Articolul Lui Munzner

Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

Translator’s Note: This is an unofficial translation of the original paper which was

written in German. All references should be made to the original paper.

Translation of Cliffordalgebren und neue isoparametrische Hyperflächen, Math. Z.


177, 479–502 (1981) by Dirk Ferus, Hermann Karcher and Hans-Friedrich Münzner.

1. Translated by Thomas E. Cecil, Department of Mathematics and Computer Science,


arXiv:1112.2780v1 [math.DG] 13 Dec 2011

College of the Holy Cross, Worcester, MA 01610, USA;


E-mail address: cecil@mathcs.holycross.edu

2. Typed by Wenjiao Yan, School of Mathematical Sciences, Laboratory of Mathemat-


ics and Complex Systems, Beijing Normal University, Beijing 100875, China.
E-mail address: wjyan@mail.bnu.edu.cn

1
Clifford Algebras and New Isoparametric Hypersurfaces
Dirk Ferus1 , Hermann Karcher2 , Hans-Friedrich Münzner3

1. Fachbereich Mathematik der Technischen Universität, Straße des 17.


Juni 135, D-1000 Berlin 12, Bundesrepublik Deutschland
2. Mathematisches Institut der Universität, Wegelerstraße 10, D-5300 Bonn,
Bundesrepublik Deutschland
3. Studienbereich Mathematik der Universität, Kufsteinerstraße, D-2800
Bremen 33, Bundesrepublik Deutschland

1. Introduction.

Isoparametric hypersurfaces, i.e. those with constant principal curvatures, are of special
interest because of their comparatively simple geometry. While in spaces of non-positive
constant curvature, they reduce to the examples of tubes over totally geodesic subman-
ifolds, the theory of isoparametric hypersurfaces in spheres is rich: the homogeneous
examples are completely known through the work of Hsiang and Lawson [2], moreover
Takeuchi and Ozeki [5] found two infinite series of inhomogeneous examples with g = 4
distinct principal curvatures. In the following work, by means of representations of
Clifford algebras, for each natural number m1 , we will construct an infinite series of
isoparametric hypersurfaces with g = 4 distinct prinicpal curvatures with multiplicities
(m1 , m2 , m1 , m2 ) where m2 grows monotonically in the series. Our examples include all
previously known homogeneous and inhomogeneous (g = 4) examples with the excep-
tion of two individual homogeneous cases of dimension 8 and 18. Particular attention
will be given to the case m1 ≡ 0 mod 4. For this case, as the dimension increases there
are more and more families with the same multiplicities (m1 , m2 , m1 , m2 ) and more
and more non-isometric compact Riemannian manifolds of the same dimension, which
(modulo an isometry of the tangent space) have the same curvature tensor pointwise.
The incongruence of families with the same multiplicities – as well as their inhomo-
geneity in many cases – can be seen through the second fundamental forms of the focal
submanifolds, which are quite accessible in our representation of the families.
On isoparametric hypersurfaces, the curvature tensor, as a field of endomorphisms
on bivectors also has constant eigenvalues. Our examples (as well as the series of

2
Takeuchi and Ozeki) provide a large number of Riemannian manifolds with these prop-
erties. This construction by means of Clifford representations carries over to spaces
with indefinite scalar product and provides examples of isoparametric hypersurfaces in
Lorentz spaces of constant curvature which Nomizu has recently begun to study.
Topologically, our examples of multiplicities (m1 , m2 , m1 , m2 ) have the form of
m2 -sphere bundles over an (m1 + m2 )-sphere bundle over an m1 -sphere.
In the two following sections, we give a brief summary of the necessary differential
geometric and algebraic foundations. Section 4 contains the definition of the new
examples; and in the rest we study their geometry.

2. Isoparametric hypersurfaces in spheres.

In this section, we put together the fundamental properties of isoparametric hy-


persurfaces which are important to us. For further details see [4].

2.1 Definitions: An oriented hypersurface in a sphere with constant principal cur-


vatures is called an isoparametric hypersurface. ( If one only requires local constancy,
then this definition makes sense for non-orientable hypersurfaces as well, but – as can
be shown – not globally: so the more general definition is furnished only for orientable
hypersurfaces. )

2.2 Parallel surfaces: Given an oriented hypersurface M in the sphere S n+1 with
principal curvatures cot(φi ), 0 < φi < π with multiplicities mi , then the parallel
hypersurface M at a distance  in the direction of the normal has principal curvatures
cot(φi − ) with the same multiplicities, and M and M have the same normal great
circles. The parallel hypersurfaces obtained this way are likewise isoparametric; they
form an isoparametric family.
The mean curvature n1
P
mi cot(φi − ) as an analytic function of  determines
the φi through its poles (modulo π). If all the hypersurfaces in a parallel family have
constant mean curvature, then they have constant principal curvatures and so are
isoparametric.

2.3 Principal foliations ( curvature foliations. ) The mi -dimensional eigendis-


tribution of the second fundamental form ( Weingarten map ) S of an isoparametric
hypersurface corresponding to the principal curvature cot(φi ) is integrable and autopar-
allel. It integrates to a totally geodesic foliation of M by mi -spheres of radius φi in
S n+1 . The parallel surface at distance φi is a focal manifold of the isoparametric family,
and the hypersurfaces of the family are tubes over the focal manifolds.

2.4 Distribution of values of the principal curvatures. Consider the principal

3
curvature cot φi of some hypersurface of the isoparametric family. Its curvature leaf is a
sphere and its center is a point of the “corresponding” focal submanifold, with the radii
of the sphere as normals. The eigenvalues of the 2d fundamental tensors are cot(φj −φi ),
j 6= i, independent of the point and the normal direction. From this, one gets through
an argument not given here, that the principal curvatures of the hypersurfaces are of
the form cot(α+ g−ig π) with multiplicities mi . Here, g is the number of distinct principal
curvatures, 1 ≤ i ≤ g, and α ∈ (0, π). Moreover, it turns out that mi = mi+2 with
indices modulo g, see Münzner [4].

2.5 Isoparametric functions: If f : S n+1 → R is a function whose gradient on each


level surface
. has constant length, then the regular levels are parallel hypersurfaces with
grad f kgrad f k as unit normal field. The corresponding second fundamental tensor
is given by
grad f −1
SV = −∇V = ∇V grad f ;
kgrad f k kgrad f k
∆f
the mean curvature is − n1 kgrad fk .
Thus the levels form an isoparametric family precisely when ∆f is constant on
the levels, that is, when both of “the first two differential parameter” functions are
functions of f itself:
kgrad f k2 = a(f ), ∆f = b(f ).
Whence the name “isoparametric”.

2.6 Differential equations of Cartan and Münzner: Conversely, given an isopara-


metric family and a function f of the oriented distance from a fixed hypersurface of the
family, then f obviously satisfies the above differential equations. The information in
2.4 can be exploited to specify the right hand side of the differential equations further.
Through Cartan and Münzner
f = cot gt,
where t is the spherical distance to a focal manifold and as usual g is the number of
distinct principal curvatures, one is led to a particularly good standardization: extend
f to a homogeneous function F of degree g on a cone in Rn+2 , then the differential
equations for F ( with Euclidean differential operators ) are written as:

kgrad F k2 = g 2 r2g−2 , ∆ F = crg−2 ,

with c = g 2 m2 −m
2
1
( = 0 for odd g ) and r(x) = kxk.
One sees as a simple consequence that F is a homogeneous polynomial of degree
g, see [4]. In particular, one can extend each piece of isoparametric hypersurface to a
compact algebraic hypersurface.

4
2.7 Rigidity. Isoparametric hypersurfaces are rigid: each isometry of such a hypersur-
face can be extended to an isometry of the sphere; hence the parallel hypersurfaces are
mapped to themselves, thus causing an isometry of the whole family. For hypersurfaces
with g ≥ 4, which are of interest to us here, this follows from the classical rigidity the-
orem since the rank of the second fundamental tensor is obviously ≥ 3. For g ≤ 3, one
can make use of the classification of Cartan [1] or otherwise carry out a direct proof.

2.8 Minimality. From the facts collected in 2.4 one obtains that the focal manifolds
of an isoparametric family are minimal submanifolds of the sphere and each such family
contains precisely one minimal hypersurface.

3. Clifford systems.

Our examples of isoparametric hypersurfaces are constructed from from so-called


Clifford systems whose definitions and properties are collected in this section.

3.1 Notation. For a Euclidean vector space V , let h(V ) resp. O(V ) resp. O(V ) denote
the symmetric resp. skew-symmetric resp. orthogonal endomorphisms of V . We let
h , i denote the canonical scalar product on Rn , and on h(V )
1
hA, Bi := Trace(AB).
dim V
E± (A) denotes the eigenspace of the eigenvalue ±1.

3.2 Definition. Let l, m be positive natural numbers.

(i) The (m + 1)-tuple (P0 , ..., Pm ) with Pi ∈ h(R2l ) is called a (symmetric) Clifford
system on R2l if for all i, j ∈ {0, ..., m} we have

Pi Pj + Pj Pi = 2δij Id.

(ii) Let (P0 , ..., Pm ) and (Q0 , ..., Qm ) be Clifford systems on R2l resp. R2n , then
(P0 ⊕ Q0 , ..., Pm ⊕ Qm ) is a Clifford system on R2(l+n) , the so-called direct sum
of (P0 , ..., Pm ) and (Q0 , ..., Qm ).

(iii) A Clifford system (P0 , ..., Pm ) on R2l is called irreducible when it is not possible
to write R2l as a direct sum of two positive dimensional subspaces which are
invariant under all of the Pi .

3.3 Representations of Clifford algebras and Clifford systems. Each represen-


tation of a Clifford algebra Cm−1 on Rl can be characterized by E1 , ..., Em−1 ∈ O(Rl )
with
Ei Ej + Ej Ei = −2δij Id.

5
One now defines (P0 , ..., Pm ) ∈ h(R2l ) by

P0 (u, v) := (u, −v), P1 (u, v) := (v, u), P1+i (u, v) := (Ei v, −Ei u), u, v ∈ Rl ,

then (P0 , ..., Pm ) is a symmetric Clifford system on R2l that is irreducible precisely
when the representation of Cm−1 is irreducible.
Conversely, one obtains all symmetric Clifford systems in this way: let (P0 , ..., Pm )
be a Clifford system on R2l , then since P02 = Id, P0 has eigenvalues ±1, and since
P0 P1 + P1 P0 = 0, P1 interchanges the corresponding eigenspaces E± (P0 ). Thus E+ (P0 )
is of dimension l.
Further, E+ (P0 ) is invariant under P0 Pi+1 , i ∈ {1, ..., m − 1}. One can identify
E+ (P0 ) with Rl , and thus define through

Ei := P1 Pi+1 |E+ (P0 )

a representation of Cm−1 on Rl .

3.4 Definition. Two Clifford systems (P0 , ..., Pm ) and (Q0 , ..., Qm ) on R2l are called
algebraically equivalent if there exists A ∈ O(R2l ) such that Qi = APi At for all i ∈
{0, ..., m}. They are called geometrically equivalent when there exists B ∈ O(Span{P0 , ...,
Pm }) ⊂ h(R2l ) such that (Q0 , ..., Qm ) and (B(P0 ), ..., B(Pm )) are algebraically equiva-
lent.

3.5 Representation theory. With the help of 3.3, we obtain from the representa-
tion theory of Clifford algebras, see [3], the following results: each Clifford system is
algebraically equivalent to a direct sum of irreducible Clifford systems. An irreducible
Clifford system (P0 , ..., Pm ) on R2l exists precisely for the following values of m and
l = δ(m):

m 1 2 3 4 5 6 7 8 · · · m+8
δ(m) 1 2 4 4 8 8 8 8 · · · 16δ(m)

For m 6≡ 0(mod 4), there exists exactly one algebraic ( and therefore exactly one
geometric ) equivalence class of irreducible systems. Thus for each positive integer k
there exists exactly one algebraic ( or geometric ) equivalence class of Clifford systems
(P0 , ..., Pm ) on R2l with l = kδ(m).
For m ≡ 0(mod 4), there exist exactly two algebraic classes of irreducible systems.
These are distinguished from each other by the choice of sign in

Trace(P0 · · · Pm ) = ±Trace Id = ±2δ(m).

6
Hence (by replacing P0 by −P0 ) there exists exactly one geometric equivalence class in
this case also.
The absolute trace
|Trace(P0 · · · Pm )|

is obviously an invariant under geometric equivalence. If one constructs all direct sums
of both of the irreducible algebraic classes with altogether k summands, then this
invariant takes on [ k2 ] + 1 different values. Thus for m ≡ 0(mod 4), there are exactly
[ k2 ] + 1 geometric equivalence class of Clifford systems on R2l with l = kδ(m).

3.6 Definition. Let (P0 , ..., Pm ) be a Clifford system on R2l . The unit sphere in
Span{P0 , ..., Pm } ⊂ h(R2l ) is called the Clifford sphere determined by the system and
is denoted Σ(P0 , ..., Pm ).

3.7 Properties of the Clifford sphere. Our construction will not depend on
(P0 , ..., Pm ) but only on Σ(P0 , ..., Pm ). We therefore begin with several important prop-
erties of Clifford spheres.
(i) For each P ∈ Σ(P0 , ..., Pm ), we have P 2 = Id. Conversely, if Σ is a unit
sphere in a linear subspace RΣ ⊂ h(R2l ) such that P 2 = Id for all P ∈ Σ, then every
orthonormal basis of RΣ is a Clifford system on R2l .
(ii) Two Clifford systems on R2l are geometrically equivalent if and only if their
Clifford spheres are conjugate to one another under an orthogonal transformation of
R2l .
(iii) The function
m
X
H(x) = hPi x, xi2
i=0

depends only on Σ(P0 , ..., Pm ), and not on the choice of orthonormal basis (P0 , ..., Pm ).
For P ∈ Σ(P0 , ..., Pm ), we have H(P x) = H(x) for all x. (To prove this, choose
P0 , ..., Pm orthonormal with P0 = P .)
(iv) For orthonormal Q1 , .., Qr ∈ Σ(P0 , ..., Pm ) since Qi Qj = −Qj Qi for all i 6= j,
we have

Q1 · · · Qr ∈ h(R2l ), f or r ≡ 0, 1(mod 4)
Q1 · · · Qr ∈ O(R2l ), f or r ≡ 2, 3(mod 4).

Further, Q1 · · · Qr is uniquely determined by an orientation of Span(Q1 , ..., Qr ).


SO(r) is generated as a group by rotations of two-dimensional coordinate planes, and
since one can bring any two Qi ’s next to each other through permutation modulo signs,
it suffices to do the proof for r = 2. This is an easy direct calculation.

7
(v) For P, Q ∈ Span{P0 , ..., Pm } and x ∈ R2l , we have

hP x, Qxi = hP, Qihx, xi.

4. The new examples. We now give the new series with g = 4 distinct principal
curvatures and study their geometry.

4.1 Theorem: Let (P0 , ..., Pm ) be a Clifford system on R2l . We define m1 :=


m, m2 := l − m − 1 and F : R2l → R by
m
X
2
F (x) := hx, xi − 2 hPi x, xi2 .
i=0

Then F satisfies the Cartan-Münzner differential equations (2.6). If m2 > 0, then the
levels of F form an isoparametric family with g = 4 distinct principal curvatures with
multiplicities (m1 , m2 ). ( Note m3 = m1 , m4 = m2 through 2.4.)
Proof. We have
X
gradx F = 4hx, xix − 8 hPi x, xiPi x

and with 3.7 (v), we have the first equation of 2.6. Further
X 
∆x F = 4(2l + 2)hx, xi − 2 2hPi x, xi∆hPi x, xi + 2kgradhPi x, xik2
= 8(l + 1)hx, xi − 16hx, xi(m + 1)
= 8(m2 − m1 )hx, xi,

and so the second equation of 2.6 is satisfied. The rest of the assertion follows from [4],
likewise see 2.5, 2.6.

4.2 The focal manifolds. The focal manifolds of an isoparametric family contain very
concentrated geometric information. We study the focal manifolds of our examples
in the next theorem and in the following paragraphs of this section. A part of the
results which we obtain is already known from the general theory of isoparametric
hypersurfaces.
Theorem: With the notation from 4.1, let f = F |S 2l−1 and Σ := Σ(P0 , ..., Pm ).
(i) For M− := f −1 ({−1}), we have

M− = {x ∈ S 2l−1 | there exists P ∈ Σ with x ∈ E+ (P )}.

In the case m2 < 0, then f = −1, thus M− = S 2l−1 ; this is only possible for m ∈
{1, 2, 4, 8}.

8
In the case m2 ≥ 0, then M− is diffeomorphic to the total space of an (l −1)-sphere
bundle
π
Γ := {(x, P ) | x ∈ S 2l−1 , P ∈ Σ, x ∈ E+ (P )} −
→ Σ, (x, P ) 7→ P.

The diffeomorphism from Γ onto M− is furnished by (x, P ) 7→ x. In particular, if f is


not constant, then M− is a – trivially connected – submanifold of codimension m2 + 1
in the sphere S 2l−1 .
In the case m2 = 0, then M− is a hypersurface; this is only possible for m ∈
{1, 3, 7}.
In the case m2 > 0, then M− is the focal manifold corresponding to the principal
curvatures of the family of multiplicity m2 . The hypersurfaces are m2 -sphere bundles
over the connected sphere bundle space M− .
Suppose (P0 , ..., Pm ) can be extended to a Clifford system (P0 , ..., Pm+1 ), which by
3.5 is not rare, then π : Γ → Σ is trivial and M− is diffeomorphic to S l−1 × S m .
For m ≡ 0(mod 4), the geometrically inequivalent Clifford systems (see 3.5) lead to
inequivalent sphere bundles Γ → Σ.
(ii) For M+ := f −1 ({+1}), we have

M+ = {x ∈ S 2l−1 | hP0 x, xi = ... = hPm x, xi = 0}.

In the case m2 ≥ 0, then M+ is a non-empty submanifold of codimension m1 + 1 and


a focal manifold of the level hypersurfaces corresponding to the principal curvatures of
multiplicity m1 . The normal bundle of M+ is trivial with x → (P0 x, ..., Pm x) as a basis
field. Hence the hypersurfaces are trivial sphere bundles over M+ .
(iii) For x ∈ M+ and P ∈ Σ, on the normal great circle c(t) := cos t x + sin t P x, we
have f (c(t)) = cos 4t.
The normal great circle meets M+ again after π2 , so that the hypersurface Mt =
f −1 ({t}) of the family in addition to cot t has also cot(t + π2 ) as a principal curvature of
multiplicity m1 . In the case m2 = 0, then only two of these principal curvatures arise:
one obtains an isoparametric family with g = 2. If m2 > 0, then the normal great circle
meets M− at t = π4 and 3π π 3π
4 , so that cot(t + 4 ) and cot(t + 4 ) are principal curvatures
with the same multiplicity m2 .

Proof. (i) From f (x) = −1, it follows from the definition of F that

hΣhPi x, xiPi x, xi = 1

thus
ΣhPi x, xiPi x = x,

9
and P = ΣhPi x, xiPi ∈ Σ has x as +1-eigenvector. Conversely, if P x = x for an
x ∈ S 2l−1 and P ∈ Σ, then one can assume as in 3.7(iii) that P = P0 . But then one
gets from 3.2(i) that hPi x, xi = 0 for all i > 0. It follows that

f (x) = 1 − 2hP0 x, xi2 = −1

and one has the stated characterization of M− .


In the case m2 < 0, one has l ≤ m. For x ∈ S 2l−1 , let x = x+ + x− be the
decomposition of x into eigenvectors of P0 , where x± ∈ E± (P0 ). Then P1 x± , ..., Pm x±
are orthogonal in E∓ (P0 ). Thus l = m, and a simple direct calculation shows that
F (x) = −1.
Now let m2 ≥ 0. First of all, π : Γ → Σ is actually a sphere bundle, since for
P ∈ Σ, E+ (P )×Σ → Γ, e (x, Q) 7→ (Id+Q)x is a local trivialization of the corresponding
vector bundle Γ e → Σ in a neighborhood of P . As we have already proven that the map
Γ → M− , (x, P ) 7→ x is surjective, it is easily shown that the map is a submersion.
From 3.7(v), it is also known that it is injective, and hence it is a diffeomorphism.
The restriction on the values of m is known by 3.5. Finally, suppose it is possible to
extend (P0 , ..., Pm ), then π : Γ → Σ is the restriction of the analogous bundle over
Σ(P0 , ..., Pm+1 ) to the “equatorial sphere” Σ, and is thus trivial. The vector bundle
e → Σ has as a subbundle of R2l × Σ → Σ a canonical metric and covariant derivative.
Γ
The curvature tensor R e has the form

e 1 , Q2 ) = 1 Q1 Q2 |E (P ) ,
R(Q +
2
for P ∈ Σ and Q1 , Q2 ∈ TP (Σ) = (RP )⊥ . If m ≡ 0(mod 4), then one has a characteristic
m-form χ on Σ by the following definition, for P ∈ Σ and Q1 , ..., Qm an orthonormal
basis of TP (Σ):
m
22 X
e σ(1) , Qσ(2) ) ◦ · · · ◦ R(Q

e σ(m−1) , Qσ(m) ) ∗ 1
χ := Trace sign(σ)R(Q
m! σ
= Trace(Q1 · · · Qm |E+ (P ) ) ∗ 1
= Trace(P Q1 · · · Qm |E+ (P ) ) ∗ 1
1
= Trace(P Q1 · · · Qm ) ∗ 1
2
1
= Trace(P0 · · · Pm ) ∗ 1.
2
But by (3.5) the (absolute) trace distinguishes between geometrically equivalence
classes.
(ii) From the homogeneity of F and (2.6) it follows that

kgrad f k2 = 16(1 − f 2 ).

10
For m2 ≥ 0, then by (i) f is not constant and must assume the value 1 as maximum:
so M+ 6= ∅. The remaining statements are immediately clear.
(iii) From 3.7(iii), it suffices to consider the case P = P0 . Then

F (c(t)) = 1 − 2hcos t P0 x + sin t x, cos t x + sin t P0 xi2


Xm
−2 hcos t Pi x + sin t Pi P0 x, cos t x + sin t P0 xi2
i=1
= 1 − 2(2 cos t sin t)2
= cos(4t).

The remaining statements are again clear.

4.3 Table of small multiplicities. From 3.7(ii) and (iii), geometrically equivalent
Clifford systems give congruent isoparametric families. In paragraph 4.6 we will show
a result in the converse direction. In both cases the small multiplicities play a special
role. We thus give a list of our multiplicities (m1 , m2 ) from Theorem 4.1 as in 3.5. So
l = kδ(m), m1 = m and m2 = l − m − 1.

The underlined multiplicities (m1 , m2 ) and (m1 , m2 ) denote, respectively, the two,
resp. three geometrically inequivalent Clifford systems for the multiplicities (m1 , m2 ),
see 3.5. We will show further below that all of these with m ≡ 0 mod 4 and l = kδ(m)
actually lead to incongruent isoparametric families ( of which there are [ k2 ] + 1. ) We
will also see that the families for multiplicities (2, 1), (6, 1), (5, 2) and one of the two
(4, 3)-families are congruent to those with multiplicities (1, 2), (1, 6), (2, 5) and (3, 4),
resp. These are all the coincidences under congruence.

4.4 Multiplicities of previously known examples. We give now for comparison


the multiplicities of the previously known examples with g = 4.

11
Homogeneous examples: see [6]: (1, k), (2, 2k − 1), (4, 4k − 1), (9, 6), (4, 5), (2, 2)
Inhomogeneous examples: see [5]: (3, 4k), (7, 8k).
We will show that these are all Clifford examples except for the homogeneous
examples with (4, 5) and (2, 2).

4.5 The second fundamental tensors of the focal manifolds. We now continue
the study of focal manifolds which was begun in 4.2 and will describe their second
fundamental tensors.
(i) From 4.2(iii), we have that for a unit normal N to an (mi + 1)-codimensional
focal manifold, the corresponding second fundamental tensor SN has the eigenvalue
0 of multiplicity mi and +1, respectively, −1 with multiplicities mj , {i, j} = {1, 2}.
Thus corresponding to the unit mi -sphere in the normal space, there is an mi -sphere of
symmetric endomorphisms S with S 3 = S and mi -dimensional kernel. The classification
of these algebraic structures (these exist likewise for the non-Clifford families with
g = 4) looks much harder than the Clifford case (S 2 = Id).

(ii) Let x ∈ M+ . Then by 4.2(ii),

⊥x M+ = {P x | P ∈ RΣ(P0 , ..., Pm )},

and for the second fundamental tensors for N = P x, P ∈ Σ := Σ(P0 , ..., Pm ), we have

SN v = M+ − tangential component of (−P v).

From this and with


ΣP := {Q ∈ Σ | hP, Qi = 0},

it follows very easily that

ker SN = RΣP N (N = P x)
SN |(ker SN )⊥ = −P |(ker SN )⊥ .

(iii) Let y ∈ M− and P ∈ Σ with P y = y. Let ΣP again be the equatorial sphere of


Σ orthogonal to P . Since E+ (P ) ∩ S 2l−1 ⊂ M− , from 4.2(i), then ⊥y M− ⊂ E− (P ).
Further, from 3.7(iii), for all Q ∈ ΣP , F ((cos t P + sin t Q)y) = F (y) = −1, and hence
0
(cos t P y+sin t Qy) (0) = Qy ∈ Ty M− . But since RΣP y ⊂ E− (P ) is an m1 -dimensional
subspace, it follows for dimensional reasons that

⊥y M− = {N ∈ E− (P ) | N ⊥ ΣP y}.

12
Now let N ∈⊥y M− , kN k = 1. The normal great circle cos t y + sin t N meets M+ in
x := √12 (y + N ) and P x = √12 (y − N ). Since the eigenspaces of the second fundamental
tensors of a parallel family of hypersurfaces are parallel along a normal circle, it follows
from (ii) that
E+ (SN ) = RΣP (y + N ), E− (SN ) = RΣP (y − N )

ker SN = {v ∈ E+ (P ) | v ⊥ y, v ⊥ ΣP N }.

4.6 The uniqueness theorem. The second fundamental tensors of the focal manifolds
are not only well suited to distinguish families with the same multiplicities as in the
following theorems, but also for distinguishing the geometry at different points of the
same focal submanifold, as will become clear further on in our work.
Theorem: Let (P0 , ..., Pm ) be a Clifford system on R2l and

m = m1 ≤ m2 = l − m − 1.

Let f be defined as in 4.2 and M− = f −1 ({−1}). Let P ∈ Σ := Σ(P0 , ..., Pm ) and


y ∈ E+ (P ) ∩ S 2l−1 . Then from 4.2(i) y ∈ M− and E+ (P ) = Ry ⊕ Span ∪ ker SN , where
the union is taken over all N ∈⊥y M− \{0}. Since the Clifford sphere Σ is uniquely
determined by the set {E+ (P ) | P ∈ Σ}, so also is the isoparametric family uniquely
determined. In particular, the congruence class of the hypersurface family determines
the geometric equivalence class of the representation.

Proof. We denote the orthogonal complement in Ty M− by ( )⊥ . Then it follows from


4.5(iii) that
[ \
Span ker SN = ( E+ (SN ) ⊕ E− (SN ))⊥
N 6=0 N 6=0
\
= ( RΣP (y + N ) ⊕ RΣP (y − N ))⊥
N 6=0
\
= ( RΣP y ⊕ RΣP N )⊥
N 6=0
\
= (RΣP y ⊕ RΣP N )⊥ ,
N 6=0

since ΣP y ⊂ E− (P ) and ΣP N ⊂ E+ (P ). But since Ty M− = RΣP y ⊕ (RΣP N ⊕ ker SN )


as an orthogonal direct sum with RΣP N ⊕ker SN ⊂ E+ (P ), then (RΣP y)⊥ is a subspace
of E+ (P ) of dimension dim M− − m1 = m1 + m2 , so that

E+ (P ) = Ry ⊕ (RΣP y)⊥ .

13
Hence, it suffices to prove that
\
RΣP N = {0}.
N 6=0

Given a u 6= 0 in this intersection, then one has for each N 6= 0 a Q ∈ RΣP with
QN = u.
1
From Q2 = kQk2 Id, it follows that N = kQk2 Qu, i.e. the rank of the map RΣP →
2l
R , Q 7→ Qu is at least dim ⊥y M− = m2 + 1, in contradiction to the assumption that
dim RΣP = m1 ≤ m2 .

4.7 Consequences. A glance at the table 4.3 shows that the assumption m1 ≤ m2 is
almost always satisfied. We will consider the (eight) exceptional cases later. The result
4.6 is naturally of special interest in the cases m ≡ 0 mod 4, in which there exist [ k2 ] + 1
geometrically inequivalent Clifford systems on R2l with l = kδ(m).
The corresponding families ( except (4, 3) and (8, 7) ) are now seen to be incon-
gruent. If one chooses two hypersurfaces from two incongruent families at the same
distance from M− , then they have the same principal curvatures and thus from the
Gauss equation the same curvature tensor: to two points from the two hypersurfaces,
there exists a linear isometry of the corresponding tangent spaces, which transforms one
curvature tensor into the other. Nevertheless, by the remark 2.7, these hypersurfaces
are not intrinsically isometric!

5. Inhomogeneity of Clifford families. We will later show that several of our exam-
ples are homogeneous, see 6.1 and 6.3. Most of our examples are indeed inhomogeneous,
since their multiplicities are not found in the list 4.4 of homogeneous multiplicities. In
this section, we will give–without use of the homogeneous classification–a direct geo-
metric proof of the inhomogeneity of most of the families and their focal manifolds.
Let (P0 , ..., Pm ) be a fixed Clifford system on R2l . Let the notation F , M , M±
etc.be as in 4. In particular, let

Σ := Σ(P0 , ..., Pm ).

5.1 Theorem: Let N+ := {x ∈ M+ | there exists an orthonormal Q0 , ..., Q3 ∈


Σ with Q0 · · · Q3 x = x}. Then N+ has the following geometric description:
\
N+ = {x ∈ M+ | there exists orthonormal N0 , ..., N3 ∈⊥x M+ with dim( ker SNi ) ≥ 3}.
i

14
Proof. First let x ∈ N+ , x = Q0 · · · Q3 x. From 4.5(i), the vectors Ni := Qi x ∈⊥x M+
are orthonormal and Q0 Q1 x = −Q2 Q3 x, Q0 Q2 x = Q1 Q3 x and Q0 Q3 x = −Q1 Q2 x
\
are orthonormal vectors in ker SNi . Conversely, let N0 , N1 ∈⊥x M+ with at least
i
three-dimensional intersection of their corresponding kernels. From 4.5(i) and 3.7(v),
there exists an orthonormal Q0 , Q1 ∈ Σ with Qi x = Ni . The intersection of the kernels
contains a vector orthonormal to Q0 Q1 x which by 4.5(i) and 3.7(v) must be of the
form Q0 Q2 x = Q1 Q3 x with Q2 , Q3 ∈ Σ orthonormal and orthogonal to Q0 , Q1 . Hence
Q0 · · · Q3 x = x.

5.2 Theorem: Suppose 9 ≤ 3m1 < m2 + 9 and for m1 = 4 suppose that the additional
identity P0 · · · P4 6= ±Id holds, then ∅ 6= N+ 6= M+ . Thus the focal manifold M+ and
the whole family are not homogeneously embedded.
Proof. The endomorphism P := P0 · · · P3 is involutive, symmetric commuting with
P4 , ..., Pm and anti-commuting with P0 , ..., P3 . Let S+ (P ) := E+ (P ) ∩ S 2l−1 be the unit
sphere in the +1 eigenspace.
For x ∈ E+ (P ), we have
m
X
2
F (x) = hx, xi − 2 hPi x, xi2 ,
i=4

since hP0 x, xi = ... = hP3 x, xi = 0. Thus for m = m1 = 3, we have that S+ (P ) ⊂ M+ .


For m = 4, it follows from the assumption that P4 is indefinite on E+ (P ), and so
S+ (P ) ∩ M+ = (F |S+ (P ) )−1 ({1}) is (l − 2)-dimensional. For m > 4, P4 , ..., Pm is a
Clifford system on E+ (P ), whose (+)-focal manifold of dimension l − m + 2 is just
S+ (P ) ∩ M+ .
In all three cases, thus we have

N+ ⊃ S+ (P ) ∩ M+ 6= ∅ and dim(S+ (P ) ∩ M+ ) = l − m + 2.

On the other hand, by 3.7(iv), E+ (Q0 · · · Q3 ) depends only on the orientation of


Span(Q0 , ..., Q3 ). The dimension of the Grassmann manifold of oriented 4-planes in
RΣ is 4(m + 1 − 4). Thus N+ has at most dimension

4(m − 3) + l − m + 2 = 4(m1 − 3) + m2 + 3 = dim M+ + 3m1 − m2 − 9 < dim M+ .

Thus N+ 6= M+ and the theorem is proven.

5.3 Remarks. The above theorem is not applicable to the case m ≤ 2 or m = 4


with P0 · · · P4 = ±Id, as these will be shown to be homogeneous. Moreover, because
of the assumption 3m1 < m2 + 9, this theorem gives no information about the further

15
(finite) number of exceptions with m1 ≥ 5. Thus we wish to generalize the above
method in such a way as to replace Q0 · · · Q3 in the definition of N+ by products of the
form Q0 · · · Q4µ−1 and Q0 · · · Q4µ and their eigenspaces. The geometric interpretation
of the eigenvectors will then be more complicated than in Lemma 5.1. Occasionally,
the geometric interpretation becomes clear in the process of giving the proof, but we
will not do the interpretation in general. Instead we remark here that in the case
m1 ≤ m2 , from 4.6 and 3.7(iv), the Clifford sphere Σ and also the set E+ (Q0 · · · Qν )
for orthonormal Q0 , ..., Qν is geometrically determined. Whenever the family of such
eigenspaces meets a level of an isoparametric function in a non-empty proper subset,
then that level and hence the whole family is inhomogeneous.

5.4 Inhomogeneous families. The following theorem is sharper than Theorem 5.2
when m1 ≥ 5.
Theorem: Let 5 ≤ m1 ≤ m2 . Then the isoparametric family is embedded inhomoge-
neously.
Proof. The product P := P0 · · · P4 is symmetric and involutive; it commutes with
P0 , ..., P4 and anti-commutes with P5 , ..., Pm . Thus (P0 , ..., P4 ) is a Clifford system on
the l-dimensional eigenspace E+ (P ), and for each hypersurface M of the original family,
we have dim(M ∩ E+ (P )) = l − 2. The Grassmann manifold of 5-planes in RΣ has
dimension 5(m − 4) and by inspection of 3.5, we see that 5(m − 4) + l − 2 < 2l − 2 since
5 ≤ m1 ≤ m2 . Thus M ∩ (∪E+ (Q0 · · · Q4 )), where the union is over all orthonormal
Q0 , ..., Q4 in Σ, is a non-empty, proper, isometry-invariant subset of M , and hence M
is not homogeneous.

5.5 Remarks. The only Clifford examples with m1 ≥ 5 for which m2 < m1 , and thus
to which 5.4 does not apply, have multiplicities (5, 2), (6, 1), (9, 6) or (8, 7). The first
three are homogeneous, and both (8, 7)-families are not (see Section 6).

5.6 Inhomogeneity of the focal manifolds. If an isoparametric family is inhomo-


geneous, it is not necessary that its focal manifolds be inhomogeneous, see example 6.4
and 6.7(ii). But again “most” focal manifolds are inhomogeneous:
Theorem: Let m1 ≤ m2 .
(i) If 25 ≤ 5m1 < m2 + 25, then M+ is inhomogeneously embedded;
(ii) If 35 ≤ 7m1 < m2 + 33, then M− is inhomogeneously embedded;
(iii) M+ in the (10, 21)-family and M− in the (9, 22)-family are inhomogeneously em-
bedded;
(iv) If m1 ≡ 0 mod 4 and P0 · · · Pm 6= ±Id, then both focal manifolds are inhomoge-
neously embedded.

16
Proof. (i) and (ii). On E+ (P0 · · · P4 ) we have
4
X
F (x) = hx, xi2 − 2 hPi x, xi2 ,
i=0

since, in contrast with the proof of Theorem 5.2, P0 · · · P4 commutes with P0 , ..., P4 ,
and anti-commutes with P5 , ..., Pm . The +1 eigenspace of the orthogonal five product
Q0 · · · Q4 , for m1 ≥ 5 cuts out a non-trivial isoparametric family with m0 = 4, l0 = 2l .
In particular, its intersection with M+ , resp. M− is of dimension 4 + 2( 2l − 4 − 1) =
m1 + m2 − 5, resp. 8 + ( 2l − 4 − 1) = 21 (m1 + m2 + 7). In each case, one adds the
dimension of the Grassmann manifold of oriented 5-planes in RΣ, that is 5(m1 − 4),
and compares with the dimension of M+ , resp. M− , and the results follow as in 5.2.
(iii) This follows analogously using the 9-product Q0 · · · Q8 and for
(iv) one uses the highest dimensional eigenspace of P0 · · · Pm in the corresponding
way.

5.7 Remark on the exceptional cases. In Section 6, we will show the homogeneity
of the families with (m1 , m2 ) = (9, 6), min{m1 , m2 } ≤ 2, and families with m1 = 4,
P0 · · · P4 = ±Id. If one takes these results as true, the Table 4.3 and 5.2 and 5.6
show that only for the following cases is the homogeneity or inhomogeneity of the focal
manifolds yet undetermined:
(i) M+ for the multiplicities (8, 7) and (8, 15), the latter with P0 · · · P8 = ±Id.
(ii) M− for the multiplicities (3, 4k), (6, 9), (7, 8), (7, 16), (8, 7), (10, 21), (12, 51) and
(8, 15) and (8, 23) in the case P0 · · · P8 = ±Id.
In several of these cases, the question of homogeneity remains open, but the ques-
tion of homogeneity for the whole family for (8, 7) will be clarified in the following.

5.8 The Condition (A) of Takeuchi-Ozeki. We return now to the beginning of this
section. Then (and implicitly likewise in both of the other proofs in 5) it was necessary
to consider the behavior of the kernel of the second fundamental tensor for different
normal directions N to establish “inhomogeneous properties”. A particularly extreme
case occurs, and one which is of special interest in light of Remark 4.5(i), when at some
point of a focal manifold the kernels of all second fundamental tensors SN (N 6= 0)
coincide. We point out here that this is just the condition (A) of Takeuchi and Ozeki
which was introduced in Lemma 17 [5].
Concerning Condition (A) in the Clifford case, we have:
Theorem: Let m1 ≡ 3 mod 4. For x ∈ M+ , let d(x) := dim ∩ ker SN where the

17
intersection is taken over all N ∈⊥x M+ . Then we have:

min d = 0, max d > 0.

For m1 ∈ {3, 7}, max d = m1 , i.e. there exists a point at which condition (A) is fulfilled.
Condition (A) and the weaker condition d(x) > 0 are thus “inhomogeneous properties”
on M+ .

Remark. Condition (A) holds at certain points of M− in both (8, 7) families (see
Section 6.6).

Proof. We first show the second assertion is true. For m1 = 3, max d = 3 by 5.1 and
5.2.
For m1 = 7, choose x ∈ S 2l−1 as a common eigenvector of the commuting 4-
products
P0 P1 P2 P3 , P0 P1 P4 P5 , P0 P1 P6 P7 , P0 P2 P4 P6 .

Since each Pi anti-commutes with at least one of these operators, we get x ∈ M+ . And
since x is also an eigenvector of the product of the above-mentioned operators, one
obtains the following identities.

From 4.5(ii) each of these vectors lies in the kernel of each of the second funda-
mental tensors SPi x ; from 3.7(v), these vectors are also orthonormal. Thus d(x) = 7.
Now let m1 = 4µ − 1 for any µ. Choose x ∈ S 2l−1 to be a common eigenvector of
the commuting operators

P0 · · · Pm1 and P2i P2i+1 P2j P2j+1 , 0 ≤ i < j ≤ 2µ − 1.

Then such an x ∈ M+ and P0 P1 x = ±P2i P2i+1 x lies in the intersection of all the
kernels. Thus max d > 0.

18
Finally, if d(x) > 0, then there exists Q0 , ..., Qm ∈ Σ with

P0 Q0 x = P1 Q1 x = · · · = Pm Qm x,

and hPi , Qi i = 0. After modifying P1 , ..., Pm by an orthogonal transformation, if neces-


sary, one can assume that P1 = Q0 and Q1 = −P0 . Now it follows for pairwise distinct
indices i, j, k that Pi Pj x = Pk Qk x, but

0 = hPi Pk x, Pi Pj xi = hPi Pk x, Pk Qk xi = −hPi x, Qk xi = −hPi , Qk i.

Thus Q2 , ..., Qm are orthogonal to P0 and P1 , and by the same argument, one can
assume as above that P3 = Q2 and P2 = −Q3 . By repetition of of this last procedure,
one obtains

P2i P2i+1 x = ±P2j P2j+1 x f or all i, j ∈ {0, ..., 2µ − 1}.

Thus P0 · · · Pm1 x = ±x. Thus if d(x) > 0, then x ∈ E± (P0 · · · Pm1 ). These eigenspaces
depend only on Σ and intersect S 2l−1 in spheres of dimension l−1 = m1 +m2 < dim M+ .
Thus there exists x ∈ M+ with d(x) = 0.

5.9 The condition (B) of Takeuchi-Ozeki. With the notation from 5.8, let x ∈ M+
and d(x) = m1 , i.e. condition (A) holds at x. One splits R2l = Rx⊕ ⊥x M+ ⊕ K ⊕ B
where K, resp. B, are the common kernel, resp. the common image, of the second
fundamental tensors at x, and one splits y ∈ R2l correspondingly as y = λx + N + k + b,
then one can find F (y) on the individual summands using the degree of homogeneity.
For the term that is constant on Rx and linear on ⊥x M+ one obtains:
X
−8 hPi N, kihPi b, bi.

This is just condition (B) which Takeuchi and Ozeki examined, see [5]. The main
result of their work is the classification of the (g = 4)-isoparametric families for which
conditions (A) and (B) hold at a point of a focal manifold. Together with 5.8, we thus
obtain:
Theorem: The Clifford series of multiplicities (3, 4k) and (7, 8k) are just the inho-
mogeneous series of Takeuchi and Ozeki [5].

6. Clifford-examples with small m1 . The arguments of the last section assumed


often that m1 was not too small. We will now handle a series of questions about the
exceptional cases–which have yet to be answered.

6.1 The homogeneous Clifford series.

19
Theorem: The Clifford isoparametric families of multiplicities (1, k), (2, 2k − 1) and
the families of multiplicity (4, 4k − 1) with P0 · · · P4 = ±Id are homogeneous.
Proof. Let F ∈ {R, C, H} and m := dimR F. Let e1 , ..., em−1 be the canonical imaginary
units of F and Ej : Fn → Fn the left multiplication by ej . Then as in 3.3, E1 , ..., Em−1
induce a Clifford system (P0 , ..., Pm ) on R2l = F2n , where l = n · m, n ≥ 2. One sets
c1 := 1, cj := ej−1 for j ∈ {2, ..., m} and thus the isoparametric function for (P0 , ..., Pm )
is given by
n X o
F (u, v) = (kuk2 + kvk2 )2 − 2 (kuk2 − kvk2 )2 + 4 hu, ci vi2 f or (u, v) ∈ Fn ⊕ Fn .

F is invariant under the following sets of isometries.

I1 := {cos t P0 + sin t P1 | t ∈ R},


I2 := {Id ⊕ αId | α ∈ F, |α| = 1},
I3 := {A ⊕ A | A ∈ U (n, F)}.

The invariance under I1 holds for any Clifford system, see 3.7(iii), invariance under
I2 and I3 is essentially based on the special form of F. In particular, the invariance
under I2 follows from the fact that the ci v constitute an orthogonal basis of Fv. Let
(u, v) ∈ S 2l−1 . From the invariance under I1 , we can assume without loss of generality
that kuk = kvk. The invariance under I3 further allows us to take without loss of

generality u = (1, 0, ..., 0)/ 2 and using the invariance under I2 , take v = (v1 , ..., vn )
with v1 ∈ (0, ∞). Finally, by repeated use of I3 , we get that each (u, v) ∈ S 2l−1 lies on
the same F -level as a point of the form
  √ π
(ū, v̄) = (1, 0, ..., 0), (cos t, sin t, 0, ..., 0) / 2, t ∈ [0, ].
2
But F (ū, v̄) = − cos 2t, and thus the isometries I1 , I2 , I3 operate transitively on the
levels of F , and the family is homogeneous. For F = R, C, H one gets a family with
m1 = 1, 2, 4, where in the last case P0 · · · P4 = −Id.

6.2 The canonical Killing fields. We will frequently make use of the following well-
known fact: If P, Q ∈ Σ are orthonormal, then P Q is skew-symmetric and x 7→ P Qx
is thus a Killing field on S 2l−1 . But from 3.7(iii) we have

F ((cos t Q + sin t P )Qx) = F (Qx)

and thus the Killing field is tangential to the levels of F . By the way, it is easy to
show that the vector space spanned by the products P Q with P, Q ∈ Σ orthonormal is
a Lie subalgebra of so(2l) isomorphic to so(m + 1), i.e. Spin(m + 1) operates on the
isoparametric family through isometries.

20
6.3 The homogeneity of the (9, 6) family. From 4.4 there exists a homogeneous
family with g = 4 and multiplicities 9 and 6, whose order depends on the orientation
so that we cannot say whether m1 is 6 or 9. On the other hand, from 4.3 there exist
Clifford families of multiplicities (6, 9) and (9, 6). The first is not homogeneous by 5.4.
We show that the other is homogeneous.
Theorem: The Clifford family with multiplicities (9, 6) is homogeneous. Thus, in
particular, it is not congruent to the (6, 9) family.
Proof. For the Clifford system (P0 , ..., P9 ) on R32 , we choose x ∈ S 31 to be a common
eigenvector of the commuting operators

P2i P2i+1 P2j P2j+1 , 0 ≤ i < j ≤ 4.

We set z(t) = cos t x + sin t P0 x and

A(t) := Span{P0 P1 z(t), P2 P3 z(t)}


B(t) := Span{P1 P2i z(t), P1 P2i+1 z(t), P0 P2i z(t), P0 P2i+1 z(t) | 1 ≤ i ≤ 4}
C(t) := Span{P2i P2j z(t), P2i P2j+1 z(t) | 1 ≤ i < j ≤ 4}

Then A(t), B(t), C(t) are subspaces of the tangent space to the F -level at z(t) which
are pairwise orthogonal to one another and one finds:
π
dim A(0) = 1, dim A( ) = 2
4
π
dim B(0) ≥ 8, dim B( ) = 16
4
π
dim C(0) = 12, dim C( ) ≥ 6.
4

Moreover, one notices that the generating vectors in C(0) and B( π4 ) are all or-
thogonal; in the other cases one finds sufficiently many (orthonormal) vectors of the
form Pr Ps z(t) with fixed r. Both inequalities are actually equalities since z(0) ∈ M+21 ,
z( π4 ) ∈ M−24 . The set of t-values with dim(A(t) ⊕ B(t) ⊕ C(t)) < 2 + 16 + 12 = 30
has only isolated points. Thus one finds a family of hypersurfaces with 30 linearly
independent Killing fields at a point. Thus the whole family is homogeneous.

6.4 The homogeneity of M− for the (3, 4k)-families. The families of multiplicities
(3, 4k) are inhomogeneous and by 5.2 their focal manifold M+ is also inhomogeneous.
In contrast, we have
Theorem: The focal manifolds M− of the families of multiplicities (3, 4k) are homo-
geneous.

21
Proof. We will use the notation of the proof of 6.1 with F = H and consider the Clifford
system (P1 , ..., P4 ). The corresponding function is
4
X
F (u, v) = (kuk2 + kvk2 )2 − 8 hu, ci vi2 .
i=1

For (u, v) ∈ M− , we have kuk = kvk, and so F is invariant under the set of isometries
I2 and I3 , and one obtains the conclusion as in 6.1.

6.5 Coincidence of the Clifford examples.


Theorem: The Clifford families with multiplicities (2, 1), (6, 1), (5, 2) are congruent
to those with multiplicities (1, 2), (1, 6), (2, 5). The (4, 3) family with with P0 · · · P4 6=
±Id is congruent to the family with multiplicities (3, 4), and thus from 5.2 it is inho-
mogeneous and hence by 6.1, it is not congruent to the other (4, 3) family.
Proof. For the Clifford system (P0 , ..., P8 ) on R16 we have from 4.2(i), case m2 < 0,
8
X
2 hPi x, xi2 = 2hx, xi2
i=0

and so
k
X  8
X 
hx, xi2 − 2 hPi x, xi2 = − hx, xi2 − 2 hPi x, xi2 .
i=0 i=k+1

For k ∈ {4, 5, 6}, P0 , ..., Pk and Pk+1 , ..., P8 are Clifford systems on R16 and in the case
k = 4 we have P0 · · · P4 6= ±Id, since this product anti-commutes with P5 . Thus the
result follows in the cases with m1 + m2 = 7. The (2, 1)-case can be proven analogously.

6.6 Families with multiplicities (8, 7). We will call the family with P0 · · · P8 = ±Id,
where without loss of generality we take the + sign, the definite family; the other we
will call the indefinite family.
Theorem: (i) For the indefinite (8, 7)-family, both focal manifolds are inhomoge-
neously embedded.
(ii) For the definite (8, 7)-family, M+ is homogeneously embedded, M− inhomogeneously
embedded. This and (i) imply that the two (8, 7) families are not congruent to one
another.
(iii) Neither (8, 7) family is congruent to the (7, 8) family.
Proof. (i) From the classification in 3.5, we can extend a Clifford system (P0 , ..., P8 )
on R32 to a system (P0 , ..., P9 ), and we can use the considerations from the proof of
6.3. Let x ∈ S 31 again be a common eigenvector of P2i P2i+1 P2j P2j+1 , 0 ≤ i < j ≤ 4.

22
Then, in particular, P2i P9 x = ±P2i+1 P8 x and the right side is a Killing field of our
(8, 7)-family. Thus, with the same definitions as in 6.3, one obtains 21 Killing fields
on M+ , that all lie in the span of the kernels of the second fundamental tensors, see
4.5(ii), and at x they span a 21-dimensional space A(0) ⊕ B(0) ⊕ C(0). Thus with

σ(x) := dim Span{v ∈ Tx M+ | there exists N ∈⊥x M+ \ {0}, SN v = 0}

we have σ(x) ≥ 21.


On the other hand, we choose u ∈ S 31 to be a (+1)-eigenvector of P9 , so that by
4.2(ii), u ∈ M+ . Further, we have P9 (Pi Pj u) = Pi Pj u for all i, j ∈ {0, ..., 8}. Thus
from 4.5(ii)
σ(u) ≤ dim E+ (P9 ) = 16.

Thus M+ is inhomogeneously embedded. Note that for x as chosen above we have


σ(x) = 21 = dim M+ − 1, and so the isometry group of M+ has orbits of codimension
1.

To study M− , we first choose x as above. Then y = (x + P0 x)/ 2 ∈ M− with
P0 y = y. Obviously y is likewise an eigenvector of the 4-products P2i P2i+1 P2j P2j+1
with 0 ≤ i < j ≤ 4. With the help of 3.7(iv), one verifies for 1 < i < j ≤ 9 and
1 ≤ k ≤ 9 that

hP1 Pi Pj y, Pk yi = 0
hP0 P9 y, Pk yi = 0, k 6= 9.

Thus, one sees with 4.5(iii) that for a fixed i ∈ {2, ..., 9}

⊥y M− = Span{P0 P9 y, P1 Pi Pj y | 2 ≤ j ≤ 9, j 6= i},

and the 8 listed vectors make an orthonormal basis. Further, it follows from 4.5(iii)
that

P1 Pi y ∈ ker SP0 P9 y , 2 ≤ i ≤ 8
P2 Pr y ∈ ker SP1 P2 Pj y , r, j ∈ {3, ..., 9}, r 6= j,

and that these 14 vectors are pairwise orthogonal. With σ defined as above for M−
instead of M+ , we thus have σ(y) ≥ 14. (By the way, one can show that P1 P9 y is
perpendicular to all the kernels, and thus σ(y) = 14. As for M+ , one can further show
that on the 23-dimensional M− , there are 22 Killing fields that are linearly independent
at y.)

23
Translator’s Note: Regarding the statement in the parentheses above, by the last line
above 4.6 one gets that
 
Span ker(SN ) ⊂ E+ (P0 ) ∩ y ⊥ ,

which has dimension 15. Since the vector P1 P9 y is in E+ (P0 ) ∩ y ⊥ and is perpendicular
to all the kernels, one gets σ(y) ≤ 14. Since σ(y) ≥ 14 has already been shown, it
follows that σ(y) = 14.

Finally, we show that there is a point v on M− with σ(v) < 14. Let v ∈ S 31 ∩
E+ (P0 ) ∩ E+ (P1 · · · P8 ). Since P0 and P1 · · · P8 commute, but P9 and P0 anti-commute
while P9 and P1 · · · P8 commute, there exists such a v in M− by 4.2(i). Further it
follows that the eigenspaces of P0 and P1 · · · P8 have 8-dimensional intersection. For
N ∈ E− (P0 ) ∩ E+ (P1 · · · P8 ) we have

hN, Pk vi = 0, 1 ≤ k ≤ 8,

since Pk v ∈ E− (P0 ) ∩ E− (P1 · · · P8 ), as Pk anti-commutes with both operators. Thus


on dimensional grounds,

⊥v M− = E− (P0 ) ∩ E+ (P1 · · · P8 ).

For N ∈⊥v M− \{0}, we have

ker SN = {w ∈ E+ (P0 ) | 0 = hw, vi = hw, P1 N i = ... = hw, P8 N i}.

But since Pi N ∈ E+ (P0 ) ∩ E− (P1 · · · P8 ), it follows on dimensional grounds that

ker SN = {w ∈ E+ (P0 ) ∩ E+ (P1 · · · P8 ) | hw, vi = 0}.

Since the kernel of N is not dependent on N , condition (A) holds at v and σ(v) = 7.
Thus M− is inhomogeneous.
(ii) Without loss of generality, we can take P0 · · · P8 = +Id. We first show that
there exists

x ∈ M+ ∩ E+ (P2 P4 P6 P8 ) ∩ E+ (P3 P4 P7 P8 ) ∩ E+ (P5 P6 P7 P8 ).

Since P2 P4 P6 P8 anti-commutes with P2 , E+ (P2 P4 P6 P8 ) has dimension 16. It


is an invariant subspace of the anti-commuting operators P3 P4 P7 P8 and P3 . Thus
E+ (P2 P4 P6 P8 ) ∩ E+ (P3 P4 P7 P8 ) is of dimension 8 and further it is an invariant sub-
space of the anti-commuting operators P5 P6 P7 P8 and P5 . Thus E+ (P2 P4 P6 P8 ) ∩
E+ (P3 P4 P7 P8 ) ∩ E+ (P5 P6 P7 P8 ) is of dimension 4 and on this space, we have
1
X
F (x) = hx, xi2 − 2 hPi x, xi2 .
i=0

24
This function is not constant and a maximum point lies in M+ . We choose such an
x. This is then an eigenvector of P3 P4 P5 P6 , and since P0 · · · P8 = Id, of P0 P1 P2 P2i−1 P2i ,
i ∈ {2, 3, 4}. Thus as in the proof of 5.8, one obtains the following identities.

These vectors are obviously orthonormal, and with 4.2(ii) and 2.7(iv) one shows
that these can be completed to form an orthonormal basis of Tx M+ by adjoining

P0 P1 x, ..., P0 P8 x, P1 P2 x, ..., P1 P8 x.

Since the vectors are all values at x of Killing vector fields, it follows that M+ is a
homogeneous submanifold.
To study M− , we first choose y ∈ S 31 as common eigenvector of the commuting
4-products
P0 P1 P2 P3 , P0 P1 P4 P5 , P0 P1 P6 P7 , P0 P2 P4 P6 ,

as in the proof of 5.8. Then without restriction P8 y = y and y ∈ M− .


Let N ∈⊥y M− , k ∈ {0, ..., 7} and i ∈ {1, ..., 7}. Then from the identities in 5.8,
there exists a j ∈ {1, ..., 7} with

P0 Pi y = ±Pk Pj y.

Thus hP0 Pi y, Pk N i = ±hPj y, N i = 0, and so P0 P1 y, ..., P0 P7 y ∈ ker SN , and the 7-


dimensional kernel is independent of N 6= 0, and condition (A) holds at y: so

σ(y) = 7.

To find a point v ∈ M− with other behavior of the kernel, we first establish that
from 3.5 there exist endomorphisms P˜0 , P˜9 such that {P˜0 , P1 , ..., P8 , P˜9 } is a Clifford
system. We then show that there exists a point
4
\
v ∈ S 31 ∩ E+ (P0 ) E+ (P˜0 , P˜9 P2i−1 P2i ).
1

The 16-dimensional eigenspace E+ (P0 ) is invariant under the anti-commuting operators


P˜0 and P˜0 P˜9 P1 P2 ( note P0 = P1 · · · P8 ). Thus E+ (P0 )∩E+ (P˜0 P˜9 P1 P2 ) is 8-dimensional

25
and further it is an invariant subspace of the anti-commuting operators P˜0 P˜9 P3 P4
and P4 P5 . Thus E+ (P0 ) ∩ E+ (P˜0 P˜9 P1 P2 ) ∩ E+ (P˜0 P˜9 P3 P4 ) has dimension 4 and is
invariant under the anti-commuting operators P˜0 P˜9 P5 P6 and P6 P7 . Thus E+ (P0 ) ∩
· · · ∩ E+ (P˜0 P˜9 P5 P6 ) has dimension 2. But from P0 = P1 · · · P8 , it follows that it is
likewise contained in E+ (P˜0 P˜9 P7 P8 ). Thus there exists such a v, which then naturally
lies in M− . For such a v, it follows that hP˜0 Pi v, Pk vi = 0 for all i, k ∈ {1, ..., 8}. Thus
from 4.5(iii) we have
⊥v M− = Span{P˜0 P1 v, ..., P˜0 P8 v},

and in particular, P5 P˜0 P1 v ∈ (ker SP˜0 P1 v )⊥ .


We show, on the other hand, that this vector lies in ker SP˜0 P3 v . Then the kernels
are not all the same at v and σ(v) > 7.
We must from 4.5(iii) show that

hP5 P˜0 P1 v, Pk P˜0 P3 vi = hPk P1 P3 P5 v, vi = 0,

for all k ∈ {1, ..., 8}. But by the choice of v, we have

hPk P1 P3 P5 v, vi = −hPk P2 P˜0 P˜9 P4 P˜0 P˜9 P6 P˜0 P˜9 v, vi = hPk P2 P4 P6 P˜0 P˜9 v, vi = 0,

from 3.7(iv) in the cases k 6∈ {2, 4, 6}. But if k = 2, then

hP2 P1 P3 P5 v, vi = hP3 P5 P2 P1 v, vi = hP3 P5 P˜9 P˜0 v, vi = −hP3 P6 v, vi = 0,

likewise from 3.7(iv). One shows this analogously for k = 4, 6. Thus σ(v) > 7 and M−
is inhomogeneously embedded.
(iii) Were an (8, 7)-family congruent to a (7, 8)-family, this would give Clifford
systems (P0 , ..., P8 ) and (P9 , ..., P16 ) on R32 with
8
X  16
X 
hx, xi2 − 2 hPi x, xi2 = − hx, xi2 − 2 hPi x, xi2
0 9

or
16
X
hPi x, xi2 = hx, xi2 .
0

We now show that (P0 , ..., P16 ) is a Clifford system on R32 , in contradiction to 3.5.
Through differentiation, one obtains
16
X
hPi x, xiPi x = hx, xix,
0

26
and for i ∈ {0, ..., 8} and u ∈ E± (Pi )
16
X
hu, uiu + hPi u, uiPi u = hu, uiu,
9

thus
hP9 u, ui = ... = hP16 u, ui = 0.

Thus for x = u + v with u ∈ E+ (Pi ), v ∈ E− (Pi ) and j ∈ {9, ..., 16}

hPi Pj x, xi = hPj (u + v), Pi (u + v)i = hPj u, ui − hPj v, vi = 0,

and so Pi Pj + Pj Pi = 0, and thus (P0 , ..., P16 ) is a Clifford system. Contradiction!

6.7 Summary. The results 4.6, 6.1, 6.3, 6.5 and 6.6 show: The congruence classes of
the families with multiplicities (1, 2), (1, 6), (2, 5), (3, 4) each occur a second time in
the Table 4.3, all others are listed only once.

References

[1] E. Cartan, Sur des familles remarquables d’hypersurfaces isoparamétriques dans les
espaces sphériques, Math. Z. 45 (1939), 335–367.

[2] W. Y. Hsiang and H. B. Lawson, Minimal submanifolds of low cohomogenity, J.


Diff. Geom. 5 (1971), 1–38.

[3] D. Husemoller, Fiber bundles. New York-San Francisen-London, MeGraw Hill 1966.

[4] H.F. Münzner, Isoparametric Hyperflächen in Sphären, I and II, Math. Ann. 251
(1980), 57–71 and 256 (1981), 215–232.

[5] H. Ozeki and M. Takeuchi, On some types of isoparametric hypersurfaces in spheres,


I and II, Tohoku Math. J. 27 (1975), 515–559 and 28 (1976), 7–55.

[6] R. Takagi and T. Takahashi, On the principal curvatures of homogeneous hypersur-


faces in a sphere, Differential Geometry, in honor of K. Yano, Kinokuniya, Tokyo,
1972.

Received on 24. September 1980

Added in proof
We just now received:

27
Dorfmeister, J., Neher, E.: Isoparametric triple systems of algebra type, preprint.

A major result is: In the case g = 4, it follows from Condition (A) that the isoparametric
families are either Clifford or that (m1 , m2 ) = (2, 2).

28

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy