Homogeneous Structures on Submanifolds 203
6.3 Isoparametric submanifolds of higher rank
In Section 4 .4 we introduced Thorbergsson’s Theorem 4.4.5, which states that
any irreducible full isoparametric submanifold of R
n
of rank at least three is an orbit
of an s-representation. The original proof by Thorbergsson in [319] uses Tits build-
ings and the Homogeneous Slice Theorem 4.3.6. There is an alternative proof of
Thorbergsson’s result by Olmos [256] using the theory of homogeneous structures
on submanifolds and normal holonomy. The idea is as follows. We know from The-
orem 6.2.5 that if there exists on a connected full compact submanifold M of R
n
a
canonical connection ∇
c
of type T ,thenM is an orbit of an s-representation. Given
an irreducible full isoparametric subm anifold of R
n
of codimension at least three,
we can focalize at the same time any two curvature distributions. The corresponding
fibers are, by the Homogeneous Slice Theorem, orbits of s-representations. A canon-
ical connection ∇
c
on M is constructed by gluing together the canonical connections
that occur naturally on these fibers. The proof of the compatibility between these
canonical connections is based on a relation between the normal holonomy g roups
of the different focal manifolds. The common eigenspaces of the shape operators
of M are parallel with respect to the canonical connection. This implies readily that
∇
c
α
= 0. To show that ∇
c
(∇−∇
c
)=∇
c
S
c
= 0 we have to use the geometric fact that
the ∇
c
-parallel transport alon g a horizontal cu rve with re spect to some focalization
equals the ∇
⊥
-parallel displacement in the focal manifold along the p rojection of the
curve.
6.3.1 The canonical connection on orbits of s-representations
To motivate the definition of canonical connection we discuss what properties we
would like it to have. With this in mind, we start with a homogeneous isoparametric
submanifold M.ThenM is a principal orbit K ·v of an s-representation, that is, of
the isotropy representation of a symmetric space G/K. Recall from Section 2.3 that
there is an orthogonal reductive decomposition k = k
v
⊕k
+
,wherek
+
=
∑
α
∈Ψ
+
k
α
and k
α
= {x ∈ k :ad(H)
2
x =
α
2
(H)x for all H ∈ a}. Recall also from Section 2.7
that the common eigenspaces of the shape operators of M are E
α
= p
α
+ p
2
α
,where
p
2
α
= {0} if 2
α
is not a restricted root. In particular, they correspond to positive
roots
α
of {ad(H)
2
: H ∈ a}.
Consider a focal orbit M
ξ
= K ·u with u = v +
ξ
(v) and define
¯
E = ker(A
ξ
−id).
Note that
¯
E corresponds to a sum f of restricted root spaces. Then k
u
= k
v
⊕f and k =
k
u
⊕n are r eductive decomposition of k
u
and k, respectively, where n is the orthogonal
complement of k
u
in k.
Proposition 6.3.1 Let M be an orbit of an s-representation, ∇
c
the canonical con-
nection on M associated with the reductive decomposition k = k
u
⊕n and
¯
∇
c
the