Math. Ann.

264, 283-302 (1983)

9 Springer-Veflag1983

Isoparametric Hypersurfaces
with Four or Six Distinct Principal Curvatures
Necessary Conditions on the Multiplicities

U. Abresch
Mathematisches Institut der Universifiit, Wegelerstrasse 10, D-5300 Bonn, Federal Republic of
Germany

Introduction
Isoparametric hypersurfaces, i.e. hypersurfaces with constant principal curvatures,
are in some sense the simplest examples for the theory of hypersurfaces ; however,
a complete classification of isoparametric hypersurfaces in spheres has not been
achieved. The difficulties arise from the fact that in contrast to the situation in
euclidean or in hyperbolic space the number g of distinct principal curvatures can
be greater than 2.
Cartan, who considered isoparametric hypersurfaces first, has solved the
classification problem in case ge {1, 2, 3} [CAR]. When studying minimal surfaces
of low cohomogeneity, Hsiang and Lawson remarked that any homogeneous
isoparametric hypersurface must stem from the isotropy representation of some
symmetric space of rank2 [HSI]. Takagi and Takahashi then studied the
geometry of these examples in detail [TAK] and noticed that there occur two
cases with 0 = 6 and several cases with 9=4. Quite a mass of isoparametric
hypersurfaces with g = 4 - many of them are non-homogeneous - can be obtained
from orthogonal representations of Clifford algebras [FKM].
Necessary conditions have been established by Mtinzner [M~N, MON 2] ; in
view of Cartan's results the problem is reduced to studying the cases g = 4 and
g = 6. It is our aim to prove some properties of all isoparametric hypersurfaces with
4 or 6 distinct principal curvatures. We point out that according to [MON] there
are two numbers m_ and m+, each of them being equal to the multiplicities of-0
curvatures. 2
Main Theorem. i) Given an isoparametric hypersurface in S "+ 1 with g =4, then the
pair (m_,m+) - w.r.g, we may assume that m_ <m+ - satisfies one of the three
conditions below:
4A m+ +m_ + 1 is divisible by 2~:=min {2~[2O>m_,tT~N}.
4B1 m_ is a power of 2, and 2m_ divides m+ + 1.
4B2 m_ is a power of 2, and 3m_ =2.(m+ +1).
Each condition corresponds to a topologically different kind of examples.
284 U. Abresch

ii) Given an isoparametric hypersurface in S ~+ 1 with g = 6, then:

m_ =m+e{1,2}.
Its proof is sketched in the Sects. 1 till 4. Basically we proceed in the same way
as Miinzer: instead of the (local) geometric problem we consider an associated
topological problem, which is then turned into an algebraic one by the coho-
mology functor (coefficients in 7z2); there result some algebraic equations which
finally yield the necessary conditions claimed in the theorem.
Our topological object (cf. 1.7) is a quotient underlying the setting studied in
[ M O N 2]. This change accounts for the sharper result in the end as well as for the
greater effort necessary to compute the cohomology (cf. Sect. 2); from
Proposition 2.6 on the calculations split into 4 distinct cases, called the coho-
mology types 4A, 4B, 6A, and 6B. For each type a reconstruction of the Steenrod
squares and the Stiefel-Whitney classes imposes some conditions of self-
consistency: thus the type 6B is excluded, the remaining 7/2-parameters in the
other cases are determined, and the conditions on m_ and m+ are established. The
detailed results are described in 3.9 (type 6A), in 4.5 (type4A), and in 4.12
(type 4B).
The purpose of the fifth section is to relate the topological result even closer to
the geometry of the isoparametric hypersurfaces. The Stiefel-Whitney classes of the
curvature distributions (1.3 and 5.1) and the equator spheres [1.8ii) and 5.2] are
two topics discussed there. To finish with we compare our main result to the set of
known examples (cf. 5.3).
It should be mentioned that this paper is a shortened version of the author's
dissertation [ABR], which may be consulted for detailed proofs. It is a pleasure to
thank Prof. H. Karcher for suggesting the topic as well as for some helpful
discussions.

1. Isoparametric Hypersurfaces in Spheres

and Two Topological Problems
The purpose of this section is to present those properties of isoparametric
hypersurfaces we need in the subsequent arguments. Proofs may be found in
[MONI.
1.1. Definition. A hypersurface M in a sphere S n+t is called isoparametric, if and
only if all its principal curvatures are constant.
Let
g be the number of distinct principal curvatures (= :cot~i),
/~iC T~t be the distribution of the curvature cot~0g, and
m~:= dim/~ be the multiplicity of this curvature.
1.2. Leaves
Codazzi and GauB equations yield:
i) the distribution /~iC T~t is integrable; the leaves are spheres of intrinsic
curvature 1 + cot 2 tpi. They are totally geodesic submanifolds of ~t.
Isoparametric Hypersurfaces 285

ii) the second covariant derivative of the Weingarten map S:Ti~,I~Tf'I

satisfies the following relation
DE,,e S "X = 2" De,S. (S-cot ~0i)- 1. DeS "X + (S-cot ~oi). R(X, ei)ei
ei~ (/~i)e C T~,/~r. Note that D~SIff~i= O.
The above differential equation shows that after reenumerating we have (with
indices taken modg):

iii) ~Pl= q~(~r)+ - - mod zr,

g

and m i = m i + 2.

Here ~o(~r) denotes a constant depending only on the hypersurface.

1.3. Parallel Surfaces

Let N be a unit normal vector field on /~r. We consider the projection maps
pr~ : / ~ / ~ r , pv-~expe e .N.
i) If er 1 < i<9} +71.~, then pr~ is a diffeomorphism, and the parallel surface
~ r turns out to be isoparametric with curvature cot(~oi-e ) on dpr~(/~i). As a
normal 9rent circle of M is orthogonal on all parallel surfaces M~, we have obtained
an isoparametric family. Note that pr,2. pr~l = pr~l + ~2"_
ii) If e equals some q~kmOdrt, then the subbundle E k is the kernel of dpr~, and
M~ is a focal manifold of dimension d i m M - m k ; the fibres of pr~ are the integral
surfaces of the distribution/~k" Thus we have Sink-bundles M p.... ~/~,p.
iii) A different result proved in this context is the
Proposition. Given a family f4, of parallel surfaces in S n+ l such that each member
has constant mean curvature, then each member iQI~ is isoparametric.
1.4. Starting with an oriented isoparametric hypersurface/~, we choose one cpk
and define a function f on S "+1 by setting f(Q):=coso.(e--q~k) for Q ~ r . We
extend it to a function on IR"+2 by f ( x ) : = l x l g . f ( ~ l ).

Proposition. This function f obeys the Cartan-Miinzner' s differential equations:

i) igrad f12 = g2. ix120- 2,
ii) A f = 89 g2. (m x - m2)" Ixlg- 2,
where the parameters g, m 1, m2e]N are related to the dimension of the domain by
n = -# .(m 1 +mE).
2
Remark. By 1.3iii) the converse is true, too. The level surfaces of a solution f of i)
and ii) define an isoparametric family.
1.5. Global Objects
All the properties stated so far are essentially properties on germs (of isoparamet-
ric hypersurfaces, isoparametric families, and of the functions f and f). For brevity
286 U. Abresch

we'll call homogeneous polynomials of degree g on IR"+2 which satisfy the

differential equations 1.4i) and 1.4ii) Caftan polynomials. Note that we have taken
a slightly different definition than [MON]. Now one can prove the
Proposition. All local solutions o f the Cartan-Mi~nzner differential equations may be
written as a sum o f a constant and a Cartan polynomial. Thus they may be extended
to global solutions in a unique way.
There are two immediate corollaries:
i) Any germ of an isoparametric hypersurface extends uniquely to a globally
defined isoparametric family in S ~+ 1.
ii) There exists a bijection:
{isoparametric families} ~ {Cartan polynomials}.
Finally we should list some properties of Cartan polynomials f (again we write
f : = f l S , + 1):
iii) f(S "+ 1) = [ _ 1, 1] ; the regular values of f are ( - 1, 1).
X
iv, s onnecte
v) There are exactly two focal manifolds:
~ :=y-l{-l} and P§
Altogether any isoparametric situation induces a
1.6. Bundle Decomposition o f N" + 1
Let k e { - 1 , 1 } = { - , +}.
T h e n / Q : = f - x{O} and/~k : = f - l(k. [0, 1]) are the total spaces of two fibre pair
bundles over the focal manifolds:

The maps qk and Pk are induced by the projections pr~ alon_g the normal great
circles. Furthermore (/~+,/]_) is an excisive couple such that B + u / ~ = S "+ 1 and
B + r~B_ =Pl.
1.7. Additional Structure, Provided g is Even
Under this special assumption the antipodal map commutes with f by 1.5 ; hence
it also commutes with the mappings pr~, qk, and Pk, and the standard projection
map ~ : S "+ 1 ~ , + 1 induces an underlying decomposition of the real projective
space:
M : = {[x]e F "+ l l f ( x ) =0}, h) = re- I(M),
B k : = { [ x ] e IP~ + 11k. f ( x ) >- 0}, /~k = g - 1(Bk),

Fk := {[x]e~n+ lJf(x)-k.(x,x)~ , ff k=7~- l(Fk).

We denote the canonical inclusion maps in the following way:

i : M ~ I P "+1 ik ;Fk'-'~]Pn+l bk:Bk'-*]P n+l

Jk : M "+B k s k : F k "--~Bk .
IsoparametricHypersurfaces 287

As there are also induced projection maps qk and Pk, we obtain the two bundles:

~k '(~mk +1, smk)...~(Bk,M) tqk,p~ , Fk"

Immediate Properties. i) (k = (nlFk)*(~)"
ii) The S~ tf4~M, Fk~Fk, and Bk~B~ may be regarded as liftings of
S o ~ s ~ + 1~ip~+ 1 with respect to the inclusions i, ik, and bk.
iii) (B§ B_) is an excisive couple in IP*+ t.

1.8. The Range of g

The existence of a decomposition of 5 ~§ as described in Sect. 1.6 is a non-trivial
topological problem. The cohomology turns it into an algebraic problem. Most
algebraic equations result from the multiplicative structure of cohomology.
Classifying all solutions of the algebraic problem yields the
Theorem (Miinzner). Any isoparametric hypersurface 1VI in a sphere has the
following properties:
i) ge {1,2, 3,4,6}.
ii) ge{3,6} ~ m+ = m _ e { 1 } w 2 . N (we write m : = m + =m_).
iii) g = 4 ~ m++m +liseven,
or m+ = m _ e 2 . N , or le{m+,m_}.
We point out that by Steenrod squares one could improve ii) and obtain:
ge{3,6}} =~ m m u s t b e a p o w e r o f 2.
Cartan has completely classified the cases gs {1, 2, 3} by geometric computations.
Thus g--4 and g = 6 are the only cases to be considered further on. Section 1.7
describes how to assign a sharper topological problem to them. We are going to
use the cohomology of that object in order to prove our main theorem.
Remarks. i) It is sufficient to use Z2-coefficients, as all the other information is
already contained in the cohomology of 1.6.
ii) As equator spheres play an essential role in the geometry of the focal
manifolds of the known examples, one may already expect that the additional
conditions in the underlying situation yield some further results on isoparametric
hypersurfaces.

2. Computing the Z2-Cohomology

The previous chapter has provided us with the background information we need.
We use cohomology theory and in the subsequent chapters also the theory of
characteristic classes to such an extent, as may be found in the books by [SPA]
and I-MIL] e.g. For brevity we adopt the convention that any coefficient ring is Z2,
except we explicitly state a different ring.
We start recalling the
2.1 Proposition (Miinzner). Let
i) R = Z 2 , or
ii) m+,m_ 22, and let R be an arbitrary ring with unit; then:
288 U. Abresch

R; q ~ { O ; m _ k } + ( m + +rn_).~q o and q<n

H~(ffk;R)= O; else.

Keeping in mind part i) of this proposition, we now look at the commutative

diagram of Gysin sequences :

2.2 --+ Hq(Fk) =l,.. Hq(ff k) --+ H,(FR) ~ H,+ I(Fk) ,~I~.+.' >

2.2' --~ Hq(F "+ x) ~" , H"(S"+ x)---~Hq(IP "+1) ~ ) H q+ I(Ip"+I) ~'+' )

There remains only a short list of sequences (dimz Hq(Fk))q=>o which are com-
patible with both, the top exact row of the above diagram and the 772-Poincar6
duality:

2.3 q 0 ... m k ... m~ +m ... 2m_~+m k ... 4m .. 5m

41 lc 1 lk 0 lc 1 lk 0 0 0 0
41I lc 1 2c 2 2k 1 lk 0 0 0 0
61 lc 1 lk 0 lc 1 lk 0 lc 1 lk
611 lc 1 2c 2 2k 1 2c 2 2k 1 lk
6III lc 1 2c 2 3c 3 3k 2 2k 1 lk

In this table, which describes the structure of H*(Fk) as a graded Z2-vector space, a
"c" denotes that dim coker (lp~-1)= 1, and a "k" denotes that dim ker(~o~,)= 1. The
absence of any of these letters implies that the corresponding dimension is zero.
Let w denote the non-zero element of HI(~ '"+ 1). By naturality Wk:= i*(w) is the
characteristic class such that the homomorphism ~o, is given by ~Pk(U)=U'W k. We
remind the reader that H*(IP "+ 1)=Z2[w]/(w,+ 2). Hence each of the vector space
types of H*(Fk) admits only a few structures as im/*-modules. They are generated
by 1 and ~9 - 1 other elements. H*(F+) and H*(F_) must have one of these
structures over the subaloebra imi* each. Compatibility conditions on the pairs of
im/,-module structures arise from the two long exact sequences"

2.4 ~ Hq(IP" + 1) !~ ) Hq(Fk)___. H q - m- ~(F_ k) ~ H q + 1(]pn + 1) _...~.

They result from the following commutative diagram (Tk denotes the Thorn
isomorphism of the bundle ~k):

_ . H~(lp,+ t) b_..~Hq(Bk)__~ Hq+ l(lp.+ 1, Bk)._. , Hq+ l(ip.+ 1)__0

2.5
1 1-
~ Hq(B_k) i-~,Hq(M)--~ Hq+X(B_R,M) ~ Hq+l(B_k) --~
l
l
Hq-m-~(F k).
T-k
IsoparametricHypersurfaces 289

The above considerations exclude all but four non-ordered pairs of imi*-module
structures. In the remaining cases we try to get as many relations as possible on the
products of the non-trivial generators of H * ( F k) over its subalgebra im i*. Towards
this purpose we choose a suitable normalisation and again make use of
Z2-Poincar6 duality. Thus we obtain the
2.6 Proposition. Let Xk, Yk, and z k be homogeneous elements o f degree re_k,
m+ + m _ , and 3m, respectively. 7k, 7'k, and e k will denote some parameters in Z2.
I[ g = 4 or g = 6, then as a module over im ix there are only the .[bllowing possibilities
for the graded algebra H*(Fk) :
4I H*(Fk)=~2[Wk, Yk]/(Wkm-~+l , Yk),
2

4IIa) n*(Fk) = ~2[Wk, Xk]/(W ~ ++m- + X' "~'k

. 2 --
-t- ~k" Xk" Wk~- ~ -~ 7k" '~k" Wk2m- ) '
4IIb) H * tt F k!~-7Z
- - 2 rL .'Vk~
. . . ~kJIt'Vk ....
1/(aA,2m-k+mk+l ~~k" ~Vk +1, Xk2_~_~k. Xk. Wr~k-k---t- W k2m-kx
}

61 H*(Fk)= ~ [ w k , y~q/(w~ + ~, y3),

611 n*(Fk)= ~ 2 [ W k , Xk, Zk]/(WSk m + 1 Xk " Wkm + l Zk. W~k+l ~ Zk2 '
2 m m 4ra
Xk "~k'Xk'Wk,Xk'Zk'3t'~'Zk'Wk " ~ W k ),
* F - 3m+1 3 2m 3m
6III H ( k)--7/2[Wk, Xk]/(Wk Xk+~k'XR'W k +ek'W k ).

N o t e that the homomorphism i x is given by iX(w )= w k. Furthermore the types of

F+ and F_ can be matched only to the four combinations (non-ordered pairs) :

4A : = (4IIa, 4IIa), 4B : = (4I, 4IIb),

6A : = (6III, 6III), 6B : = (6I, 6II).

Remarks. i) To distinguish the types 4IIa and 4IIb, it is in general not sufficient to
know the isomorphism class of H*(FR) as a graded algebra. The Clifford series
contain explicit examples for the subalgebra imi* playing the essential role. In this
context we observe that the geometries of certain subseries (m 1e { 1, 2, 4, 8}) are of
type 4A and 4B in turn.
ii) In a more technical way it had been possible to do the proof completely
without relating our problem to the covering situation via the Gysin sequences 2.2.
Instead one could study the images of products under i"* k and under the
homomorphisms Hq-m-k(F_k)-~H ~+ l(~n+ 1) in the sequences 2.4.
iii) The Alexander duality theorem will not give any additional information, as
there is a commutative diagram:

H
~-mk
(Fk)
Tk
~ 'Hq+
I(B" M[
" ',' "
.~
de,or-
Hq+1(lpn+ l
~+~ I (F+uF))
F_k,F
[ Poincar6 matron [
[duality ~ excision
[ /
H n _ ~(Fk ) ~ Alexander duality H q + l(len + 1, Ien + 1 __ F k ) .

Now it is easy to compute H * ( M ) as a graded Z f v e c t o r space from the Mayer-

Vietoris sequence :

2.7 --~ Hq(IP"+ 1) (~+,~-~ H~(F +)~Hq(F_) (p~,-p~-), Hq(M) aq ~Hq + l(lpn+ 1)___~,
290 U. Abresch

which may be decomposed into (split) short exact sequences, as we know the maps
i* in all four cases.
Another tool is given by the Gysin sequences of the sphere bundles ~k" As they
stem from the commutative diagrams:

2.8 ~ H~(M) ~ ' H q+ l(Bk, M) ~§ , H q+ l(Bk) J~§ Hq+ X(M)--~

-~I Tk
H q- mk(Fk) H + (Fk),
the homomorphisms tp* *
k are given by tPk(U)=U" com~+l(r Here
o~,.~+l(~k ) ---(qk)
* - l z * k(Uk)--
_ Tk-l(U2)
is the top Stiefel-Whitney class of the bundle ~k, and U k = T~(1) denotes its unique
Z2-orientation class.
If we now determine kerp"_~ k+ t by sequence 2.7, we obtain the

2.9 Proposition. i) I f F k is of type I, then ~o,,_k+ l(~_k)=W "-k+l

ii) I f F k is not of type I, then com_~+l(~_k)=0, and thus H*(M) is a free module
over H*(F_k) via P*k with basis 1, Pk (Xk).
Remark. The reader may compare the diagrams 2.5, 2.7, and 2.8 in order to realize
why these computations cannot restrict the algebras H*(Fk) any further than the
sequences 2.4 do.
In case g = 4 the above proposition determines the algebra H*(M), because the
relation for the square of the non-trivial generator p~(Xk) may be read off from
Proposition 2.6 immediately. In case g = 6 we may find out that relation by
applying the 7/2-Poincar6 duality theorem on H*(M).
We also simplify our notation and write:
w instead o f w kand Pk*(w~)-
- Pk*'*
z~( w )--' *z (w),
x k instead of p~(Xk), and
YRinstead of P*(Yk)"
The context will always make clear in which algebra the element lies. Notice
that there may arise some additional relations, when one passes from H*(~ "+ 1) to
H*(Fk), or from H*(Fk) to H*(M).
Next we adopt the convention that
in the cases A notation is chosen such that m_ <m+, and that
in the cases B notation is chosen such that F_ is of type L
Then the further details of the cohomology of our topological problem
(compare 1.7) run as follows:

2.10 Proposition. An explicit description of H*(M), which of course depends on the

cohomology types we have distinguished in Proposition 2.6, runs:
4A H*(M)=TI2[w,x+,x_]/(wm+ +m- + 1,X2 +7+ "X+ "Wm- +7+ "e+ "WEra-,
X 2 "Jr-7 _ "X_ "Wm+ " ~ - 7 - " / ~ - " w 2 m + ) ,

4B H*(M)=Z2[w,x+,y_]/(w"§ x2 +7+ "x+ .w"- +w2"-,y2_),

Isoparametric Hypersurfaces 291

6A H*(M)=Z2[w,x+,x_]/(w 3m+ 1, x2+ + x + "x_ + x 2_+)~'w 2m,

X 3+ "~t-~ " X + " W 2m -~- ~." W 3m , X 3 -~- ~) " X _ " W 2m "JI- F." w 3 m ) .

Here ~:=T+ =~_ and e:=e+ =e_.

By our notation this list implies the knowledge of the algebra homomorphisms
i*, i*, and p~' (or j*, respectively). Some more details on the long exact cohomology
sequence of the pair (B k, M) (see diagram 2.8, too) are given,
i) except for k = + in the Cases B, by:
H*(B k, M) = t~*H*(M), t*k = O,
q~(v). Uk---k(pk(v)'x_k)
,~* * for all v~H*(Fk);

ii) for k = + in Case 4B, by:

H*(B+,M)4=~*H*(M), I*(U+)=q*(wm++l),
q*(x+). U+ =fi*(y_),
q*+(w2m-)" U+ =q*(x+ +T+ "w~-)'6*(y_)
=,~*+(x+ .y_ + ~+ .y_ "win-).

In this section we have established a description of the 7Z2-cohomology,

including its structure as a graded algebra. But cohomology carries even more
structure; there are the Steenrod squaring operations which permit an almost
independent calculation of the tangential and the normal Stiefel-Whitney classes
of the submanifolds F k in IPn+ ~. These tools will subsequently turn out to impose
the required restrictions on the mk. As the Case 6B will be shown to be self-
contradicting and hence be excluded, we have not given all its details in
Proposition 2.10.

3. Steenrod Squares and Stiefel-Whitney Classes in the Case g----6

As has been mentioned at the end of the previous section the principal idea is to
apply the Whitney duality formula to the tangent and the normal Stiefel-Whitney
classes of the focal manifolds Fk:
3.1 ~o(IP"+ t, Fk). 09(Fk)= lk
.. o)(lP , + 1)=(1 + W)n+2.
In the Case 6B we restrict our attention to F_. The axioms for the Steenrod
squares imply that Sq*w= w.(1 + w) and that the Wu class V2m(F_) equals y_.
Hence Wu formula asserts that o~2m(F_)=y_. On the contrary the normal
Whitney class ~o(Fn+~,F_)=o(~_) is contained in imi* (compare for the
Propositions 2.6 and 2.9), and the duality formula forces co(F_) to lie in the same
subalgebra. That is the announced self-contradiction, which excludes this case.
It remains to consider the type 6A. The basic idea stays to be the same; we may
also restrict to one focal manifold F k. Again the axioms for Steenrod squares yield :
Sq*w=w'(1 +w),
Sq* x~ -m-X k2 "{- Xk " O~k+ Wm " ~k ,
292 U. Abresch

where ~k, f l k U l m l k a r e polynomials of degree strictly less than m. By Cartan

formula the Steenrod squares of all elements of the cohomology algebra H*(FR)
may be determined, depending only on 0~k and ilk" Before proceeding, it is
convenient to notice that for 0_-<i < m the Sq i m a p the imi~-submodule of H*(Fk)
generated by 1 and x k into itself, while Sq m does not. But if m is not a power of 2,
the Adem relations give an expression of Sq" as a sum of iterated Sq i (0 < i < m)9
This yields the
3.2 Lemma. m is a power o f 2.
With the aid of Wu formula we compute:
3.3 09(Fk) ----(1 + w) 3ra + 1 9(~.w 2ra +~k2 + X k ' ~ k + W m ' f l k + X 2 ) 9

Let us now assume m > 2 ! As F k has codimension m + 1, the normal Whitney class
must look like:
3.4 r + 1, Fk ) = Zk § .~k.t~k ,

where Zk, tTk~-l"m ~k

'* satisfy: degz k --<m + 1 and dega k< 1.
Using the fact that the Sq i are natural and commute with the connecting
homomorphisms, one could read off from Proposition 2.10 that z k = c t k and o"k 1 =

and even circumvent the assumption on m. We however prefer to ignore these little
restrictions - ~t_ k isn't known, too - and head towards the
3.5 Proposition. I f a submanifold o f ~6,,+ 1 has a cohomology algebra o f type 6III,
then m < 2.
Instead 9
Proof 9 We insert 3.3 and 3.4 into the duality formula 3 9 and comparing
coefficients with respect to the basis 1, x k, x k2 over 1m/k,
9 "~ we obtain the following
system of equations :

0 = ~k.(~2 + w"./3~ + ~. w2")+(1 +w) am+l +~.ak.w 3,.

62 ra
3.6 0=trk'( k § " f l k ) § 7~k'O~k

O=Zk § k9

We conclude that z k, trk, and 0~k are units, while fig = 0. Hence instead of the above
system we m a y as well consider the equation"
3.7 O=O~k.(~2k + ]).w2rn)§ e.wam § § w)am+ 1. t7k- 1

Notice that the w-adic Newton algorithm determines the unit 0~k in imi*, as soon as
one has chosen y, e, and trk. The case t r k = l § gives 0~k=(l+w)mmodw 2m in
contrast to the requirement on degct k ; therefore only the possibility o k 1 remains, =

and it is easy to check that the unique solution of 3.7 is given by :

3.8 0~k_~_(l § w)ra + c § y. w2m.(1 § W ) 2 c - r a - 1 § l?,.w3ra ,

where the integer c equals ~-(4m+ 1) or al.(8m+ 1), depending on the residue class
o f m modulo 3. Looking at the degree o f ~ k in the formula 3.8, we finish our proof,
and furthermore we get the values of ), and 5.
Isoparametric Hypersurfaces 293

Now we collect all the details obtained so far, and we arrive at the
3.9 Result
Given an isoparametric hypersurface with 6 distinct principal curvatures, either all
their multiplicities equal 1, or all their multiplicities equal 2. The 7Z2-cohomology
of the associated decomposition of F 7 or ]p13 (compare for 1.7) runs:
degxk=m, d e g w = 1, Sq*w= w.(1 +w).
m = 1" H*(M)=TZ2[w,x+,x_]/(w4, x 2 + x + " x _ + x _2, x + , 3x _ ) 3

H * ( F k ) = subalgebra generated by w and x k

Sq* x k = x k. (1 + Xk)
~O(Fk) = 1 + X k + X~

wOW, FR) = 1 + )Ok "~

m=2: H*(M)=Z2[w,x+,x_]/(w7, x2+ + x + . x _ + x 2_ 4 - w 4,
X3+ "-[- X + 9w4 +w 6 ,x_
3 + x _ . w 4 + w 6)
H * ( F k ) = subalgebra generated by w and x k
2
Sq* (X: k = X k "~- X k " W -~- X k

09(Fk) = (1 + W) 7. ( 1 + W 2 + w 4 + X k + X k 9W + X 2 )

o~(F13, Fk ) = 1 + w + x k .

In either case both the f o c a l manifolds are non-orientable.

There is known one 9 e o m e t r y for e a c h case. Therefore, as far as our main
theorem is concerned, there is no point in checking any further conditions.

4. Steenrod Squares and the Results for g = 4

In principal the idea stays to be the same here, too. But the analogue of the system
3.6 will contain much less information. In order to cope with that fact, we shall
consider both focal manifolds. The axioms for the Steenrod squares assert that in
each F k :
Sq*w=w.(1 +w)
4.1 S q * Xk ~- Xk" ak q'- Wm - k " flk

Sq*yk=yk'O~k (in Case 4B only),

where O~k,flkelml k a r e polynomials of degree <=mk, O~k is a unit, fig is a zero-
divisor, and ))k is the coefficient of wm-k in a k.
Furthermore here it is not only for convenience but essential in proof to make
use of the equations arising from the requirement of self-consistency. Again 4.1
enables us to compute the Steenrod squares of all elements in the cohomology
algebras as functions of the coefficients of 0~k and ilk.
294 U. Abresch

In each H*(Fk) we require the A d e m relations to be valid m o d imi*, and thus

we get the

4.2 Lemma. For each polynomial ~k there is a non-negative integer a k such that:
ctk = (1 + W ) ok 9

Obviously a k may be chosen in a way that it is strictly less than 2 . d e g x k or 2 - d e g y k,

respectively.
F r o m n o w on we treat the Cases 4A and 4B separately. We start with the type
4A. The crucial relation is:
4.3i) (l+w)m++m-+l=(l+w) a§ modw m*+m-+l,
which we also write in the form:
4.3ii) ( l + w ) ~§ +~- + l - a k - ~ , m o d w m++m-+l

This relation can be derived in several different ways:

i) Steenrod squares are natural c o h o m o l o g y operations and c o m m u t e with the
connecting h o m o m o r p h i s m s Aq of the Mayer-Vietoris sequence 2.7. We apply Sq*
to d*(x + 9x _ ) = w m+ + m_ + 1, and c o m p u t e each term separately.
ii) We m a y use the duality formula 3.1 for one of the three manifolds M, F+, or
F _ , and calculate the tangent Whitney class with the aid of W u formula. We k n o w
that ~o(IP"r 1, M) = 1, or we c o m p u t e ~o(IP"§ 1, Fk ) from the data given in Proposition
2.10.
L o o k i n g at 4.3ii), the requirements on the degrees of the ~k yield:
4.4 ak=O or ak=m_ k or m r + m _ <ak < 2 m _ k.
N o w 4.3i) immediately imposes some restrictions on the m k, and we arrive at the
4.5 Result
Let 2 K be the smallest power of 2 which is strictly greater than min {m+, m_ }. The
bundle decomposition of IP"+1 which is induced by an isoparametric hypersurface
with 4 distinct principal curvatures m a y be of cohomology type 4A, if and only if
2~l(m+ + m _ + 1). Let m_ < m r (w.r.g.), then a detailed description runs:

m+
H*(Fk) = subalgebra generated by w and x k
degw=l and degxk=m_ k
Sq*w=w.(1 +w)
Sq*x+ = x + + w ~- "fl+
S q * x _ = x _ "(1 + w ) m§ +x +w~§ .fl_,
where flk is an appropriate element of 7Z2[w]/(w mkr 1).
~o(F_) =(1 + w)2'n" + 2m- + 2 ; o)(IW+ 1, F _ ) = 1

o)(F+) = mOP"+ 1, F + ) =(1 + w)"+ + ' - + 1

Isoparametric Hypersurfaces 295

Both the focal manifolds are orientable.

We should point out that the Adem relations may also be used to restrict the
fig. Thus one obtains the formula:
flk =r/k, 1"( 1 +(1 + W)m-k)
+ r/k,2"((1 + W)~k+(1 + W)-X~)+ r/k,3. W~.(1 +W)m-k,
where 2k is the greatest power of 2 which divides m_ k, and vk is the greatest power
of 2 which does not exceed m R. The three Z2-parameters qk, X' r/k,Z' and r/k,3 are not
free, but they must also satisfy certain conditions depending on m+ and m_.
Obviously all the parameters equal to zero is a valid combination.
It remains to treat the Case 4B. We again apply Sq* on both sides of
A*(x+ .y_) W2m- +ra+ + 1 and obtain :
=

4.6 ~+.~ = ( l + w ) 2"-+m§ modw "§

When here we define the Sq ~by 4.1 and the Cartan formula, there are even more
relations to be checked for self-consistency. The identity x + 9w" § + 1= 0 accounts
for the
4.7 Lemma. i) fl+ = w ~-, and
ii) m_ is a power of 2.
We apply Sq* to each term of the identity xZ+ +7+ .x+ .w"- + w 2-- = 0 and
compare coefficients, keeping in mind that H*(F+) is the direct sum of the
component generated by 1 over imi* and the one generated by x+. This yields:
7+.w"- = ( l + 7 + . w " - + ~ + ) 2 modw "++I
4.8 0 = 7+'(1+w'-+~+) mod~e " §
7+=1 ~ rn++l<2a-m_.
Now we are in a position to compute the normal Whitney class ~o(IP"+ 1,F+),
although U+ is not contained in im6* (see 2.10). Towards this purpose we look at:
q*(x+ .~_). U+ = 6 " Sq*y_ = q * ( S q * x + ) . S q * U+
=q*(x+).q*(ot+ +7+ -w"- + x + ) . S q * U+.
Hence in H*(F+)/(kernel of the multiplication by x+) we have:
co(iP.+ 1, F + ) = ~ _ -(~+ +7+ .w"- + x + ) - 1.
A detailed analysis of this relation using 4.6, 4.7, and 4.8 yields
4.9 eo(~.+ 1, F+) =(1 + w)m+ + 1 .(1 + x + .~+ 1).

By Proposition 2.9 the very last formula is not only true in the quotient
H.(F+)/(w,.§ 1) but also in the cohomology algebra H*(F+). Again in view of
Proposition 2.9 this shows that m_ < m+, and then by 4.2 and 4.8 the values of ~+
may be determined:
y + = 0 =~ ~ + = 1
4.10
y + = l =~ 0 c + = ( l + w ) 3c,
where c: = 89 m_ tIN.
296 U. Abreseh

It remains to insert them into 4.9 and again to check whether

deg(w m§ + 1 + ~o(lP"+ 1, F +)) =<m +.

This gives the desired restrictions on m+.

For the sake of completeness we compute with the aid of Proposition 2.10 that:
4.11 coOP"+ X,F_)= T_- l(Sq * U ) = ~ + .
The tangent Stiefel-Whitney classes may now be calculated from the duality
formula 3.1 or as well with the aid of Wu formula. Putting all the information
together, we get the
4.12 Result
The decomposition of IP"+ 1, which is induced by an isoparametric hypersurface
with 4 distinct principal curvatures (see 1.7), may have cohomology of type 4B, if
and only if it is covered by one of the two subcases described below :
4Bl: 2m_l(m++l) and m_ is a power of 2.
H * ( F _ ) = 7/a [w, y_]/(w m+ + 1, y2 )
H.(F+)=7Z.2[w,x+]/(wZm- +m+ + 1, X + .W,,+ + 1,.~+.
Z + warn-)
H * ( m ) = 7 Z z [ W , x +, y_ ]/(wm+ + l, y _ , x- 2 ..~..w2m-)

Sq*x+ = x + + w 2"- Sq*y_ = y _ .(1 +w) m+ + 1+2m-

co(F_) = ( 1 + W)2m++2m-+2 tO(F"+ 1, F _ ) = 1
t o ( F + ) = ( l + x + ) . ( l + w ) m++l ~o(IP"+1, F+) = ~o(F+).

Notice that F_ is orientable for all admissible multiplicities, while F+ is

orientable if and only if m_ 4:1.
4B2: 3.m =2.(m++1) and m is a power of 2.
H * ( F _ ) =712[w, y_ ]/(w m+ + a, y2)
H*(F+)=TZ2[w,x+]/(wZ"-+m++I,x+ "w"+ +1 ,:~+
.2 + x + . w m - + w 2m-)
H * ( M ) = 7/2 [w, x +, y_ ]/(w m§ + 1, x 2 + x +" w"- + w 2"-, y2 )
Sq*x+ = x + "(I+W)m++I+w 2m- Sq*y_ : y _ . ( l + w ) 2m-
~o(F_) = (1 + w) 2m- +" ++ 1 o~(ip,+ 1, F _ ) = (1 + W) m + + 1
o~(F+)=(l+W)m§ "- ~o(F"+l,F+)=(l+w)"++l+x+.

In particular both the focal manifolds F k are orientable for m_ >4, while
they are both non-orientable in the case m+ = m = 2.
In both these subcases the grading is given by:
degw=l, degx+ = m _ , degy_ = m + + m .
Remark. For the cohomology type 4B the Adem relations also impose conditions
on fl+; but these equations are satisfied by the value of fl+ which we have
calculated, and hence there are no further restrictions on the multiplicities which
may be obtained from this idea.
IsoparametricHypersurfaces 297

Finally we should point out that the sets of admissible (non-ordered) pairs of
multiplicities, which can be read off from 4.5 and 4.12, are disjoint. Therefore we
may use them to distinguish the types 4A, 4B 1, and 4B2. That finishes the proof of
the main theorem stated in the introduction.

5. Conclusions
It is one aim of this section to give a more direct interpretation of the topological
results, which have been computed in order to prove our main theorem, in terms of
the geometry of isoparametric families. Furthermore we want to compare the
theorem to the known examples in order to get an idea how sharp our necessary
conditions are. To start with we calculate the
5.1. Stiefel-Whitney Classes of the Curvature Distributions
/ \

We remind the reader that for each principal curvature c o t ( ~ + i . - ~ ) o f the

isoparametric hypersurface M there is a projection mapping p i : M ~ F i (ie7to),
which is induced by Pr(2i+l).~/2g such that kerdp~ is the distribution EiC T M (cf.
1.3). W.r.g. we may choose the orientation of the normal bundle N M on M in such
a way that
~F+; if i=0 mod2 (i.e." p+ =po)
F~=[F_; if i=1 rood2 fi.e.'p.=p_l).
I m m e d i a t e Consequences. i) o~(F i) = o~(N M O E I) = o~(p*( N F i)) =p*co(lP"+ 1, F i)
ii) co(IP"+ ~) = ~o(TM) = ~ ~(Ei),
iii) p * ( w ) = w,

iv) kerp* = {0 in the Cases A

w ~+ + 2 . H , ( F i ) ; in the Cases B.
With the Results 3.9, 4.5, and 4.12 in mind these properties are sufficient to
obtain the parts concerning the case g = 4 in the following

Theorem.
4A : ~O(Eo)= co(E2) = (1 + w) m ++m- + x
t~(E1) = co(Ea) = 1
the compatibility of Sq* with p* as well as p* yields: Po*-P2--
*
4B1 : cO(Eo)=co(E2)=(1 + x + ) . ( 1 +w) m+ +
o~(E1) = oJ(E3) = 1
it can be read off that p ' x + = x + , hence: P o - P 2 .
4B2 : ea(Eo) = (1 + w)m++ 1 + x+
~(E,) = co(Ea) =(1 + w)'~+ + 1
co(E2) =(1 + w) m+ + l + x + + w m-
298 u. Abresch

it can be read off that p'x+ = x + + w ~-, hence: p~ 4:p*.

6A: ~o(E o) = (1 + w ) ' - 1 + x + r = ( 1 + 09)m- 1 +X_
~(E2) ---(1 + w) ra - 1 + x _ co(E1) = ( 1 + w ) m - 1 + X+ + X _

co(E4) = ( 1 + w)m- 1 "~-X+ "~-X_ o9(E3) --(1 q- w ) m - 1 d-x+

all the p* must be distinct.

+ +

Fig. 1 -t- "--1---' +

To prove Part 6A, it is necessary to consider the sphere bundles

~m f t ~M P' ~ F i

in some more detail. We have depicted the geometry along a normal great circle.
The points of intersection with one of the focal manifolds F§ or F_ are marked
with a " + " or a ~ respectively. The thick arcs are the intersections with/~+, the
other arcs are the intersections with/~_, prn/6 interchanges the end points of each
thick arc (they are th e points of intersection with M), while pr_ ~/6 interchanges the
end points of each thin arc.

Observations
i) There are induced isometries p r , . ~ / 6 : M ~ M (#eTZ6) in the quotient; they
constitute the dihedral group D3, which is generated by the involutions pr~/6 and
pr_,,/6 for instance.
ii) pr~/3 = pr n/6 .pr~/6 induces F,-bundle homomorphisms:

(Pr~/3), : Ei--~ E i - 2.

iii) P4* - - Pr2~/a

* "Po* P3* = Pr2~/3
* "P5*
9 -- * * *-- * *
P2 -- Pr2n/3 "P4 Pl -- Pr2~/3 "P3
A* =fo*- prL/3 f? =f~"prL~3
f,* =f~'" pr~/3 f~ =f?"prL/~ 9
Especially all p* are injective and all fi* are surjective. Notice that ,j~..p~ = 0.
Isoparametric Hypersurfaces 299

iv) Let A denote the antipodal map of S". There are the following com-
mutative diagrams :
TS" ~. , TS m TS ~ A, ~TS ~

Eo (prn/6), ~Eo E5 (pr-~/6), ~E5

Lemma. Let "A~utn*( M) be the sub#roup {re A ut n*( M)lz( w) = w }. Then ~A-'~H*( M)
is the dihedral ffroup D 3 #enerated by:
z~ : w ~ w and "C2 ~Wb--*W
X+ I-+X+ X+ b"+X+-}-X_
X_~+X+ +X_ X_b-+X_.
Proof. With the aid of 3.9 it is easily verified
i) that ,:, z2eXfi-tH*(M ), and
ii) that any , e AutH*(M) which induces the identity on Hm(M)/7Z2 9w m must be
the identity itself because of the relation :

Notice that the claimed expressions for og(Eo) and co(Es) are immediate from
3.9. From observation iv) we may now deduce that:
pr*/6e{id, zl} and pr*~/6e{id, z2} ,
hence:
9 9 9 9 . .
(pr~/6, pr_~/6, Pr2~/3)e {(ld, ld, ld), (~ 1, ~ 2, ( ~ . z 2 ) 2 ) }

For both these possibilities we may calculate all the Stiefel-Whitney classes co(Ei),
then check whether the relation co(TM)= I-I c~ holds, and thus finish the proof
of the theorem.

Remarks. i) This sort of computation also yields the homomorphisms fi* and p*.
ii) It is possible to carry out the same argument for isoparametric hyper-
surfaces/~ C S "+ 1 and the distributions/~i C T/~r. In fact in case O= 3, me { 1, 2, 4, 8}
it turns out that the classes ~o,(/~i) do not vanish, and thus the bundles/~i and N F i
as well are nontrivial.
iii) In the case 9=6, m = 2 we also change and use the Z-cohomology of
MI2CSx3, which has been computed in [ M U N 2 ] . Then the special isometries
{pru.,/61l~eZ12 } and AutH*(M) constitute two copies of the dihedral group D 6
instead of D 3. Making appropriate choices for the orientations of the/~, we obtain
expressions for the Euler classes e(/~). In order to avoid a lengthy description of
Miinzner's calculations, we apply the e(/~) to the homology classes represented by
the fibres S m of the bundles
Sm fJ ' M PJ >F~,
300 U. Abresch

and state the result as a table for the fpe(ff~i)IS"] :

, j ~ 0 1 2 3 4 5

0 2 I -1 0 -1 -1
1 3 2 -3 -1 0 -1
2 -1 -I 2 1 -1 0
3 0 -1 3 2 -3 -1
4 -1 0 -I -1 2 1
5 -3 -1 0 -1 3 2

5.2. Equator Spheres

In the known examples with 4 distinct principal curvatures equator spheres play a
somehow crucial role in the geometry of the focal manifolds ffk" In our setting they
are blown down to some real projective spaces IP"k of dimension r k >-.mk, which are
embedded into F k in such a way that the compositions IP'k~Fk ' '~ ,IP "§ are
linear embeddings. Although these facts provide an intuitive understanding why
the key idea of identifying antipodal points works and proves our main theorem,
there are even in the case g = 4 no more restrictions on the multiplicities, which
could be obtained with the aid of the equator spheres. But vice versa our results -
eventually after applying the Whitney duality formula to the embedding IPr~ ~ F k -
give rise to some upper bounds for the rk:
4A: r+_--<m+ + m _ r__--<m++ m _
4Bl: r+ <m+ +m_ r_ <m+
4B2: r+ < 2 m _ - 1 r_<m+
6A: r+<3m r_<3m.
5.3. The K n o w n Examples
6A: m = 1 (92, so4)
m=2 (g21~)g2, g2)

4B2: (m_,m+)=(2,2) (sos~)sos, sos)

(m_,m+) =(4, 5) (SOlo,Us).

All four geometries have transitive isometry groups. Each of the examples is
constructed out of the isotropy representation of the symmetric space listed behind.
A classification of all isoparametric hypersurfaces whiqh may be constructed out
of such isotropy representations is carried out in [TAK].
4A, 4BI : The Clifford series.
There examples may be obtained from orthogonal representations of the Clifford
algebra C1~ ml + x : let P . ~ SO (F." + 2), 0 _-<p < m 1, be matrices such that p2 = id and
P~.P~ + P~.P. = 0 for # 4: v; obviously the P. are symmetric and the function
f(x):=(x,x) 2- ~, ( P . ( x ) , x ) 2 ; Xe~x n+2
#~_o
satisfies the Cartan-Mtinz'ner differential equations (cf. 1.4). It is a result of
representation theory that all irreducible representations of C1~ t have the
Isoparametric Hypersurfaces 301

same dimension 2.6(ml), where the number 6(ml) is defined by:

mI I 2 3 4 5 6 7 8 ... rnl+8

6(mj) 1 2 4 4 8 8 8 8 ... 16.6(ml)

The whole construction is studied in detail in [FKM]. We list only the following
properties:
i) Put m z : = x . f ( m ~ ) - m 1 - 1 for some xelN. If ml, m 2 >0, then f defines an
isoparametric family with 4 distinct principal curvatures and multiplicities m~ and
m 2. In general we have m I =m_ and m z = m § ; the only exceptions are:
(mx,mz)=(2, 1), (4, 3), (5, 2), (6, 1), (8, 7), (9, 6).
ii) Precisely the geometries with role{i,2,4,8 } and with an odd quotient
(m 1 + m 2 + 1) : 6(m1) have cohomology type 4B! ; they cover all the admissible pairs
of multiplicities for this type, as long as m_e{1,2,4,8}. The other Clifford
examples have cohomology type 4A each, i.e. 6(ml) divides m 1 + m z + 1 and 6(ml) is
greater than min {ml, m2}. However, 6(m~) is strictly greater than the power of 2 in
our estimate, as soon as m 1 > 10.
iii) The subseries where m 1 equals 1, 2, or 4 and the example with (m~, m2)
= (9, 6) are homogeneous.
iv) The 2m~ +mz-dimensional focal manifold (in general F+) also carries a
structure as a bundle:
]pmi+m2._,F2ml + m 2 , F , 2:" (Clifford sphere),
or
~ m I + m 2 . ~ . ] ~ 2 m i + rn2 ~ F ~, ~ r n I "

v) The algebra Cl ~ has inequivalent irreducible representations if and

only if 4 divides m a. The different representations produce geometrically different
examples having the same multiplicities. In [ F K M ] they are distinguished by the
Pontrjagin class p,,1/4(z)e H* (2:r~1;Z). Here z denotes the m I + m2 + 1-dimensional
vector bundle associated to ~v; p2,,,+,,2 is a hypersurface in the total space of z.
Let us denote their trivial normal bundle by N~:
~*(T)e ~*(TZ ~') = N , ~ TF 2"' +~'.
A short calculation, using Chern classes, shows:
c(TF 2"1 +"2 |162 = ~(c(z | c(TX"' |162
= 1 + ~*(p.,l/,(~)),

c(NF 2"' +~ | C) = c(TS" + xl... | c(TF 2"' +"~ | C)- 1

= 1 - ~*(p /,(T))

and
c(P.,|162= c((~,$ N ~)|162
=~*c(NP,|162
302 U. Abresch

As k* a n d ~* are injective, the g e o m e t r i c a l l y different examples with the same

multiplicities can also be distinguished by the P o n t r j a g i n classes of p2ml +m2 o r by
the P o n t r j a g i n classes of two c u r v a t u r e distributions/~i.
vi) P r o v i d e d that m_ > 1, it follows from the structure of H*(ffk;Tl) (cf.
[MI~IN 2]) t h a t i s o p a r a m e t r i c families m a y differ in the P o n t r j a g i n classes of the
focal m a n i f o l d s Fk (or in the classes of s o m e c u r v a t u r e distributions), only if:

41m_, or 41m+, or 41(2m_+m+).

This s t a t e m e n t does n o t change, if one m a k e s use of the c o r r e s p o n d i n g characteris-
tic classes in the u n d e r l y i n g setting. In o r d e r to see this, notice t h a t

n,n* : n*(Fk)--+ n*(ff k)--+n*(Fk)

is the s a m e as m u l t i p l i c a t i o n by 2 a n d that there is an identity c(VB|
=o~(VB) 2 m o d 2 , where VB denotes a n y real vector bundle.
If # = 4, even on the level of a d m i s s i b l e pairs of multiplicities the classification
p r o b l e m for i s o p a r a m e t r i c hypersurfaces r e m a i n s unsolved. In c o n t r a s t for # = 6
the possible multiplicities and, b y T h e o r e m 5.1 a n d the c o r r e s p o n d i n g r e m a r k ,
even the i s o m o r p h i s m classes of the c u r v a t u r e d i s t r i b u t i o n s have been d e t e r m i n e d
completely. In view o f 5.3v) the l a t t e r fact m i g h t be r e g a r d e d as a hint that n o
further i s o p a r a m e t r i c hypersurface with 6 distinct principal c u r v a t u r e s c o u l d exist.

References
Isoparametric Hypersurfaces
ABR Abresch, U. : Notwendige Bedingungen f'tir isoparametrische Hyperfl~ichenin Sph'aren mit
mehr als drei verschiedenen Hauptkriimmungen. Bonn. Math. Schr. 146 (1983)
CAR Cartan, E. : Sur des families remarquables d'hypersurfaces isoparametriques dans les espaces
spheriques. Math. Z. 45, 335-367 (1939)
FKM Ferus, D., Karcher, H., Miinzner, H.-F.: Cliffordalgebren und neue isoparametrische
Hyperfl~ichen. Math. Z. 177, 479-502 (1981)
HSI Hsiang, W.Y., Lawson,H.B., Jr. : Minimal submanifolds of low cohomogeneity. J. Diff.
Geometry 5, 1-38 (1971)
M1]N Miinzner, H.-F. :-Isoparametrische Hyperfl~ichenin Sph~iren(geometrischer Teil). Math. Ann.
251, 57-71 (1980)
MON 2 Miinzner, H.-F. : l~ber die Zerlegung der Sph~e in Ballbiindel. II. Math. Ann. 256, 215-232
(1981)
OZE Ozeki, H., Takeuchi, M. : On some types of isoparametric hypersurfaces in spheres. I : Tohoku
Math. J. 27, 515-559 (1975); II. 28, 7-55 (1976)
Tak Takagi, g., Takahashi, T. : On the principal curvatures of homogeneous hypersurfaces in a
sphere. Differential geometry, in honor of K. Yano, pp. 469-481. Tokyo: Kinokuniya 1972

Algebraic Topology
DOLD Dold, A. : Lectures on algebraic topology. Grundlehren der Mathematischen Wissenschaften.
Berlin, Heidelberg, New York: Springer 1972
HUS Husemoller, D.: Fibre bundles. New York : McGraw-Hill 1966; 2 no edition : Graduated Texts
in Mathematics No. 20, Berlin, Heidelberg, New York : Springer
MIL Milnor, J.W., Stasheff, J.D. : Lectures on characteristic classes. Ann. of Math. Studies 76.
Princeton, NJ : Princeton University Press 1974
SPA Spanier, E.H. : Algebraic topology. New York: McGraw-Hill 1966
STE Steenrod, N.E., Epstein, D.B.A.: Cohomology operations. Ann. of Math. Studies 50.
Princeton, NJ : Princeton University Press 1962

Received November 6, 1982