(12) \ /
K*(X)
which resolves in an exact hexagon
K\X, Y) -» K\X)

K\X) ^- K\X, Y)
T
and w hich has, so to speak, the exact sequence (11) as "universal covering."
For a triple X, Y, Z (see 1.8) we have the exact triangle
K*(Y,Z)-*K*(X, Y)
(12*) \ /
K*(X, Z)
and the corresponding hexagon.
1.10. The Chern character. For each complex vector bundle { over the space
X t tl the Chern character ch(£) is defined as an element of the rational co-
homology ring H*(X, Q), [5, §9.1]. If H"(Xy Q) denotes the direct sum of
the even dimensional cohomology groups (which is a commutative subring of
H*(X, Q))} then cA(f) t H"(X, Q). The definition of ch($ uses only the total
Chern class c(f). The classes ch(£) and c(f), both regarded as elements of
H"(X, Q), determine each other. The Chern character induces a ring homo-
morphism [16, §12.1 (5)]
(13) eh: K{X) - K\X) - H'\X, (?) C H*(X, Q)
with
ch(R(X)) C 8*(X, Q) = Kernel (HHX, Q) -> H*({x0), Q)).

205

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
We are now going to define a group homomorphism
(14) ch: K'\X, Y) -> H*(X, Y; Q), ((X, 7 ) e 8 , n ^ 0 ) .
n n
By definition, K~ (X, Y) = R(S (X/Y)). We have the suspension isomor-

<r": 5*(A, Q) -* # * ( S m Q), A e «,

which raises degrees by n and is defined by ten soring a t ff*(A, Q) (from the
left) with the canonical generator of Hn(Sn, Z), If f c K~n(X, Y), let {' be
the "corresponding element1' of R(Sn(X/Y)). Then ch(?) t 0*(S*(X/Y), Q)
and (<r")"1 ch(?) t ff*(X/Y; Q). We have the canonical isomorphism
a:B*(X/Y,Q)->H*(X, Y;Q)
and we define

In 1.7 and 1.8 we described the Bott isomorphism. Since ch([ri] — 1) is the
canonical generator of H2(S2, Z) and since ch preserves products, it follows
easily, that cA(/3({)) = ch(£) for £ t K~n(X, Y). Therefore we can define ch(£)
for f e K*(X, F), n any integer. Using the notation of 1.9 we have now defined
the Chern character as a homomorphism
ch;K*(X, Y)-+H*(X, Y;Q).
eh maps K°(X, Y) into H"{X, F; Q) and K\X, F) into Hod(X, F; Q) which
denotes the direct sum of the odd-dimensional cohomology groups. The fol-
lowing theorem is easy to check.
THEOREM. The Chern character is a "natural transformation" of the <{co-
homology theory" described in 1.9 into the ordinary cohomology theory with rational
coefficients for which one only considers the Z2-grading H* = H'* © H°d. In
particular, ch preserves products, commutes with maps, ch o f = /* o ch, and
one has commutative diagrams
K\Y) A K\X, Y) K\Y)± K\X, Y)
ch\ ch\ ch\ ch\
H'XY, Q) + H-\X, F; Q), W\Y, Q) + H'\Xt Y; Q).
The commutativity of these diagrams can be deduced from the fact that
the 6 of both theories is induced from the map Cf —> SlY (compare 1.4). One
has to be careful with the signs. We hope to have chosen the various definitions
such that commutativity (not only commutativity up to sign) holds in these
diagrams.

2. The spectral sequence. Let X be a finite simplicial complex. We shall

206

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
establish a spectral sequence relating the integral cohomology ring of Ar with
K*(X).
2.1. Let Xn be the n-skeleton of X. We use the it-theory defined in 1.8.
We filter JC{X) by defining
Knp(X) = Kernel [Kn(X) ->K n (X" 1 )].
THEOREM. Let X be a finite simplicial complex. Let x0 be the space consisting
o] a single point, so that K9(x0) = Z ij q is even and K9(xo) « 0 if q is odd. There
exists a spectral sequence E*r'9 (r££ 1, — oo < p , q < » ) with
(1) E?'S*C'(X,
dx being the ordinary coboundary operator.
(2) Erc*H\X,
(3) EZ' 9* GJC+'iX) - Kl+'(X)/Klll(X).
The differential dr : E r' -> £j -«- vanishes for r even since E^'9 = 0 for all
p 9 +r r+l

odd values of q.
PROOF. We use the method of [12, Chapter XV, §7] and define the graded
group
H(p, q) = £ H"(p, 9) = Z K"(X-\ X'-), 9 £ p.
These H(p, q) satisfy the axiom (SP.1)-(SP.5) of [12, loc. cit.]. For axiom
(SP.4) see 1.8 (11*).

where a* runs through all p-simplices. But a*/o* = /S". Therefore Xp4*(<rj, <r?) ^
Jf*+€(iS') S i^(>S0) ^ ^(xo). This proves (1). To get (2) one has to check
that dx is the ordinary coboundary operator.
2.2. REMARK. The preceding spectral sequence can be generalized to a
fibre bundle (F, X, F) with projection TT : Y ~» X. If this fibre bundle satisfies
certain conditions, then there is a spectral sequence with E\'9 ^ C*(Xy K*(F))
and Ep2'9 S # P (X, 2T(F)) (local coefficients). Furthermore El9 S GJC^9{Y)
with respect to a certain filtration of 2£p+a(F). This spectral sequence specializes
to the one of the theorem for Y — X and rr the identity.
2.3. The whole spectral sequence of 2.1 is compatible with the Bott periodicity.
This makes it possible to forget about the grading and to use the notation of 1.9.
THEOREM. Let X be a finite simplicial complex. Let K*(X) be the kernel of
K*(X) -+ K*(Kv~l). There exists a spectral sequence E*r{X)t r ^ 1, with

El(X) s QJW3Q - K*(X)/K*^(X).

The differentials dr vanish for even r.

207

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
This spectral sequence could also be obtained directly by the method of
[12, Chapter XV, §7] by putting H(p, q) = K*(X9"\ X9'1).
REMARK. It is easy to show (by the notion of 1-equivalence, [12, p. 336])
that the E^X) together with the differentials dr are homotopy type invariants
of X for r ^ 2. Also K*9(X) is a homotopy type invariant. It can be invariantly
defined as follows: an element £ of K*(X) belongs to K*(X) if and only if for
any finite simplicial complex Y of dimension ^ q — 1 and any continuous
map / : Y -> X we have /'{ = 0. Thus the spectral sequence [E'(X), r ^ 2}
is well-defined for any space X of the homotopy type of a finite simplicial
complex. By a theorem of J. H. C. Whitehead [19, p. 239, Theorem 13] any
finite CW-complex is of the homotopy type of a finite simplicial complex.
Hence the spectral sequence {E"r(X), r ^ 2} is well-defined for spaces of the
class U (see 1.1).
The differentials dr are certain (higher order) cohomoiogy operations.
d> : El ^ mX, Z) -» E?* 2* # P+3 (X, Z) is the Steenrod operation 8q\
2.4. Let X be a finite simpiicial complex. We propose to study the spectral
sequence of 2.3 in its relation with the Chern character. Let 'E*r be the spectral
sequence with
f
E\ = C9(X, Q), dx the ordinary coboundary operator,
'K = H*(X, Q) for r g: 2, 'dr - 0 for r > 2.
This trivial spectral sequence is obtained by the method of [12, Chapter XV, §7]
by putting 'H(r, s) = H*(X"\ Xr~l; Q) for s ^ r. The spectral sequence
of 2.3 comes from H(r, s) = K*(X'~l, Xr~l). The Chern character gives a
homomorphism
ch:H(r,s)->'H(r,s),
and since ch is a natural transformation from the 2£*-theory to the rational
cohomoiogy theory, we get a homomorphism ch from the spectral sequence
[El] of 2.3 into the spectral sequence [rE*r\. Using ch we can prove:
THEOREM. Suppose X t% (see 1.1). The spectral sequence {2£?(X)} collapses
(i.e., dr = 0 for r ^ 2 and thus E$(X) ^ EZ(X)) if one of the following conditions
is satisfied:
(i) H*(X, Z) has no torsion,
(ii) H*(X, Z) = 0 for all odd integers i.
PROOF. We may assume that X is a finite simplicial complex, ch : E*r -* 'E*r
is always injective for r = 1, since then it is just the coefficient homomorphism
C(X, Z) -* C(X, Q). For r = 2 it is the homomorphism £P(X, Z) -» #*(X, Q)
which is injective if X has no torsion. Since the 'dr vanish for r ^ 2 it follows
by induction on r that the dr also vanish for r ^ 2 if El —• 'W% is injective.
This proves the theorem under the assumption (i). If (ii) holds, then dr(r ^ 3,
odd) vanishes since it maps E*r(X) in E*r*r(X), and one of these groups is 0.
The dr (r even) vanish anyhow.

208

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
THEOREM. Suppose X t 21 (see 1.1). The spectral sequence \E*(X) (x) Q}
collapses (i.e., dr ® Q = 0 for r ^ 2).

w fo>dwe and maps K°(X) (x) Q onto # " ( * , Q) and £ ' ( * ) (x) Q onto H"(X, Q).
PROOF. We may assume that X is a finite simplicial complex. The spectral
sequence {E*(X) (g) Q] is obtained by putting "Hip, q) = K*(X9'\ Xp~l) (x) Q.
The Chern character gives a homomorphism of this spectral sequence into the
spectral sequence {'El(X)\ which is bijective for r = 1. This implies the
theorem (compare [12, Chapter XV, Theorem 3.2]).
COROLLARY. Suppose X t SI (see 1.1). / / K*(X) has no torsiont then
ch: K*(X) -> H*(X, Q)
is injective.
2.5. The preceding results on the spectral sequence imply:
COROLLARY. Let X be a space belonging to 81 (see 1.1). Then K*(X) is addi-
lively a finitely generated abelian group. The rank of K°(X) equals the sum of
the even dimensional Betti numbers of X, whereas the rank of Kl(X) is the sum
of the odd dimensional Betti numbers of X.
For any { t K*(X) let chn(£) be the n-dimensional component of ch(£).
COROLLARY. Suppose X t H and that H*(X, Z) has no torsion. Then
(i) { t K*9(X) if and only if chr(() = 0 for r < p, in particular
ch: K*(X) -+ H*(X, Q)
is injective and K*(X) is without torsion, i.e., free abelian.
(ii) / / { E K*(X), then chv(£) t HV(X, Q) comes from an integral class which
is uniquely determined and equal to the image of £ in K*(X)/K*^(X) ^ HP(X, Z).
To every integral p-dimensional class x, there exists £ t K*V(X) with cAp({) = x%
i.e., ch(£) = x + higher terms.
(iii) Let A be a subgroup of K*(X). If for every x t HV(X, Z)t p ^ 0, there
exists it A with ch(£) = x + higher terms, then A = K*(X).
2.6. So far we have not studied the behaviour of the spectral sequence (2.3)
with respect to the product structure of K*(X). We have only been able
to get a partial result which we summarize without proof in the following
theorem.
THEOREM. Suppose X t U. We consider the spectral sequence E*r(X) (r ^ 2)
with the operators dr. Let Z* be the kernel and Bp the image of dr. There exist
pairings J[r : E>(X) (x) E*r(X) -» £* + f (X) with

(4)
•*B**\X) and B*(X)®Z'r(X)

209

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)

and such that I l r + i ** induced from JJ, in virtue of (4). Moreover} J][3 is the
cup-product and PJ« is the product in GK*(X) induced by the ring structure of
K*(X) far which

(5) K*(X).K:.(X)CK:+AX).
We conjecture that dr is an anti-derivation. This would imply (4). We
shall only need (5) in the sequel. (5) admits a straightforward proof.
By (5) the mth power of an element of K*i(X) belongs to K%(X). If m is
sufficiently large then K*>(X) is zero, hence any element of K*X{X) is nilpotent.
Clearly, £ t K^(X) if and only if cAoft) = 0. This special case of 2.5 (i) holds
for any X e 8. We conclude:
PROPOSITION. An element £ of K*(X) is nilpotent if and only if c/io(£) = 0.
An element rj of K*(X) is inuertitle if and only if chQ(r}) — ± 1 .
PROOF. It remains to show that rj is invertible if cho(ri) = dbl. In this
case, dirj — 1 — £ with cho(%) = 0 and thus £ nilpotent. Then i?~l =
i + £ + • • • + r*1) ^ r = o.
3. The differentiable Riemann-Roch theorem and some applications.
3.1. We recall the Riemann-Roch theorem given in [1] in a slightly more
general formulation. Let X, Y be compact oriented differentiable manifolds.
By the triangulation theorem of Cairns, X and Y belong to the class 51 of 1.1.
A continuous map / : Y —> X will be called a c^map if we are given an element
d(/) z H\Y, Z) such that cx(f) a w2(Y) - f*w2(X) mod 2 where w%(Y) and
w2(X) are the second Stiefel-Whitney classes of Y and X respectively
(w2 t H2( , Z 2 )). As in [5; 1], if J is a real vector bundle with finite-dimensional
base B € we define
4© - IX (*.-/2)/(sinh (*,/2)) e #*(£„ Q)
where the Pontrjagin classes of £ are the elementary symmetric functions in
the x<. If £ is the tangent bundle of the differentiable manifold X we write
& instead of 8ft).
THEOREM. Le/ F and X be as before. Let f : Y —* X 6e a d-rwap. T^en //iere
homomorphism
g : K*(Y)-> K*(X)
such that
(i) ch(g(y))-UX) - /,(c/iW(/)/2-•«(Y))f V * tf*(*0,
^ is //ic Gysm homomorphism (Poincare dual of the homology homomorphism).
(ii) g wiaps K°(F) into K°(X) and Kl(Y) into K\X) if dim Y = dim X
(mod 2).
0 maps K°(Y) into K\X) and Kl(Y) into K°(X) if dim Y # dim X
(mod 2).

210

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
(iii) g is related to the homomorphism /' : K*(X) —» K*(Y) by the formula
9(j\x)-y) = x-g{y), xtK*(X), ytK*(Y),
Ij we define &(j) = t(F)-/*(S(X)"'), then (i) may be written as
0') ch(g(y)) = L(ch(.y)e"(""-&(J))-
This theorem is slightly more general than Theorem 1 of [1] which was formu-
lated for K°(X). Here we have stated it for K*(X) which makes the assumption
dim Y ss dim X (mod 2) superfluous. The proof does not have to be changed
once one has developed the cohomology theory of §1. Moreover we assert
here the existence of the homomorphism g satisfying (iii). This brings no
additional difficulty. One just has to follow up the proof of Theorem 1 of [1]
(see also [16]). Something new would be involved if we tried to choose g in a
natural way (call it then /,) and prove certain functorial properties of it. We
shall take up this question in a more detailed exposition. Nevertheless we
permit ourselves to call the g of the theorem /,. But we are not allowed then
to use for Z —•' Y —*f X the formula (/ o / ) , = /, o /,. (The composition of
two (vmaps is a Cx-map in a natural way.) The formula (i') shows that ch(g(y))
is uniquely determined for a Ci-map /. Therefore (2.4, 2.5), g = /, is given
without ambiguity if K*(X) or H*(X, Z) has no torsion.
It follows easily from (i') that
ch((J oj)xz) = <*(/,(,*)) for ztK*(Z).
By (2.4, 2.5)
(Jo})* - M/*) if K*(X) or H*(X,Z) has no torsion.
3.2. Let Y be a compact oriented difTerentiable manifold. It is called a
Ci-manifold if we are given an element cx(Y) t H2(Y, Z) whose restriction mod 2
is wa(Y). For a Cj-manifold Y the Todd genus T(Y) is defined. It is equal
to the value of the top-dimensional component of eei(Y)n • S(F) on the funda-
mental cycle of Y. By definition, T(Y) is a rational number. It is an integer
as follows by applying Theorem 3.1 to the map of Y onto a point. Compare
[1], see also [6]. If Y is almost-complex and cx(Y) the first Chern class, then
T(Y) is the usual Todd genus which is equal to the arithmetic genus if Y is
a projective algebraic manifold [15).
3.3. Let { = (2£{, Bh Fh T^) be a difTerentiable fibre bundle in the sense
of [5, §7.4]. Assume that E^Bit F^ are compact oriented difTerentiable manifolds.
As in [5] we let | be the bundle along the fibres. This is a real vector bundle
over E | whose second Stiefel-Whitney class w2(£) equals w2{Et) — Tr*w2{Bk).
Assume that ir = X| is a Cj-map. Then Ci(x) s= w2(£) mod 2. If i : Fi —> E^
is the injection of a fibre in the total space then
i*Cl(x) s w2(Ft) mod 2.
Therefore if we put Ci(F|) = i*Ci(w), the manifold F$ becomes a c,-manifold

211

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
and we can speak of the Todd genus T(F^). Assume that | is endowed with
a complex structure, i.e., we are given a complex vector bundle rj over E^ which
considered as real vector bundle is |. Then F$ is almost complex in a natural
way. Furthermore w is a d-map with d(ir) = Ci(iy). The Todd genus
is then the same whether we consider F^ as cx-manifold with cx(F^ = i
or as almost complex manifold.
3.4. THEOREM. Let £ be a differenliable fibre bundle as in 3.3. Let *
be a Ci-map. If the Todd genus T(F^) = ± 1 then the homomorphism

is injective. Moreover ir! identifies K*(B^) with a direct summand of (^

The endomorphism TX O T1 of K*(Bs) is the multiplication with a fixed invertible
element of
PROOF. We shall use Theorem 3.1 for x with Y = 2?{ and X = £«. First
we observe that

Therefore with ^ = TTI we have by 3.1 (i')

ch{icx(y)) - / # (c%) .«- lCO/> .*(&), ye
Now put y = 1, the unit of K*(E{). Then cft(t/) = 1 and it follows easily that
the zero-dimensional component of ch(ir{l) equals T(F^). Since T(F { ) = ± 1 ,
T J is an invertible element in K*(B^) (see 2.6) whose inverse we denote by a.
Now let h be the homomorphism K*(E^) —* K*(Bs) equal to x, followed by
multiplication with a; then (iii) of 3.1 gives
h(r\x)) - s for all xe
which proves the theorem.
The preceding theorem admits various generalisations. For example, if the
Todd genus T(JF{) = m j* 0, (m e Z), then x l is injective on the direct sum of
those p-primary components of /£*(£$) with p =» 0 or a prime not dividing m.
This type of theorem is analogous to 3.2 of [4].
3.5. Let G be a compact connected Lie group and T a maximal torus of G.
Let £ be a principal G-bundle whose base space 2?* is a compact oriented dif-
ferentiable manifold. Consider the associated bundle with G/T as fibre. Its
total space is E^/T, its base space is B^. With these assumptions we have:
PROPOSITION. Let * be the projection E^/T —• B f , Then

is injective. T!2iL*(B{) is a direct summand of K^iE^T).

PROOF. We may assume that £ is differentiable. The bundle along the fibres
of EJT admits a complex structure such that G/T has Todd genus 1 (see

212

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)

[5, §§7.4, 22.3]). The complex structure along the fibres and the orientation
of B | define an orientation for the compact differentiate manifold EJT. The
proposition follows from 3.4.
THEOREM. We make the preceding assumptions. Let U be a closed connected
subgroup of G of maximal rank, i.e., we may assume U D T~ Then E^/U is the
total space of the bundle associated to £ and with G/ U as fibre. Let a be the projection
Ei/U-*Bs. Then

is injective. <rlK*(Bi) is a direct summand of K*(Ei/U).

PROOF. We have the diagram
Ex/T -A Ei/U -Z>Biy (rop = r, T 1 = p! o <r!.
By the above proposition T 1 is injeetive which implies a1 is injective. Also
the last statement of the theorem follows immediately.
REMARK. We have proved this theorem under the assumption that B^ is
a compact oriented differentiate manifold. A small generalization of the
Riemann-Roch Theorem 3.1 makes it possible to drop the assumption on
orientability. It is probably also true when B^ is any finite CW-complex.
The preceding theorem holds in particular for bundles with an even dimen-
sional sphere as fibre and the special orthogonal group as structure group.
If x.: Y —> X is such a bundle (X compact oriented differentiable), then ir1 :
K*(X) —» K*(Y) is injective. The corresponding theorem for integral co-
homology holds if X has no 2-torsion (more generally, T* is injective on the
direct sum of the p-primary components of H*(X, Z) with p = 0 or p an odd
prime).
3.6. THEOREM. Let G be a compact connected Lie group, U a closed connected
subgroup of G of maximal rank. Then K\O/U) = 0 and K°(G/U) is a free
abelian group with rank equal to the quotient of the order of the Weyl group of G
by the order of the Weyl group of U.
PROOF. The theorem is true if U = T (maximal torus of G). In this case
G/T has no torsion in integral cohomology and its odd dimensional cohomology
groups vanish [7], The theorem follows then from 2.5 if one takes into account
that the order of W(G) (Weyl group of G) is the Euler number of G/T which
equals dim 0 H0W(G/Tt Q). For the general case, we assume that U D T and
consider the map x : G/T —> G/U* Then ir! is injective by 3.5. It follows
that K\G/U) - 0 and that K*(fi/U) has no torsion. It is well-known [3]
that the odd-dimensional Betti numbers of G/U vanish and that the Euler
number of G/U equals ord TF(G)/ord W(U). Thus dim 0 H"(G/U, Q) =
ord W(G)/ord W(U) which completes the proof in virtue of 2.5.
REMARK. AS in the case of G/T, Theorem 3.6 follows immediately from
2.5 if H*(G/Ut Z) has no torsion.

213

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
4. The classifying space of a compact connected Lie group.
4.1. Completions of modules. We shall summarize here some known results
of commutative algebra which we learned from J. P. Serre. These results are
needed in the sequel. For references see Zariski and Samuel, Commutative
algebra, Van Nostrand, and [13, Expose* 18 (Godement)].
Let A be a Noetherian ring, a an ideal of A. We give every finitely generated
A -module M the topology defined by the submodules a*M. The completion
of M for this "a-adic topology" is by definition
iff = lim M/anM (inverse limit).

(i) Let N be a submodule of M. Then the a-adic topology of N coincides with

the topology induced on N by the a-adic topology of M.
This is a consequence of the lemma of Artin-Rees which says that there
exists a positive integer h such that (anHf) AiV = an~h((ahM) C\ N) for n ^ h\
see [13, Expose 2, The"oreme 2].
(ii) Let 0 —> N —* M —* P —+ 0 be an exact sequence of (finitely generated)
A-modulcs) then

is exact. Thus "completion" is an exact functor [12, Chapter II, §4].

PROOF. We have the exact sequence
0 -* N/(anM H N) -> M/auM -* P/oT - • 0.
By (i), $ is the inverse limit of the first inverse system in this sequence.
Since N/(an+kM HiV)-> N/(anM Pi N) is onto for all n and all k > 0, this
inverse system satisfies the "Mittag-Leffler condition.1' According to the forth-
coming book of Dieudonne-Grothendieck (Complements to Chapter 0) the
assertion (ii) follows. There is, of course, a direct proof along the lines of [11, §3].
(iii) Let B be a commutative ring, G a finite group of automorphisms of B and
let A = Ba be the subring of those elements of B which are invariant under all
automorphisms of G. Assume B is, as an A0-algebraf finitely generated over a
Noetherian subring Ao of A. Then B and A are Noetherian and B is a finitely
generated A-module.
PROOF. Since .40 is Noetherian, B (as a quotient ring of a polynomial ring
over .40) is also Noetherian. Tf x t B, then Yl»<o (z — <?(z)) = 0. Thus x
is integral over A. Let x,, • • • , x. be generators of B over Ao. Then we have
equations
x\ + anxr1 + • • • + o.-« = 0, an t A, q = order of G.
Thus B is generated as an A -module by the monomials xp • • • xZm (tn, ^ q — 1),
hence is a finitely generated ^-module. Let A' be the subring of A generated
over Ao by the a,,-. The ring A' is Noetherian since it is a finitely generated

214

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
. B is even a finitely generated 4'-module. Thus also A is a finitely
generated A '-module. If c,, • • • , ck are generators of A as module over A',
then the cr and the a,, generate A as A0-algebra. Hence A is a Noetherian
ring,
(iv) We make the assumptions of (iii). Let b be an ideal of B which is stable
under G (a(b) = bforazG). Put a = b C\ A. Let b' = a-B be the ideal of B
generated by a. Then there exists a positive integer n such that bn C bf C b. Thus
b and V define the same topology on B.
PROOF. In a noetherian ring, to prove that a power of the ideal b is contained
in b', it is enough to show that all prime ideals p containing b' also contain b
(see for example [13, Expose 2]). Let p be a prime containing b' and let x z b.
Then x' = I I , . * <r(x) z A H b = a C b'. Hence x' z p. Hence there is a a-
with cr(x) z p and thus x t o-'^p). Hence b is contained in the union of the
prime ideals <r(p), a t G. But it is an easy lemma (see Northcott, Ideal theory,
Cambridge Tracts, pp. 12-13), true in any ring, that if an ideal b is contained
in the union of a finite number of prime ideals, it is contained in one of them.
Thus in our case, b C <Kp) for some <r z G. But b = <r~l(b) by assumption.
Thus b C P as contended.
We consider A and B both as A-modules and complete them with respect
to the a-adic topology. We have a map A —> B which is injective by (ii). In
view of (iv) B is also the completion of B with respect to the b-adic topology
of B. The group G operates naturally on B. Let (B)G be the ring of invariants,
(y) Under the preceding assumptions the map A —• B maps A (bijectively)
onto (Bf. Thus (Bay = (B)°.
PROOF. Let B(G) be the ring of all maps from G into B. This is a direct
sum of g copies of B where g is the order of G. We consider the exact sequence

where a(b)f b z B, is the map which attaches to <r z G the element 6 — a(b) z B.
All rings in this exact sequence have to be considered as A-modules (A = B°).
We complete them with respect to the a-adic topology. "Completion" is an
exact functor, hence (B((?))" = B(G) and the resulting sequence

is exact which proves (v).

4.2. The representation ring of a compact Lie group. Let G be a compact
lie group. Let (p,, p2, • • •) be the (equivalence classes of) irreducible complex
representations of G. Let R(G) be the free abelian group generated by the p,.
The tensor product of representations makes R(G) into a ring which we shall
call the representation ring of G. The complex representations of G may be
identified with the elements X) W<P< °* R(@) where the n{ are non-negative
integers.
Let e: R(G) -» Z be the "augmentation homomorphism" obtained by attaching
to each representation of G its dimension. Let I(G) be the kernel of e; it will

215

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
be called the augmentation ideal of R(G). We define the completed represen-
tation ring with respect to the /((r)-adic topology:
lim R(G)/I(G)n (inverse limit).

Let G, H be compact Lie groups and G —• H a homomorphism; then we have

an induced homomorphism R(H) —> R(G) which maps I(H) in I(G) and is
therefore continuous with respect to the /(F)-adic topology of R(H) and the
/((?)-adic topology of R(G). It induces therefore a homomorphism R(H) —>
R(G). Suppose now G = H. Then any automorphism of G induces auto-
morphisms of R(G) and of R(G). An inner automorphism induces the identity.
If G is connected and T a maximal torus of (?, then the Weyl group W(G) is
a group of automorphisms of T and thus operates also on R(T) and R(T).
4.3. The completed representation ring of a torus. Let T be a torus. We
write it as the group offc-tuplesof reals mod 1. Every irreducible represen-
tation of T is 1-dimensional and given by a homomorphism
(xlt • • •, xk) -> exp
of T into £7(1). TheringR(T) may be identified with that subring of the ring
of formal power series C[\xl7 • • • , xk]] which is generated over Z by
exp (2*tzi), exp (—2*ia?i), • • •, exp (2irtx*), exp (—2irizk).
Hence ^(T) t* Noetherian.
Let «j, • • • , zk be indeterminates. We give the polynomial ring Z[zu
the (2i, • • • , z*)-adic topology, and define a ring homomorphism

by setting ^(z,-) = exp(2ntx}-) — 1. Then <f>(Zj) t I(T), thus <f> is continuous

and induces a homomorphism tj> of the completed rings.
PROPOSITION. The homomorphism

is bijective. (Z][zu • • • , zk}] is the ring of formal power series.)

PROOF. Put A = Z[zu • • • , zj. Under <£ we may identify A with a subring
of R(T). For the latter ring we may write
RCD = z[zu-.., Zk) a + Zlrii..., a + **n.
We have then I(T) = fe, • • • , zk, (1 + ZiY1 - 1, • • • , (1 + zkyx - 1). Thus
the ideal I{T)n of R(T) contains only formal power series with lowest term
of degree ^ n. Thus I(T)* O A contains only polynomials with lowest term
of degree > n. Therefore, I(T)n r\ A C (2,, • • • , 2*)*- Clearly, (*„ • • • , zk)n C
I(T)nr\A. Thus
KIT H i = fe, • • •, zk)* = (7(T) H A)\
This shows that the (zlf * • • , 2jt)-adic topology of A coincides with the topology

216

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
induced from the embedding of A in R(T). Thus 4> is injective. Since {
contains R(T), the map $ is surjective.
NOTE. We have just considered R(T) as subring of R(T). This is all right,
since R(T) is Hausdorff in its /(!F)-adic topology. In fact, (\ I(T)n = 0,
since an element of this intersection would be a power series whose lowest
term has an arbitrarily high degree.
4.4. The completed representation ring of a compact connected Lie group.
Let G b e a compact connected' Lie group and T a maximal torus of G. The
Weyl group W(G) operates on R{T)\ see 4.2. We have a ring homomorphism
R(G) —> R(T) (by the restriction map) which is injective. R(G) maps (bi-
jectively) onto the ring of invariants of R(T) under the action of W(G). This
classical result follows from the fact that the highest weight of an irreducible
representation has multiplicity one (compare [5, §3.4]). We denote this ring
of invariants by R{T)W{0) and identify R(G) with it. We have the situation
of 4.1 (Hi). Ao is here the ring Z of integers. Thus we know that R(G) is
Noetherian and that R(T) is a finitely generated module over R(G).
W(G) operates naturally on fi(T) and we have an induced map fi(G) —> tt{T)
(see 4.2).
THEOREM. Let G be a compact connected Lie group, T a maximal torus of G.
Then R(G) -> R(T) maps tt(G) bijectively onto (^(T 7 ))*^, the ring of invariants
of W(G) in R(T).
PROOF. We are exactly in the situation of 4.1. Here R(T) plays the role
of B, R(G) of A, W(G) of Gy and Z of Ao. The ideal b corresponds to I(T),
the ideal a to I(T) n R(G) = /((?).
NOTE. R{G) is Hausdorff, since (\ I{G)n C (\ I(T)n = 0. The homo-
morphism R(G) —> ft(G) is injective. This is in general not true if G is not
connected (Atiyah, Characters and cohomology, in preparation).
4.5. Let X be a space belonging to the class % of 1.1. Let £ be a principal
<?-bundIe over X where G is a compact Lie group. £ induces a ring homo-
morphism

in the following way. Consider a representation of G viewed as a homomorphism

p : G —* U(m). Then p(£) is a principal C/(m)-bundle and defines an element
«i(p) of K°(X). Since the (equivalence classes of) irreducible representations
are free generators of the additive group R(G) the homomorphism cx€ is well-
defined.
If we have a map / : Y —* X (Y, X t H), if £ is a principal G-bundle over X
and r\ = /*£ the principal (r-bundle over Y induced from £ by /, then we have
the commutative diagram

(1)

217

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)

If Y consists of a single point, then K*(Y) ^ Z and a, is just the augmentation

c : R(G) -> Z. This shows that the ideal ./(G) is mapped by at into K\(X)
(see 2.3 Remark). By 2.6 (5) there exists an n0 such that <*{(/(£)") = 0 for
n ^ n0. Since $ ( ( J ) is the inverse limit of the R(G)/(I(G))n with n ^ n0, we
have a natural ring homomorphism

Obviously, at is #(G) --• £(G) followed by &t.

If we have as before a map / : Y —> X, then we have the commutative diagram

4 6. Classifying spaces. Let F be a contravariant functor on the class H

(see 1.1), i.e., F attaches to each X t 21 an algebraic object of a given type, say
an abelian group for convenience, and for each continuous map / : Y —» X
(Y, X z 21) there is given a homomorphism /* : F(X) --» F(F) satisfying the
functorial properties and the homotopy axiom (/* = #* if the maps /, g : Y —* X
are homotopic).
Let (J be a compact Lie group, BG its (infinite) classifying space determined
up to homotopy type. We shall define 3(B0) to be an algebraic object of the
same type as all the F(X), X t 21. The definition will be such that an element
of $(Ba) is completely given by the group G. The classifying space Bo is not
needed for the definition, but we write 3(Ba) rather than 5(G) to avoid the
confusion with F(G).
DEFINITION. An element a of 5(BC) is an operator which attaches to each
X and each principal G-bundle £ over X an element a(£) £ F(X) depending only
on the equivalence class of £ such that the following holds: for a map f : Y —> X
(Y, X t 21), a principal G-bundle { over X and the principal G-bundle /*£ over Y
induced from £ by /, we have a(/*{) = /*(«(?))• Using the notation of [15, §3]
this means that the diagram

/• : F(X) -* F(Y)
is commutative.
If U% G are compact Lie groups and p : U —> G a homomorphism, then we
have the induced homomorphism

For a t J ( 5 O ) , p*a : H\X} Uc) -> F(X) is the composition Hl(X, Ut)
X a
H (X, Gc) -> F(X).

218

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
If U = G and p is an inner automorphism of Gt then p* is the identity since
p : Hl(X, Gc) -> # l (X, Ge) is the identity.
According to the classification theorem [18, §19] we can choose a principal
<?-bundle £n which is classifying up to n, i.e., *\(J£e.) = 0 for i g n, and whose
base space Bim = £» belongs to %. Let n, < n2 < n3 < • • • be a sequence
of positive integers such that dim Bni g n l + l . Then £R, is induced from £Bi+l
by a map Bni —• BB<+1 uniquely determined up to homotopy. Thus we have
a homomorphism F(Bni+t) —> F(Bni). This enables us to write 5(B0) as an
inverse limit
(3) *(*<,) S l i m F C B J ,

This isomorphism is canonical. In particular, we have:

(4) An element a of 5(BO) vanishes if and only if there exists for every n0 an
integer n <z n0 and a principal G-bundle f. classifying up to n such that a(£n) = 0.
If we take for F the ordinary cohomology theory with coefficients in some
abelian group, then $(BO) becomes H**(B0, A); see [5, §6.1]. If we take for
F the /C*-theory of 1.9 then we define the ring
3C*(£0) - X\BO) 0 J C W = 5(B0)
3Z°(B0) is the X(B0) mentioned in the introduction. In this theory we write
p1 instead of p*. The Chern character ch : X*(B0) -> H**(BO, Q) is clearly
defined.
4.7. Because of the diagrams (1) and (2) of 4.5 we have canonical ring homo-
morphisms
a : R(G) -» 3C*CB0), & : &(G) -> JC*(50).
a equals R(jG) ~> ft(G) followed by d. Of course, a and & map into 3Z°(B0).
We sometimes write more explicitly ao instead of a and &o instead of &.
Let G and H be compact Lie groups and p : G —> H a homomorphism; then
we have a commutative diagram
R(G) -»&(G) -*> 5C*(BG) A H**(B0, Q)

! I
R(H) -> R(H) A X*(BH) A #••(£*, Q).
I'1 I--
4.8 We state now the main theorem of §4 and give a corollary. The proof
of the theorem will be given in the following sections.
THEOREM. Let Gbea compact connected Lie group. Then & is an isomorphism
of R(G) onto X*(BG).
l
COROLLARY. Let G be a compact connected Lie group. Then X (Ba) = 0.
Moreover, JC*(BO) = 3C°(2?O) has no torsion and no zero divisors.

219

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
We have seen in 4.4 that R(G) is a subring of R(T) which is a ring of formal
power series over Z. Thus the corollary follows from the theorem.
REMARK. We conjecture the theorem to hold for any compact Lie group.
It holds if G is finite (Atiyah, loc. cit. in 4.4 Note).
4.9. We prove Theorem 4.8 first for the case where G is a torus T which we
describe as in 4.3 as the group of fc-tuples of reals mod 1. Let Pn be the complex
projective space of complex dimension n. Over Pn we take the £/(l)-bundle
7), whose first Chern class is the canonical generator g of H2(Pny Z) ^ Z\ see
[15, §4.2], rin is induced from i^+i by the embedding Pn —» P«+i. Let B2n be
the cartesian product of k copies of Pn. Over B2n we have the T-bundle f2n
which is the Whitney sum of the w*-(t7n), 1 ^ i ^ k, where ir, is the projection
of B2n on its ith factor. £2n is classifying up to dimension 2n. We have the
embedding B2n —» B2n+2 which induces &• from {2*+2 and which gives rise to
the homomorphism K*(B2n+2) -> K*(B2n). It follows from 4.6 (3) that
X*(BT) 9* lim K*(B2Jf

the inverse limit being taken with respect to the maps X*(l?2.+j) —» K*(B2n)
just defined.
Let us denote by xt the first Chern class of *i(i?»), i.e., £. = **(0). Then
(5) H*(B2niZ) = Z[xlt ..-
where /. + l is the ideal (x?1, • • • , xVl).
We consider the map ch o a^n o 0 of the polynomial ring Z[zu • • • , zJ into
H*(B2ni Q), see 4.3 and 4.5. It maps z, onto e*' — 1. Since e" — 1 = a;/ + higher
terms, it follows from 2.5 (iii) that «$,. o 0 maps Z[2lf • • • , zk] onto /IC*(jB2n) =
K°(B2%), the kernel being the ideal Jn+l = (z^1* • • • , 2*+1) as follows from (5).
Thus
g* Z[zlt
and
(6) 3C*(£r) ^ lim

If we identify /J(T) with Z[[zly • • • , «J] (Proposition 4.3) and X*(BT) with the
above inverse limit (6), then & : R(T) —> K*(BT) is just the natural map
Z[[zi , • • • , zk]] -> lim Z[«! , • • • , zJ/J n + i .

To prove that this map is bijective, one has to check that the (zlf • • • , zft)-adic
topology of Z[zu • • • , zk] and the topology defined by the sequence Jn of ideals
coincide. But this is easy to do.
4.10. PROPOSITION. Let G be a compact connected Lie group, T a maximal
torus of G and p : T -*G the embedding. Then the map p! : K*(BG) -> 3C*(£r)
(see 4.6) is injective.

220

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
PROOF. We first observe that there exist principal (r-bundles which are
classifying up to n (n arbitrary) and which have a compact oriented differentiable
manifold as base. This is true for G = £/(ra), since then we have the complex
Grassmannians as "universal" base spaces. An arbitrary G may be embedded
in U(m) for m sufficiently large. G has thus "universal" base spaces which
arefibredwith U(m)/G as typicalfibreand complex Grassmannians. The bundle
along the fibres is orientable, since it is an extension of a principal G-bundle
and G is connected [5, §7.5]. Hence we have constructed universal base spaces
for G with the desired properties (compare [18, §19.6]).
Let a be an element of X*(B0) for which pl(a) = 0. Then we must show
that a = 0. By 4.6 (4) and the above observation on classifying bundles, it
suffices to prove that a(£) = 0 where £ is any principal G-bundle over an arbitrary
compact oriented differentiable manifold X. Using the notation and the propo-
sition of 3.5 with Bi = X it suffices to prove that ir!a(£) = 0. But 7r!a(£) =
a(x*£), the lifted bundle *•*£ equals p(rj) where rj is a principal T-bundle. Now
l
«(P0?)) - (P a)(ri) = 0.
4.11. PROOF OF THEOREM 4.8. We have the commutative diagram

The vertical maps are injective, the upper horizontal one is bijective (4.4,
4.10, 4.9). Thus da is injective. The Weyl group W(G) as group of auto-
morphisms of T operates on 3Z*(BT) (see definition of pl in 4.6). Since these
automorphisms come from inner automorphisms of G} every element of pl3Z*(Bo)
is invariant under W{G), The operation of W(G) on &(T) and 3Z*(BT) is the
same if one identifies the two rings under &T; this follows from the diagram
in 4.7. Therefore by 4.4

P1X*(BO) C

Since p1 is injective, X*(Ba) C dQtt{G) which completes the proof.

221

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020
 (19)
REFERENCES
1. M. F. Atiyah and F. Hirzebruch, Riemann-Roch theorems for differcntiable manifolds, Bull.
Amer. Math Soc. vol. 65 (1959) pp. 276-281.
2. , Quelques th4oremes de non-plongement pour Us varUUs diffirentiables, Bull. Soc.
Math. France vol. 87 (1959) pp. 383-396.
3. A. Borel, Sur la cohomologie des espacesfibre'sprincipanx et des espaces homogenes de
groupes de Lie compact, Ann, of Math. vol. 57 (1953) pp. 115-207.
4. , Sur la torsion des groupes de Lie, J. Math. Pures. Appl. vol. 35 (1956) pp. 127-139.
5. A. Borel and F. Hirzebruch, Characteristic classes and homogeneous spaces. 1, II, Amer.
J. Math. vol. 80 (1958) pp. 458-538 and vol. 81 (1959) pp. 315-382.
g. f Characteristic classes and homogeneous spaces. Ill, Amer. J. Math., to appear.
7. R. Bott, An application of the Morse theory to the topology of Lie groups, Bull. Soc. Math.
France vol. 84 (1956) pp. 251-281.
8. , The space of loops on a Lie group, Michigan Math. J. vol. 5 (1958) pp. 35-61.
9. 1 fhe stable homoiopy of the classical groups, Ann. of Math. vol. 70 (1959) pp.
313-337.
10. , Quelques remarques sur les thtorhmes de pirwdiciU, Bull. Soc. Math. France vol.
87 (1959) pp. 293-310.
11. N. Bourbaki, Topologie gentraU, Chapter 9, 2nd ed., Paris, Hermann, 1958.
12. H. Cartan and S. Eilenberg, Homological algebra, Princeton University Press, 1956.
13. H. Cartan and C. Chevalley, G&me'trie algebrique, Seminaire Ecole Norm. Sup., Paris,
1956.
14. S. Eilenberg and N. Steenrod, Foundations of algebraic topology, Princeton University
Press, 1952.
15. F. Hirzebruch, Neue topologische Methoden in der algebraischen Geometric, Berlin-
Gottingen-Heidelberg, Springer-Verlag, 1956.
16, 9 A Riemann-Roch theorem for differenliable manifolds, Seminaire Bourbaki,
Fevrier, 1959.
17. D. Puppe, Homotopiemengen und ihre indunerten Abbildungen. I, Math. Z. voL 69 (1958)
pp. 299-344.
18. N. Steenrod, The topology of fibre bundles, Princeton University Press, 1951.
19. J. H. C. Whitehead, Combinatorial homotopy. I, Bull. Amer. Math. Soc. vol. 55 (1949)
pp. 213-245.

222

Downloaded from https://www.cambridge.org/core. Barnard College Library, on 25 Apr 2021 at 05:36:57, subject to the Cambridge Core terms
of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511662584.020