Abstract
It is known that an isometric action of a Lie group on a compact symmetric space gives rise to a proper Fredholm action of a path group on a path space via the gauge transformations. In this paper, supposing that the isometric action is a Hermann action (i.e., an isometric action of a symmetric subgroup of the isometry group), we give an explicit formula for the principal curvatures of orbits of the path group action and study the condition for those orbits to be austere, that is, the set of principal curvatures in the direction of each normal vector is invariant under the multiplication by − 1. To prove the results, we essentially use the facts that Hermann actions are hyperpolar and all orbits of Hermann actions are curvature-adapted submanifolds. The results greatly extend the author’s previous result in the case of the standard sphere and show that there exist a larger number of infinite dimensional austere submanifolds in Hilbert spaces.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In [24] and [29], Palais and Terng introduced a class of proper Fredholm (PF) submanifolds in Hilbert spaces. These are submanifolds in Hilbert spaces which have finite codimensions and generalize properly immersed submanifolds in Euclidean spaces. By definition, the shape operators of PF submanifolds are compact self-adjoint operators. Moreover, the infinite dimensional differential topology and Morse theory [23, 26, 27] can be applied to PF submanifolds. Typical examples of PF submanifolds are orbits of the gauge transformations. More precisely, let G be a connected compact Lie group with a bi-invariant Riemannian metric, \(\mathfrak {g}\) its Lie algebra, and P the trivial principal G-bundle over the circle S1. The loop group H1(S1,G) of Sobolev H1-loops is the gauge group of P and it acts on the Hilbert space \(L^{2}(S^{1}, \mathfrak {g})\) of L2-connections by the gauge transformations:
They showed that this action is isometric, proper and Fredholm (PF) [24]. Thus, every orbit of this action is a PF submanifold of \(L^{2}(S^{1}, \mathfrak {g})\). Moreover, they essentially considered an equivariant map \({\Phi }:L^{2}(S^{1}, \mathfrak {g}) \rightarrow G\), which is nowadays called the parallel transport map, and showed that the above gauge group action is hyperpolar and the principal orbits are isoparametric PF submanifolds. Here, the action is called hyperpolar if there exists a closed affine subspace Σ of \(L^{2}(S^{1}, \mathfrak {g})\) which meets every orbit orthogonally.
The above examples were extended by Pinkall and Thorbergsson [25] and eventually reformulated by Terng [30] as follows. Let \(\mathcal {G} := H^{1}([0,1], G)\) denote the path group of all H1-paths from [0,1] to G and \(V_{\mathfrak {g}} := L^{2}([0,1], \mathfrak {g})\) the Hilbert space of all L2-paths from [0,1] to \(\mathfrak {g}\). For any closed subgroup L of G × G, the subgroup
acts on \(V_{\mathfrak {g}}\) isometrically by the gauge transformations. If L is the diagonal ΔG, then the P(G,L)-action is identified with the action given by Palais and Terng above. If L = K × K for a symmetric subgroup K of G, then it is just the one given by Pinkall and Thorbergsson [25]. Terng [30] generally proved that the P(G,L)-action is PF and showed that the parallel transport map \({\Phi } : V_{\mathfrak {g}} \rightarrow G\) is equivariant with respect to the L- and P(G,L)-actions where L acts on G by (b,c) ⋅ a := bac− 1, and each P(G,L)-orbit is expressed as the inverse image of an L-orbit under Φ. Using these results, she showed that if the L-action is hyperpolar then the P(G,L)-action is also hyperpolar and the principal orbits are isoparametric. Here, the L-action (or more generally, a proper isometric action on a Riemannian manifold) is called hyperpolar if there exists a closed connected totally geodesic submanifold Σ which is flat in the induced metric and meets every orbit orthogonally [6]. In this way, the structure of the P(G,L)-action is understood through the parallel transport map.
Terng and Thorbergsson [31] investigated the parallel transport map \({\Phi }: V_{\mathfrak {g}} \rightarrow G\) and showed that it is a Riemannian submersion. Moreover, they gave an interesting application to the submanifold geometry in symmetric spaces. Let G/K be a symmetric space of compact type with projection \(\pi : G \rightarrow G/K\). They considered the composition \({\Phi }_{K} := \pi \circ {\Phi } : V_{\mathfrak {g}} \rightarrow G \rightarrow G/K\) and proved that if N is a closed submanifold of G/K then the inverse image \({\Phi }_{K}^{-1}(N)\) is a PF submanifold of \(V_{\mathfrak {g}}\). Moreover, they showed that if N is an equifocal submanifold of G/K then each component of \({\Phi }_{K}^{-1}(N)\) is an isoparametric PF submanifold of \(V_{\mathfrak {g}}\) and the converse is also true. Although \({\Phi }_{K}^{-1}(N)\) is infinite dimensional, many techniques and results in the finite dimensional Euclidean case are still valid in the case of Hilbert space \(V_{\mathfrak {g}}\). Applying techniques for isoparametric submanifolds to \({\Phi }_{K}^{-1}(N)\), they studied equifocal submanifolds in G/K. In this way, the parallel transport map is also known as a tool to study the submanifold geometry in symmetric spaces. It is a fundamental problem to show the geometrical relation between N and \({\Phi }_{K}^{-1}(N)\).
Afterwards, Koike [13] gave a formula for the principal curvatures of \({\Phi }_{K}^{-1}(N)\) under the assumption that N is curvature-adapted, that is, for each normal vector v at each p ∈ N the Jacobi operator Rv leaves the tangent space TpN invariant and the restriction \(R_{v}\vert _{T_{p} N}\) commutes with the shape operator Av. Recently, the author [18] corrected inaccuracies in that formula and studied the relation between two conditions:
-
(A)
N is an austere submanifold of G/K,
-
(B)
\({\Phi }_{K}^{-1}(N)\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
Here, a submanifold is called austere [4] if the set of principal curvatures with multiplicities in the direction of each normal vector is invariant under the multiplication by (− 1). The author showed [18, Theorem 4.1]:
Theorem 1
[18] If G/K is the standard sphere then (A) and (B) are equivalent.
The purpose of this paper is to extend this theorem to the case that G/K is not necessarily the standard sphere. However there are two difficulties to do this. The first one is that N must be curvature-adapted in order to use the formula for the principal curvatures of \({\Phi }_{K}^{-1}(N)\), otherwise there is no way to compute those curvatures. The second one is that even if N is curvature-adapted the principal curvatures of \({\Phi }_{K}^{-1}(N)\) and their multiplicities are complicated in general and it is not clear whether the austere properties of N and \({\Phi }_{K}^{-1}(N)\) are equivalent or not.
In this paper, we let G/K be a symmetric space of compact type and suppose that N is an orbit of a Hermann action [7], that is, an isometric action of a symmetric subgroup H of G on G/K. We know that all orbits of Hermann actions are curvature-adapted submanifolds [2]. Moreover, we can explicitly describe the principal curvatures of orbits of Hermann actions [22]. Furthermore, \({\Phi }_{K}^{-1}(N)\) is expressed as an orbit of the P(G,H × K)-action and the conditions (A) and (B) are restated as follows:
-
(A)
The orbit \(H \cdot (\exp w) K\) through \((\exp w) K\) is an austere submanifold of G/K,
-
(B)
The orbit \(P(G, H \times K)* \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\),
where \(w \in \mathfrak {g}\) and \(\hat {w}\) denotes the constant path with value w. Note that since the Hermann action is hyperpolar [8], the P(G,H × K)-action is also hyperpolar. This implies that we only have to consider normal vectors which are tangent to a fixed Σ when studying the above conditions (see Lemma 7.2).
In this paper, we first derive an explicit formula for the principal curvatures of P(G,H × K)-orbits (Theorem 6.1), which unifies and generalizes some results by Terng [29], Pinkall-Thorbergsson [25], and Koike [14] (see Remark 6.4). Then, using this explicit formula, we study the relation between (A) and (B). To explain the results, we write σ and τ for the involutions of G associated to the symmetric subgroups K and H respectively. We denote by \(\mathfrak {g} = \mathfrak {k} \oplus \mathfrak {m}\) (resp. HCode \(\mathfrak {g} = \mathfrak {h} \oplus \mathfrak {p}\)) the decomposition into the (± 1)-eigenspaces of the differential of σ (resp. τ). Take a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\) and denote by Δ = Δ(σ,τ) the associated root system of \(\mathfrak {t}\). We will prove the following theorem (Theorem 7.1):
Theorem I 1
If Δ is a reduced root system, then (A) and (B) are equivalent.
Then, without supposing that Δ is a reduced root system, we will prove the following theorem (Theorems 8.2, 8.5, and 8.8):
Theorem II 1
-
(i)
Suppose that σ = τ. Then (A) and (B) are equivalent.
-
(ii)
Suppose that σ and τ commute. Then (A) implies (B).
-
(iii)
Suppose that G is simple. Then (A) implies (B).
Note that (B) does not imply (A) in the cases (ii) and (iii). In fact, we will show a counterexample of a minimal H-orbit which is not austere but the corresponding P(G,H × K)-orbit is austere (cf. Section 9). Without the assumption of (ii) or (iii), we do not know whether (A) implies (B) or not because in the non-simple case there exist many non-commutative pairs of involutive automorphisms of G [17]. However, the above theorems greatly extend the previous theorem in the spherical case and cover all known examples of austere orbits of Hermann actions [9, 22]. Applying the above theorems to those examples, we obtain a larger number of infinite dimensional austere PF submanifolds in Hilbert spaces. Notice that so obtained austere PF submanifolds are not totally geodesic due to [19].
This paper is organized as follows. In Section 2, we review basic knowledge on P(G,H × K)-actions and the parallel transport map. In Section 3, we review fundamental results on the submanifold geometry of orbits of Hermann actions. In Section 4, we introduce a hierarchy of curvature-adapted submanifolds in symmetric spaces and formulate the curvature-adapted property of orbits of Hermann actions. In Section 5, we refine the formula for the principal curvatures of \({\Phi }_{K}^{-1}(N)\) [13, 18] so that it can be applied to orbits of Hermann actions. In Section 6, by applying the refined formula to orbits of Hermann actions, we derive an explicit formula for the principal curvatures of P(G,H × K)-orbits. In Section 7, using this explicit formula, we formulate the conditions (A) and (B) in terms of roots in Δ and prove Theorem I. In Section 8, we show an inequality between the multiplicities of roots α and 2α in Δ and prove Theorem II. Finally, in Section 9, we show a counterexample to the converse of (ii) and (iii) of Theorem II and mention further remarks on the converse.
2 Preliminaries
Let G be a connected compact semisimple Lie group and K a closed subgroup of G. Suppose that K is a symmetric subgroup of G, that is, there exists an involutive automorphism σ of G satisfying the condition \(G^{\sigma }_{0} \subset K \subset G^{\sigma }\), where Gσ denotes the fixed point subgroup of G and \(G^{\sigma }_{0}\) its identity component. We denote by \(\mathfrak {g}\) and \(\mathfrak {k}\) the Lie algebras of G and K respectively. The differential of σ induces an involutive automorphism of \(\mathfrak {g}\), which is still denoted by σ. The direct sum decomposition \(\mathfrak {g} = \mathfrak {k} \oplus \mathfrak {m}\) into the (± 1)-eigenspaces of σ is called the canonical decomposition. We fix an Ad(G)-invariant inner product 〈⋅,⋅〉 of \(\mathfrak {g}\) which is a negative multiple of the Killing form of \(\mathfrak {g}\). Then, it is invariant under all automorphisms of \(\mathfrak {g}\) and the canonical decomposition is orthogonal. We equip G with the corresponding bi-invariant Riemannian metric and the homogeneous space G/K with the corresponding G-invariant metric. Then M := G/K is a symmetric space of compact type and the projection \(\pi : G \rightarrow M\) is a Riemannian submersion with totally geodesic fiber.
We denote by \(\mathcal {G} := H^{1}([0,1], G)\) the path group of all Sobolev H1-paths from [0,1] to G and by \(V_{\mathfrak {g}} := L^{2}([0,1], \mathfrak {g})\) the path space of all L2-paths from [0,1] to \(\mathfrak {g}\). Then \(\mathcal {G}\) is a Hilbert Lie group and \(V_{\mathfrak {g}}\) a separable Hilbert space. We consider the isometric action of \(\mathcal {G}\) on \(V_{\mathfrak {g}}\) given by the gauge transformations
where \(g \in \mathcal {G}\) and \(u \in V_{\mathfrak {g}}\). We know that this action is proper and Fredholm [24, Theorem 5.8.1], [29, Section 4]. For any closed subgroup L of G × G the subgroup
acts on \(V_{\mathfrak {g}}\) by the same formula. It follows that the P(G,L)-action is also proper and Fredholm [30, p. 132]. Thus every orbit of the P(G,L)-action is a proper Fredholm (PF) submanifold of \(V_{\mathfrak {g}}\) [24, Theorem 7.1.6]. We know that the P(G,{e}× G)-action on \(V_{\mathfrak {g}}\) is simply transitive [31, Corollary 4.2] and that the P(G,G ×{e})-action on \(V_{\mathfrak {g}}\) is also simply transitive [19, Section 5].
The parallel transport map [30, 31] \({\Phi }: V_{\mathfrak {g}} \rightarrow G\) is defined by
where \(g_{u} \in \mathcal {G}\) is the unique solution to the linear ordinary differential equation
We know that Φ is a Riemannian submersion and a principal Ωe(G)-bundle, where Ωe(G) = P(G,{e}×{e}) denotes the based loop group [31, Corollary 4.4, Theorem 4.5]. The normal space of the fiber Φ− 1(e) at \(\hat {0} \in V_{\mathfrak {g}}\) is identified with the subspace \(\hat {\mathfrak {g}} = \{\hat {x} \mid x \in \mathfrak {g}\}\) of \(V_{\mathfrak {g}}\), where \(\hat {x}\) denotes the constant path with value x. It follows that \({\Phi }(\hat {x}) = \exp x\). The composition
is a Riemannian submersion which is also called the parallel transport map.
We consider the isometric action of G on M defined by
where b ∈ G and aK ∈ M, and the isometric action of G × G on G defined by
where (b,c) ∈ G × G and a ∈ G. Then π and Φ have the following equivariant properties [30, Proposition 1.1 (i)]:
From these, we have
Let H be a closed subgroup of G; in later sections, we will suppose that it is a symmetric subgroup of G. We denote by \(\mathfrak {h}\) the Lie algebra of H and \(\mathfrak {g} = \mathfrak {h} \oplus \mathfrak {p}\) the orthogonal direct sum decomposition. Then H acts on M, the subgroup H × K acts on G and the subgroup P(G,H × K) acts on \(V_{\mathfrak {g}}\). We know the following relations for orbits [30, Proposition 1.1 (ii)]:
Thus, we have
Then, we obtain the commutative diagram
where p denotes the projection onto the first component and ψ the submersion defined by ψ(g) := (g(0),g(1)) for \(g \in \mathcal {G}\). We know that the following conditions are equivalent: the orbit H ⋅ aK is a minimal submanifold of M, the orbit (H × K) ⋅ a is a minimal submanifold of G, and the orbit P(G,H × K) ∗ u through u ∈Φ− 1(a) is a minimal PF submanifold of \(V_{\mathfrak {g}}\) [12, Theorem, 4.12], [5, Lemma 5.2].
Recall that an isometric action of a compact Lie group A on a Riemannian manifold X is called polar if there exists a closed connected submanifold Σ of X which meets every A-orbit and is orthogonal to the A-orbits at every point of intersection. Such a Σ is called a section, which is automatically totally geodesic in X. If Σ is flat in the induced metric, then the action is called hyperpolar [6]. For a proper Fredholm action on a Hilbert space, we can define it to be hyperpolar in the similar way. We know that the following conditions are equivalent [6, Proposition 2.11], [30, Theorem 1.2], [3, Lemma 4]:
-
(i)
The H-action on M is hyperpolar,
-
(ii)
The H × K-action on G is hyperpolar,
-
(iii)
The P(G,H × K)-action on \(V_{\mathfrak {g}}\) is hyperpolar.
Since we fixed a bi-invariant Riemannian metric on G induced by a negative multiple of the Killing form of \(\mathfrak {g}\), the condition (ii) is equivalent to the existence of a c-dimensional abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\) where c is the cohomogeneity of the H × K-action [6, Theorem 2.1]. Then \(\pi (\exp \mathfrak {t})\), \(\exp \mathfrak {t}\), and \(\hat {\mathfrak {t}} = \{\hat {x} \mid x \in \mathfrak {t}\}\) are sections of the H-action, the H × K-action, and the P(G,H × K)-action respectively. If the actions are hyperpolar then the following conditions are equivalent [30, Theorem 1.2]: aK ∈ M is a regular point of the H-action, a ∈ G is a regular point of the H × K-action, and u ∈Φ− 1(a) is a regular point of the P(G,H × K)-action. Here, a point is called regular if the orbit though it is principal.
3 Submanifold Geometry of Orbits of Hermann Actions
In this section, we review fundamental results on the submanifold geometry of orbits of Hermann actions. For details, see Ohno [22] (see also Goertsches-Thorbergsson [2] and Ikawa [9]). Throughout this section, M = G/K denotes a symmetric space of compact type and H a symmetric subgroup of G. We denote by σ and τ the involutions of G associated to K and H respectively and by \(\mathfrak {g} = \mathfrak {k} \oplus \mathfrak {m}\) and \(\mathfrak {g} = \mathfrak {h} \oplus \mathfrak {p}\) the canonical decompositions. We choose and fix a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\) so that \({\Sigma } := \pi (\exp \mathfrak {t})\) is a section of the Hermann action.
Consider the root space decomposition of \(\mathfrak {g}^{\mathbb {C}}\) with respect to \(\mathfrak {t}\):
where
Here \({\Delta } = \{\alpha \in \mathfrak {t} \backslash \{0\} \mid \mathfrak {g}(\alpha ) \neq \{0\}\}\) is a root system of \(\mathfrak {t}\) [9, Lemma 4.12]. Since \(\overline {\mathfrak {g}(\alpha )} = \mathfrak {g}(- \alpha )\), where the bar denotes the complex conjugation, the real form is
where
Note that
Since σ commutes with ad(η)2 for all \(\eta \in \mathfrak {t}\), we have
where
We define a linear orthogonal transformation ψα of \(\mathfrak {g}_{\alpha }\) by
An equivalent definition is that
Since σ ∘ ψα = −(ψα ∘ σ), we have a linear isometry \(\psi _{\alpha }: \mathfrak {m}_{\alpha } \rightarrow \mathfrak {k}_{\alpha }\). Set
By setting \(x^{\alpha }_{i} := \psi _{\alpha }\left (y^{\alpha }_{i}\right )\), we can take bases \(\left \{x^{\alpha }_{i}\right \}_{i = 1}^{m(\alpha )}\) of \(\mathfrak {k}_{\alpha }\) and \(\left \{y^{\alpha }_{i}\right \}_{i = 1}^{m(\alpha )}\) of \(\mathfrak {m}_{\alpha }\) satisfying
for any \(\eta \in \mathfrak {t}\).
The root space decompositions above are refined by combining a decomposition derived from the involutions σ and τ. More precisely, we consider the composition
and the eigenspace decomposition
where
Here, the eigenvalues belong to \(U(1) = \{\epsilon \in \mathbb {C} \mid \vert \epsilon \vert = 1\}\). For each 𝜖 ∈ U(1), we denote by \(\arg \epsilon \) its argument satisfying \(- \pi < \arg \epsilon \leq \pi \). Since σ ∘ τ commutes with ad(η) for all \(\eta \in \mathfrak {t}\), we have
where
Since \(\overline {\mathfrak {g}(\alpha , \epsilon )} = \mathfrak {g}(- \alpha , \epsilon ^{-1})\) the real form is
where
Setting ρ+ = σ ∘ τ + τ ∘ σ and ρ− = σ ∘ τ − τ ∘ σ we can write
where Re(𝜖) and Im(𝜖) denote the real and imaginary parts of 𝜖 respectively. Since \(\mathfrak {g}_{0, \epsilon }\) and \(\mathfrak {g}_{\alpha , \epsilon }\) are invariant under σ, we have
where
Since \(\mathfrak {g}_{\alpha , \epsilon }\) is invariant under ψα, we have a linear isometry \(\psi _{\alpha }: \mathfrak {m}_{\alpha , \epsilon } \rightarrow \mathfrak {k}_{\alpha , \epsilon }\). Set
Then, similarly, we can take bases \(\left \{x^{\alpha , \epsilon }_{i}\right \}_{i = 1}^{m(\alpha ,\epsilon )}\) of \(\mathfrak {k}_{\alpha , \epsilon }\) and \(\left \{y^{\alpha , \epsilon }_{i}\right \}_{i = 1}^{m(\alpha , \epsilon )}\) of \(\mathfrak {m}_{\alpha , \epsilon }\) satisfying
for any \(\eta \in \mathfrak {t}\).
We now take \(w \in \mathfrak {t}\), set \(a := \exp w\) and consider the orbit N := H ⋅ aK through aK. Denote by La the isometry of M defined by La(bK) := (ab)K. Identifying TeKM with \(\mathfrak {m}\), we can describe the tangent space and the normal space of N as follows [22, p. 12]:
Moreover, the decomposition (3.4) is just the common eigenspace decomposition of the family of shape operators \(\{A^{N}_{dL_{a}(\xi )}\}_{\xi \in \mathfrak {t}}\). In fact [22, p. 17]:
for each \(\xi \in \mathfrak {t}\). If σ and τ commute then 𝜖 = ± 1 and thus we get [2, Theorem 5.3]:
where
In particular if σ = τ then we have [28, p. 122]
where
4 The Curvature-Adapted Property
In this section, we formulate the curvature-adapted property of orbits of Hermann actions.
First, we recall the concept of curvature-adapted submanifolds [1]. Let N be a submanifold of a Riemannian manifold M. For each \(v \in T^{\perp }_{p} N\) at each p ∈ N, the Jacobi operator Rv is a symmetric linear transformation of TpM defined by
where RM denotes the curvature tensor of M. Then N is called curvature-adapted if for every \(v \in T^{\perp }_{p} N\) at each p ∈ N the Jacobi operator Rv leaves TpN invariant and the restriction \(R_{v}\vert _{T_{p} N}\) commutes with the shape operator \({A_{v}^{N}}\) of N.
We now make the following definition:
Definition 4.1.
Let M = G/K be a symmetric space of compact type and N a submanifold of M. For an integer c satisfying 1 ≤ c ≤ codimN, we say that N is c-curvature-adapted if for each aK ∈ N the following two conditions are satisfied:
-
(i)
For every \(v \in T^{\perp }_{aK} N\), the Jacobi operator Rv leaves TaKN invariant,
-
(ii)
For each \(v \in T^{\perp }_{aK} N\), there exists a c-dimensional abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m}\) satisfying \(v \in dL_{a}(\mathfrak {t}) \subset T^{\perp }_{aK} N \) such that the union
$$ \{R_{dL_{a}(\xi)}\vert_{T_{aK} N}\}_{\xi \in \mathfrak{t}} \cup \left\{A^{N}_{dL_{a}(\xi)}\right\}_{\xi \in \mathfrak{t}} $$is a commuting family of endomorphisms of TaKN.
Note that 1-curvature-adapted submanifolds are just curvature-adapted submanifolds in the original sense. Note also that if aK = eK then Rv is identified with − ad(v)2 since M is a symmetric space. Typical examples of c-curvature-adapted submanifolds are given by the following proposition, which was essentially shown by Goertsches and Thorbergsson [2, Corollaries 3.3 and 3.4]:
Proposition 4.2 (Goertsches-Thorbergsson 2)
All orbits of Hermann actions of cohomogeneity c are c-curvature-adapted submanifolds.
Proof
Let N be an orbit of a Hermann action \(H \curvearrowright M\) of cohomogeneity c. Take aK ∈ N. Since \(L_{a}^{-1} N = (a^{-1}Ka) \cdot eK\) we can assume aK = eK without loss of generality. Take \(v \in T^{\perp }_{eK} N\). Choose a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p} = T^{\perp }_{eK} N\) containing v. Since \(\pi (\exp \mathfrak {t})\) is a section of the Hermann action, we have \(\dim \mathfrak {t} = c\). Then it follows from the decomposition (3.4) that the Jacobi operator Rv = −ad(v)2 leaves TeKN invariant and
hold for any \(v, w \in \mathfrak {t}\). Thus, N is a c-curvature-adapted submanifold of M. □
Remark 4.3
We do not know whether all orbits of hyperpolar actions of cohomogeneity c are c-curvature-adapted submanifolds or not. We know that any indecomposable hyperpolar action of cohomogeneity at least two on M is orbit equivalent to a Hermann action [16]. We also know that any cohomogeneity one action on M is automatically hyperpolar [6, Corollary 2.13]. There exist examples of cohomogeneity one actions on the standard sphere which are different from Hermann actions [6, 15]. Since the standard sphere is of constant sectional curvature, all orbits of such cohomogeneity one actions are 1-curvature-adapted submanifolds.
Let N be a c-curvature-adapted submanifold of a symmetric space M = G/K of compact type. Take aK ∈ N. Choose and fix an arbitrary c-dimensional abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m}\) satisfying the condition (ii) of Definition 4.1 for some \(v \in T^{\perp }_{aK} N\). Consider the root space decomposition
where
These are just the common eigenspace decompositions of the commuting operators \(\left \{\operatorname {ad}(\xi )^{2}\right \}_{\xi \in \mathfrak {t}}\). On the other hand, the following lemma concerns the common eigenspace decomposition of the commuting operators \(\left \{A^{N}_{dL_{a}(\xi )}\right \}_{\xi \in \mathfrak {t}}\).
Lemma 4.4
There exists a unique finite subset Λ of \(\mathfrak {t}\) such that
where Sλ is a nonzero subspace of TaKN defined by
Proof
By left translation, we can assume aK = eK without loss of generality. It is easy to see that such a subset Λ is unique. To see the existence, we take a basis \(\{\eta _{i}\}_{i = 1}^{c}\) of \(\mathfrak {t}\). We denote by {λ(ηi)1,⋯,λ(ηi)m(i)} the set of all distinct eigenvalues of the shape operator \(A^{N}_{\eta _{i}}\) and by \(W_{\lambda (\eta _{i})_{1}}, {\cdots } , W_{\lambda (\eta _{i})_{m(i)}}\) their eigenspaces. Since \(\{A^{N}_{\eta _{i}}\}_{i = 1}^{c}\) is a commuting family, we have the common eigenspace decomposition
Define a linear functional \(\lambda _{j_{1} {\cdots } j_{c}} : \mathfrak {t} \rightarrow \mathbb {R}\) by
Then for each \(\eta = a_{1} \eta _{1} + {\cdots } + a_{c} \eta _{c} \in \mathfrak {t}\) and \(x \in W_{\lambda (\eta _{1})_{j_{1}}} \cap {\cdots } \cap W_{\lambda (\eta _{c})_{j_{c}}}\) we have
Set \({\Lambda } := \{\lambda _{j_{1} {\cdots } j_{c}}\}_{1 \leq j_{1} \leq m(1) ,\ \cdots , \ 1 \leq j_{c} \leq m(c)}\) and \(S_{\lambda _{j_{1} {\cdots } j_{c}}} := W_{\lambda (\eta _{1})_{j_{1}}} \cap {\cdots } \cap W_{\lambda (\eta _{c})_{j_{c}}}\). Identifying \(\mathfrak {t}\) with the dual space \(\mathfrak {t}^{*}\) we obtain the desired subset \({\Lambda } \subset \mathfrak {t}\) and the decomposition \(T_{eK} N = \mathop {\bigoplus }_{\lambda \in {\Lambda }} S_{\lambda }\). This proves the lemma. □
The following proposition concerns the common eigenspace decomposition of the union of commuting operators \(\{R_{dL_{a}(\xi )}\}_{\xi \in \mathfrak {t}} \cup \left \{A^{N}_{dL_{a}(\xi )}\right \}_{\xi \in \mathfrak {t}}\).
Proposition 4.5
Let Λ be as in Lemma 4.4. Then the tangent space and the normal space of N are decomposed as follows:
where \({\Lambda }_{0} := \{\lambda \in {\Lambda } \mid dL_{a}(\mathfrak {m}_{0}) \cap S_{\lambda } \neq \{0\}\}\) and \({\Lambda }_{\alpha } :=\{\lambda \in {\Lambda } \mid dL_{a}(\mathfrak {m}_{\alpha }) \cap S_{\lambda } \neq \{0\}\}\).
Proof
By left translation, we can assume aK = eK without loss of generality. Since the tangent space is invariant under \(\{R_{\xi }\}_{\xi \in \mathfrak {t}}\) the normal space is also invariant under \(\{R_{\xi }\}_{\xi \in \mathfrak {t}}\) and we have the decompositions
By the curvature-adapted property, \(\mathfrak {m}_{0} \cap T_{eK} N\) and \(\mathfrak {m}_{\alpha } \cap T_{eK} N\) are invariant under \(\{A^{N}_{\xi }\}_{\xi \in \mathfrak {t}}\). Thus, by similar arguments as in the proof of Lemma 4.4, we obtain the common eigenspace decompositions
and the assertion follows. □
Example 4.6
Let \(H \curvearrowright M\) be a Hermann action of cohomogeneity c. Choose a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\). Take \(w \in \mathfrak {t}\), set \(a := \exp w\) and consider the orbit N = H ⋅ aK through aK. From the decomposition (3.4), it is clear that \(\mathfrak {t}\) is a c-dimensional abelian subspace in \(\mathfrak {m}\) satisfying the condition (ii) of Definition 4.1 for any \(v \in \{dL_{a}(\xi )\}_{\xi \in \mathfrak {t}}\). We set \(U(1)_{0} := \{\epsilon \in U(1)_{\geq 0} \mid \mathfrak {m}_{0, \epsilon } \neq \{0\}\}\) and
We also set \(U(1)_{\alpha } := \{\epsilon \in U(1) \mid \mathfrak {m}_{\alpha , \epsilon } \neq \{0\}\}\) and
Then, we can rewrite the decompositions (3.4) and (3.5) as follows:
For each α ∈Δ+ and \(\epsilon \in U(1)_{\alpha }^{\top }\), we set
Then, Λ0 and Λα in Proposition 4.5 are
Note that the correspondence \(U(1)_{\alpha }^{\top } \ni \epsilon \mapsto \lambda (\alpha , \epsilon ) \in {\Lambda }_{\alpha }\) is one-to-one because \(\cot x\) is strictly decreasing on \(\mathbb {R}/\pi \mathbb {Z}\). Thus, we have
and therefore the decompositions (4.3) and (4.4) are expressed as
5 Principal Curvatures via the Parallel Transport Map
Let M = G/K be a symmetric space of compact type and \({\Phi }_{K}: V_{\mathfrak {g}} \rightarrow M\) the parallel transport map. In [13] and [18], an explicit formula for the principal curvatures of the PF submanifold \({\Phi }_{K}^{-1}(N)\) of \(V_{\mathfrak {g}}\) was given under the assumption that N is a curvature-adapted submanifold of M. In this section, we refine that formula to the case of c-curvature-adapted submanifolds so that it can be applied to orbits of Hermann actions.
Let N be a c-curvature-adapted submanifold of M. To consider the PF submanifold \({\Phi }_{K}^{-1}(N)\) of \(V_{\mathfrak {g}}\), we can assume eK ∈ N without loss of generality due to the equivariant property (2.3) of ΦK. Choose and fix an arbitrary c-dimensional abelian subspace \(\mathfrak {t}\) in \(T^{\perp }_{eK} N\) satisfying the condition (ii) of Definition 4.1 for some \(v \in T^{\perp }_{eK} N\). Recall the decompositions given in Proposition 4.5:
Set
Take bases
Then we obtain a basis
of \(\mathfrak {m}_{0}\) and a basis
of \(\mathfrak {m}_{\alpha }\). Via an isometry \(\psi _{\alpha }: \mathfrak {m}_{\alpha } \rightarrow \mathfrak {k}_{\alpha }\) defined by (3.1) we take a basis
of \(\mathfrak {k}_{\alpha }\). Finally we choose a basis \(\{{x_{i}^{0}}\}_{i= 1}^{\dim \mathfrak {k}_{0}}\) of \(\mathfrak {k}_{0}\). Then the relations
hold for any \(\xi \in \mathfrak {t}\).
We write \(V(\mathfrak {g})\) for the Hilbert space \(V_{\mathfrak {g}} = L^{2}([0,1], \mathfrak {g})\) and decompose
We equip a suitable basis with each term above. Recall that in addition to
there are two other kinds of orthonormal bases of \(L^{2}([0,1], \mathbb {R})\), namely
We consider bases
Then all these bases form a basis of \(V(\mathfrak {g}) = V_{\mathfrak {g}} \cong T_{\hat {0}} V_{\mathfrak {g}}\). Since \({\Phi } : V_{\mathfrak {g}} \rightarrow G\) is a Riemannian submersion with the orthogonal direct sum decomposition [31, p. 686]
we have the orthogonal direct sum decomposition
where ⊥ denotes the projection from \(\mathfrak {g} = \mathfrak {k} \oplus T_{eK} N \oplus T^{\perp }_{eK} N\) onto \(T^{\perp }_{eK} N\). Thus, we obtain a basis
of \(T_{\hat {0}} {\Phi }_{K}^{-1}(N)\).
For each \(\xi \in \mathfrak {t}\), we denote by \(A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }}\) the shape operator of \({\Phi }_{K}^{-1}(N)\) in the direction of \(\hat {\xi }\). Similarly to [18, Lemma 3.1], the following lemma holds.
Lemma 5.1
-
(i)
\(A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }}\left ({x^{0}_{i}} {\sin \limits } n \pi t\right ) = 0 \), HCode \(A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }}\left (y^{0, \lambda }_{j}\right ) = \langle \lambda , \xi \rangle y^{0, \lambda }_{j}, \)
-
(ii)
\( A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }}\left (y^{0, \lambda }_{j} {\cos \limits } n \pi t\right ) = A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }}\left (y^{0, \perp }_{l} {\cos \limits } n \pi t\right ) =0, \)
-
(iii)
\( A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }} (x^{\alpha , \perp }_{r} {\sin \limits } n \pi t) = - \frac {\langle \alpha , \xi \rangle }{n \pi } y^{\alpha , \perp }_{r} {\cos \limits } n \pi t, \) \( A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }} (y^{\alpha , \perp }_{r} {\cos \limits } n \pi t) = - \frac {\langle \alpha , \xi \rangle }{n \pi } x^{\alpha , \perp }_{r} {\sin \limits } n \pi t, \)
-
(iv)
\( A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }}\left (y^{\alpha , \lambda }_{k}\right ) = \langle \lambda , \xi \rangle y^{\alpha , \lambda }_{k} + \frac {2 \langle \alpha , \xi \rangle }{\pi } \sum \limits _{n = 1}^{\infty } \frac {1}{n} \left (x^{\alpha , \lambda }_{k}{\sin \limits } n \pi t \right ) , \)
-
(v)
\( A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }}\left (x^{\alpha , \lambda }_{k} {\sin \limits } n \pi t\right ) = - \frac {\langle \alpha , \xi \rangle }{n \pi } y^{\alpha , \lambda }_{k} (-1 + {\cos \limits } n \pi t) , \)
-
(vi)
\( A^{{\Phi }_{K}^{-1}(N)}_{\hat {\xi }}\left (y^{\alpha , \lambda }_{k} {\cos \limits } n \pi t\right ) = - \frac {\langle \alpha , \xi \rangle }{n \pi } x^{\alpha , \lambda }_{k} {\sin \limits } n \pi t \).
The following theorem describes the principal curvatures of the PF submanifold \({\Phi }_{K}^{-1}(N)\) of \(V_{\mathfrak {g}}\). This theorem refines [18, Theorem 3.2] (see also [13, Theorem 3.3]). In fact, if c = 1 then it is equivalent to the original one. It can be proven by the similar arguments using Lemma 5.1.
Theorem 5.2
Let M = G/K be a symmetric space of compact type, \({\Phi }_{K}: V_{\mathfrak {g}} \rightarrow M\) the parallel transport map, N a c-curvature-adapted submanifold of M through eK, and \(\mathfrak {t}\) an arbitrary c-dimensional abelian subspace in \(\mathfrak {m}\) satisfying the condition (ii) of Definition 4.1. Then for each \(\xi \in \mathfrak {t}\) the principal curvatures of the PF submanifold \({\Phi }^{-1}_{K}(N)\) in the direction of \(\hat {\xi }\) are given by
where we set \({\Delta }^{+}_{\xi } := \{\upbeta \in {\Delta }^{+} \mid \langle \upbeta , \xi \rangle = 0\}\) and \(\arctan \frac {\langle \alpha , \xi \rangle }{\langle \lambda , \xi \rangle } := \frac {\pi }{2}\) if 〈λ,ξ〉 = 0. The eigenfunctions and the multiplicities are given in Table 1.
6 Principal Curvatures of P(G,H × K)-Orbits
In this section, from Theorem 5.2, we derive an explicit formula for the principal curvatures of orbits of P(G,H × K)-actions induced by Hermann actions.
Let M = G/K be a symmetric space of compact type and H a symmetric subgroup of G. Choose and fix a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\). Then \(\pi (\exp \mathfrak {t})\) is a section of the Hermann action \(H \curvearrowright M\) and \(\hat {\mathfrak {t}} = \{\hat {x} \mid x \in \mathfrak {t}\}\) is a section of the hyperpolar P(G,H × K)-action on \(V_{\mathfrak {g}}\). We take arbitrary \(w , \xi \in \mathfrak {t}\) and consider the principal curvatures of \(P(G, H \times K) * \hat {w}\) in the direction of \(\hat {\xi }\).
Recall that the tangent space and the normal space of the orbit N = H ⋅ aK where \(a := \exp w\) are decomposed as follows (cf. Section 3 and Example 4.6):
where we set \(U(1)_{\alpha } := \{\epsilon \in U(1) \mid \mathfrak {m}_{\alpha , \epsilon } \neq \{0\}\}\) and
Here, \(dL_{a} (\mathfrak {m}_{0, \epsilon })\) and \(dL_{a}(\mathfrak {m}_{\alpha , \epsilon })\) are the eigenspaces of the shape operator \(A^{N}_{dL_{a}(\xi )}\) associated with the eigenvalues 0 and \(- \langle \alpha , \xi \rangle \cot \left (\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon \right )\) respectively.
Using the above information, we can describe the principal curvatures of orbits of P(G,H × K)-actions induced by Hermann actions:
Theorem 6.1
Let M = G/K be a symmetric space of compact type and H a symmetric subgroup of G. Take a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\) and \(w \in \mathfrak {t}\). Then for each \(\xi \in \mathfrak {t}\) the principal curvatures of \(P(G,H \times K) * \hat {w}\) in the direction of \(\hat {\xi }\) are given by
Taking bases \(\left \{{x^{0}_{i}}\right \}_{i}\) of \(\mathfrak {k}_{0}\), \(\left \{x^{\alpha , \epsilon }_{k}\right \}_{k}\) of \(\mathfrak {k}_{\alpha , \epsilon }\), \(\left \{y^{0, \epsilon }_{j}\right \}_{j}\) of \(\mathfrak {m}_{0, \epsilon }\), {ηl}l of \(\mathfrak {t}\) and \(\left \{y^{\alpha , \epsilon }_{k}\right \}_{k}\) of \(\mathfrak {m}_{\alpha , \epsilon }\) with the relation (3.3) we can describe the eigenfunctions and the multiplicities as in Table 2. Here, we are identifying \(T_{\hat {w}} V_{\mathfrak {g}}\) with \(T_{\hat {0}} V_{\mathfrak {g}}\) via the gauge transformation \(g*: V_{\mathfrak {g}} \rightarrow V_{\mathfrak {g}}\) for a unique g ∈ P(G,G ×{e}) satisfying \(g * \hat {0} = \hat {w}\).
In particular, if \(w \in \mathfrak {t}\) is a regular point then the term \(\frac {\langle \alpha , \xi \rangle }{n \pi }\) vanishes.
Proof
Take a unique g ∈ P(G,G ×{e}) satisfying \(g * \hat {0} = \hat {w}\). By (2.2) we have \(g(0) = \exp w = a\). From (2.3) the diagram
commutes. Thus setting \(\bar {N} := L_{a}^{-1}(N)\) we have \(g* {\Phi }_{K}^{-1}(\bar {N}) = {\Phi }_{K}^{-1}(N) = P(G, H \times K) * \hat {w}\) by (2.4). Moreover since \(w \in \mathfrak {t}\) it follows from \(g * \hat {0} = \hat {w}\) that \(g(t) \in \exp \mathfrak {t}\) for all t ∈ [0,1]. Thus we have \(d(g * ) \hat {\xi } = g \hat {\xi } g^{-1} = \hat {\xi }\). Hence it suffices to compute the principal curvatures of \({\Phi }_{K}^{-1}(\bar {N})\) in the direction of \(\hat {\xi }\). Since \(\mathfrak {t}\) is a c-dimensional abelian subspace in \(\mathfrak {m}\) satisfying the condition (ii) of Definition 4.1 we can apply Theorem 5.2 to \(\bar {N}\). From (4.3) and (4.4), the tangent space and the normal space of \(\bar {N}\) are
From (4.5) and (4.6), the above decompositions are rewritten as
where \(\bar {S}_{0} := dL_{a}^{-1}(S_{0})\) and \(\bar {S}_{\lambda } := dL_{a}^{-1}(S_{\lambda })\). Since 〈β,ξ〉 = 0 implies 〈λ(β,𝜖),ξ〉 = 0 the eigenvalue 〈λ,ξ〉 in the theorem is equal to 0. Moreover taking a unique \(m^{\prime } \in \mathbb {Z}\) satisfying \(- \pi /2 < \langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon + m^{\prime } \pi \leq \pi /2\) we have
Since \(m \in \mathbb {Z}\) in the theorem is arbitrary, the assertion follows. □
Applying Theorem 6.1 to (3.6) and (3.7), we obtain the following corollary.
Corollary 6.2
Suppose that σ ∘ τ = τ ∘ σ. Then the principal curvatures of the orbit \(P(G, H \times K) * \hat {w}\) in the direction of \(\hat {\xi }\) are given by
The multiplicities are respectively given by
In particular, if \(w \in \mathfrak {t}\) is a regular point then the term \(\frac {\langle \alpha , \xi \rangle }{n \pi }\) vanishes.
Applying Theorem 6.1 to (4.5) and (4.6), we obtain the following corollary.
Corollary 6.3
Suppose that σ = τ. Then the principal curvatures of the orbit \(P(G, H \times K) * \hat {w}\) in the direction of \(\hat {\xi }\) are given by
The multiplicities are respectively given by
In particular, if \(w \in \mathfrak {t}\) is a regular point then the term \(\frac {\langle \alpha , \xi \rangle }{n \pi }\) vanishes.
Remark 6.4
Terng [29] showed that any principal orbit of the P(G,ΔG)-action, where ΔG is the diagonal of G × G, is an isoparametric PF submanifold of \(V_{\mathfrak {g}}\) and computed its principal curvatures. This result was extended by Pinkall and Thorbergsson [25] to the case of P(G,K × K)-action, where K is a symmetric subgroup of G. (Note that in the equation (28) of [25] the term α(Y ) should be − α(Y ).) More generally, Koike [14] computed the principal curvatures of principal orbits of the P(G,H × K)-action induced by a Hermann action with the assumption that the involutions σ and τ commute [14, p. 114]. Theorem 6.1 above does not require such assumptions at all.
Remark 6.5
For each α ∈Δ+ and \(\epsilon \in U(1)_{\alpha }^{\top }\) it is clear that
We will alternatively use the latter expression to describe the principal curvatures.
7 The Austere Property: Reduced Case
In this section, we study the relation between the austere properties of H- and P(G,H × K)-orbits under the assumption that the root system Δ is reduced; the non-reduced case will be dealt with in the next section. Notice that this assumption is independent of the choice of a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\). The main result of this section is the following theorem (Theorem I in Introduction):
Theorem 7.1
Let M = G/K be a symmetric space of compact type and H a symmetric subgroup of G. Suppose that the root system Δ of a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\) is reduced. Then for \(w \in \mathfrak {g}\) the following conditions are equivalent:
-
(i)
The orbit \(H \cdot (\exp w) K\) through \((\exp w) K\) is an austere submanifold of M,
-
(ii)
The orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
To prove this theorem, we need the following lemma. The statement (i) was essentially shown by Ohno [22, Proposition 13]. Note that this lemma is still valid in the non-reduced case.
Lemma 7.2
Let \(\mathfrak {t}\) be a maximal abelian subspace in \(\mathfrak {m} \cap \mathfrak {p}\) and \(w \in \mathfrak {t}\). Set
which is a subset of \(U(1)_{\alpha }^{\top }\). Then
-
(i)
(Ohno [22]) The orbit \(H \cdot (\exp w)K\) through \((\exp w)K\) is an austere submanifold of M if and only if the set
$$ \left\{ \left. \cot \left( \langle \alpha, w \rangle + \frac{1}{2} \arg \epsilon \right) \alpha \ \right\vert\ \alpha \in {\Delta}^{+}, \ \epsilon \in U(1)_{\alpha}^{*} \right\} $$with multiplicities is invariant under the multiplication by (− 1), where the multiplicity of \(\cot (\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon ) \alpha \) is defined to be m(α,𝜖),
-
(ii)
The orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\) if and only if the set
$$ \left\{ \left. \frac{1}{\langle \alpha, w \rangle + \frac{1}{2} \arg \epsilon + m \pi} \alpha \ \right\vert\ \alpha \in {\Delta}^{+}, \ \epsilon \in U(1)_{\alpha}^{*} , \ m \in \mathbb{Z} \right\} $$with multiplicities is invariant under the multiplication by (− 1), where the multiplicity of \(\frac {1}{\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon + m \pi } \alpha \) is defined to be m(α,𝜖).
In connection with the proof of (ii), we reprove (i) here.
Proof
-
(i)
Set \(a := \exp w\) and N := H ⋅ aK. From the straightforward computations [22, pp. 15–16], the normal space of N is expressed as
$$ T^{\perp}_{aK} N = dL_{a}(\mathfrak{m} \cap \operatorname{Ad}(a)^{-1} \mathfrak{p}) = dL_{a}(\ \ \bigcup_{b \in K \cap a^{-1} Ha} \operatorname{Ad}(b)\mathfrak{t}\ \ ). $$(7.1)Thus for each \(v \in T^{\perp }_{aK} N\) there exist \(\xi \in \mathfrak {t}\) and b ∈ K ∩ a− 1Ha such that v = dLa(Ad(b)ξ). Since b belongs to a− 1Ha the isometry Lb leaves the submanifold \(\bar {N} := L_{a}^{-1}N\) invariant. Moreover since b belongs to K the differential dLb of the isometry Lb at eK is identified with Ad(b). Thus, the shape operators satisfy \(A^{\bar {N}}_{\operatorname {Ad}(b)\xi }= dL_{b} \circ A^{\bar {N}}_{\xi } \circ dL_{b}^{-1}\). From this, we obtain
$$ A^{N}_{dL_{a}(\operatorname{Ad}(b)\xi)} = dL_{a} \circ dL_{b} \circ dL_{a}^{-1} \circ A^{N}_{dL_{a} (\xi)} \circ dL_{a} \circ dL_{b}^{-1} \circ dL_{a}^{-1}. $$This shows that the eigenvalues with multiplicities of the shape operators \({A^{N}_{v}}\) and \(A^{N}_{dL_{a}(\xi )}\) coincide. Thus, to consider the austere property, it suffices to consider normal vectors \(\{dL_{a}(\xi )\}_{\xi \in \mathfrak {t}}\) of N. Thus, it follows from the common eigenspace decomposition (3.4) that the orbit \(H \cdot (\exp w) K\) is an austere submanifold of M if and only if the set
$$ \left\{ \left. \langle \alpha, \xi \rangle \cot \left( \langle \alpha, w \rangle + \frac{1}{2} \arg \epsilon \right) \ \right\vert\ \alpha \in {\Delta}^{+}, \ \epsilon \in U(1)_{\alpha}^{\top} \right\} $$with multiplicities is invariant under the multiplication by (− 1) for each \(\xi \in \mathfrak {t}\). Notice that this condition is equivalent to the condition that the set
$$ \left\{ \left. \cot \left( \langle \alpha, w \rangle + \frac{1}{2} \arg \epsilon \right) \alpha \ \right\vert\ \alpha \in {\Delta}^{+}, \ \epsilon \in U(1)_{\alpha}^{\top} \right\} $$with multiplicities is invariant under the multiplication by (− 1) cf. [11, p. 459]. Hence, the assertion follows from the fact \(\cot \left (\frac {\pi }{2} + \pi \mathbb {Z}\right ) = \{0\}\).
-
(ii)
Choose a unique g ∈ P(G,G ×{e}) satisfying \(g * \hat {0} = \hat {w}\). Then we have a = g(0) and the commutative diagram
(7.2)
Since ΦK is a Riemannian submersion it follows from (7.1) and (7.2) that each normal vector of \({\Phi }_{K}^{-1}(N)\) is expressed as \((dg* )\operatorname {Ad}(b) \hat {\xi }\) for some \(\xi \in \mathfrak {t}\) and b ∈ K ∩ a− 1Ha. Denote by \(\hat {b} \in \mathcal {G}\) the constant path with value b. Then by (2.3) we have the commutative diagram
where \(\hat {b} *\) is identified with Ad(b) acting on \(V_{\mathfrak {g}}\) by pointwise operation. Since Lb leaves \(\bar {N}\) invariant, it follows that \(\hat {b}*\) leaves \({\Phi }_{K}^{-1}(\bar {N})\) invariant. Thus, we have
This together with the equality \(g * {\Phi }_{K}^{-1}(\bar {N}) = {\Phi }_{K}^{-1}(N)\) implies
Thus, similarly, it suffices to consider normal vectors \(\{d(g*) \hat {\xi }\}_{\xi \in \mathfrak {t}}\) of \({\Phi }_{K}^{-1}(N)\). Note that \(g * \hat {0} = \hat {w}\) implies \(d(g*) \hat {\xi } = \hat {\xi }\) as mentioned in the proof of Theorem 6.1. Thus, from Theorem 6.1 and Remark 6.5, it follows that the orbit \(P(G, H \times K) * \hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\) if and only if the set
with multiplicities is invariant under the multiplication by (− 1). Hence, the assertion follows from the fact that the set \(\left \{\frac {1}{\pi /2 + m \pi } \alpha \right \}_{m \in \mathbb {Z}}\) with multiplicities is invariant under the multiplication by (− 1) due to the equality \(\frac {1}{ \pi /2 + m\pi } \alpha = (-1) \times \frac {1}{ \pi /2 +(-m-1)\pi }\alpha \). □
We are now in position to prove Theorem 7.1.
Proof Proof of Theorem 7.1
Take a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\). Since \(\pi (\exp \mathfrak {t})\) is a section of the H-action, we can assume \(w \in \mathfrak {t}\) without loss of generality.
“(i) ⇒ (ii)” : Let α ∈Δ+ and \(\epsilon \in U(1)_{\alpha }^{*}\). Since the orbit \(H \cdot (\exp w) K\) is austere, it follows from Lemma 7.2 (i) that there exist \(\alpha ^{\prime } \in {\Delta }^{+}\) and \(\epsilon ^{\prime } \in U(1)_{\alpha ^{\prime }}^{*}\) such that
Since \(\cot \left (\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon \right ) \neq 0\) and \(\cot \left (\langle \alpha ^{\prime }, w \rangle + \frac {1}{2} \arg \epsilon ^{\prime } \right ) \neq 0\), it follows from the reduced property of Δ that \(\alpha ^{\prime } = \alpha \). Moreover since the map \(\epsilon \mapsto \cot (\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon )\) is injective we have \(m(\alpha , \epsilon ) = m(\alpha , \epsilon ^{\prime })\). Then we have
Since \(\cot x\) is strictly decreasing on \(\mathbb {R}/ \pi \mathbb {Z}\), there exists a unique \(n \in \mathbb {Z}\) such that
For each \(m \in \mathbb {Z}\) we set \(m^{\prime } : = - n - m\). Then we obtain
Thus, by Lemma 7.2 (ii) the orbit \(P(G, H \times K) * \hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
“(ii) ⇒ (i)”: Since the orbit \(P(G, H \times K) * \hat {w}\) is austere, it follows from Lemma 7.2 (ii) that for each α ∈Δ+, \(\epsilon \in U(1)_{\alpha }^{*}\) and \(m \in \mathbb {Z}\) there exist \(\alpha ^{\prime } \in {\Delta }^{+}\), \(\epsilon ^{\prime } \in U(1)_{\alpha ^{\prime }}^{*}\) and \(m^{\prime } \in \mathbb {Z}\) such that
Since Δ is reduced, we have \(\alpha = \alpha ^{\prime }\). Moreover since the map \((\epsilon , m) \mapsto \frac {1}{\langle \alpha , w \rangle + (\arg \epsilon )/2 + m \pi }\) is injective we have \(m(\alpha , \epsilon ) = m(\alpha , \epsilon ^{\prime })\). Then we have
that is,
Hence, we have
Thus by Lemma 7.2 (i) the orbit \(H \cdot (\exp w) K\) is an austere submanifold of M. □
Remark 7.3
In the above proof, we essentially showed that the following conditions are equivalent when Δ is reduced:
-
(i)
The orbit \(H \cdot (\exp w)K\) through \((\exp w) K\) is an austere submanifold of M,
-
(ii)
For each α ∈Δ+ the set
$$ \left\{ \left. \cot \left( \langle \alpha, w \rangle + \frac{1}{2} \arg \epsilon \right) \alpha \ \right\vert\ \epsilon \in U(1)_{\alpha}^{*} \right\} $$with multiplicities is invariant under the multiplication by (− 1),
-
(iii)
For each α ∈Δ+ the set
$$ \left\{ \left. \frac{1}{\langle \alpha, w \rangle + \frac{1}{2} \arg \epsilon + m \pi} \alpha \ \right\vert\ \epsilon \in U(1)_{\alpha}^{*} , \ m \in \mathbb{Z} \right\} $$with multiplicities is invariant under the multiplication by (− 1).
-
(iv)
The orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\),
8 The Austere Property: General Case
In this section, without supposing that the root system Δ is reduced, we study the relation between the austere properties of H- and P(G,H × K)-orbits.
As a preliminary, we prove the following lemma, which generalizes Lemma 4.32 in [9]. In fact, if σ = τ then it is just the original one.
Lemma 8.1
Let M = G/K be a symmetric space of compact type, H a symmetric subgroup of G and \(\mathfrak {t}\) a maximal abelian subspace in \(\mathfrak {m} \cap \mathfrak {p}\). Suppose that there exists α ∈Δ satisfying 2α ∈Δ. Then the multiplicities satisfy m(α) > m(2α).
Proof
We extend the inner product of \(\mathfrak {g}\) to the complex symmetric bi-linear form on \(\mathfrak {g}^{\mathbb {C}}\) which is still denoted by 〈⋅,⋅〉. Choose 𝜖 ∈ U(1) satisfying \(\mathfrak {g}(\alpha , \epsilon ) \neq \{0\}\). Since \(\overline {\mathfrak {g}(\alpha , \epsilon )} = \sigma (\mathfrak {g}(\alpha , \epsilon )) = \mathfrak {g}(- \alpha , \epsilon ^{-1})\) the involution \(z \mapsto \sigma (\bar {z})\) leaves \(\mathfrak {g}(\alpha , \epsilon )\) invariant. Thus we have the (± 1)-eigenspace decomposition
Take \(z_{0} \in \mathfrak {g}(\alpha , \epsilon )^{+} \backslash \{0\}\). Then by definition we have
Since \([\mathfrak {g}(\alpha ), \mathfrak {g}(\alpha )] \subset \mathfrak {g}(2\alpha )\) we have the linear map \(\operatorname {ad}(z_{0}): \mathfrak {g}(\alpha ) \rightarrow \mathfrak {g}(2 \alpha )\). We restrict this map to the subspace
It suffices to show that the restriction \(\operatorname {ad}(z_{0}): (\mathbb {C}z_{0})^{\perp } \rightarrow \mathfrak {g}(2 \alpha )\) is surjective. Take arbitrary \(y \in \mathfrak {g}(2 \alpha )\). We define \(x \in (\mathbb {C} z_{0})^{\perp }\) by
Then, by the Jacobi identity, we have
where the last equality follows from \([y, z_{0}] \in [\mathfrak {g}(2 \alpha ) , \mathfrak {g}(\alpha )] \subset \mathfrak {g}(3 \alpha ) = \{0\}\). Notice that \([z_{0}, \bar {z}_{0}] \in [\mathfrak {g}(\alpha ), \mathfrak {g}(-\alpha )] \subset \mathfrak {g}(0)\). Moreover, from (8.1), we have \([z_{0}, \bar {z}_{0}] \in \mathfrak {m}^{\mathbb {C}} \cap \mathfrak {p}^{\mathbb {C}}\). Hence, we have \([z_{0}, \bar {z}_{0}] \in \mathfrak {t}^{\mathbb {C}}\) by maximality. Since
for all \(\eta \in \mathfrak {t}\) we get \([z_{0}, \bar {z}_{0}] = \sqrt {-1} \|z_{0}\|^{2} \alpha \). Applying (8.2) to this, we obtain
Therefore, we have ad(z0)(x) = y. This proves the lemma. □
Using this lemma, we study the relation between the austere properties of H- and P(G,H × K)-orbits in the rest of this section. First we consider the case σ = τ (Theorem II (i) in “Introduction”):
Theorem 8.2
Let M = G/K be a symmetric space of compact type and H a symmetric subgroup of G. Suppose that σ = τ. Then for \(w \in \mathfrak {g}\) the following conditions are equivalent:
-
(i)
The orbit \(H \cdot (\exp w)K\) through \((\exp w)K\) is an austere submanifold of M,
-
(ii)
The orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
Remark 8.3
The above conditions (i) and (ii) are also equivalent to the following conditions (see [9, Proposition 4.27, Theorem 4.31] and [19, Theorem 8]. See also [20, Theorem 1] for the irreducible case):
-
(iii)
The orbit \(H \cdot (\exp w)K\) through \((\exp w)K\) is a totally geodesic submanifold of M,
-
(iv)
The orbit \(H \cdot (\exp w)K\) through \((\exp w)K\) is a reflective submanifold of M,
-
(v)
The orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is a weakly reflective PF submanifold of \(V_{\mathfrak {g}}\).
Note that the orbit \(P(G, H \times K) * \hat {w}\) is not totally geodesic [19, Corollary 2] and thus not reflective.
Proof Proof of Theorem 8.2
Take a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} = \mathfrak {p}\). Since \(\pi (\exp \mathfrak {t})\) is a section of the H-action we can assume \(w \in \mathfrak {t}\) without loss of generality.
“(i) ⇒ (ii)”: Let α ∈Δ+ satisfy \(\langle \alpha , w \rangle \notin \pi \mathbb {Z}\). Suppose that \(\langle \alpha , w \rangle \notin \frac {\pi }{2} \mathbb {Z}\). Then from Lemma 7.2 (i) there exists \(\alpha ^{\prime } \in {\Delta }^{+}\) satisfying \(\langle \alpha ^{\prime }, w \rangle \notin \frac {\pi }{2} \mathbb {Z}\) such that
Since \(\langle \alpha , w \rangle \notin \frac {\pi }{2}\mathbb {Z}\) we have \(\alpha ^{\prime } \neq \alpha \). Then by the property of root systems \(\alpha ^{\prime }\) is either 2α or \(\frac {1}{2} \alpha \). Suppose that \(\alpha ^{\prime } = 2 \alpha \). Then the multiplicities of left and right terms of (8.3) are m(α) and m(2α) respectively. However m(α) > m(2α) holds by Lemma 8.1. Thus \(\alpha ^{\prime } \neq 2 \alpha \). Similarly \(\alpha ^{\prime } \neq \frac {1}{2}\alpha \). This is a contradiction. Thus, \(\langle \alpha , w \rangle \in \frac {\pi }{2} \mathbb {Z}\) holds for all α ∈Δ+ satisfying \(\langle \alpha , w \rangle \notin \pi \mathbb {Z}\). (Thus, N is totally geodesic.) Hence from Lemma 7.2 (ii) the orbit \(P(G, H \times K) * \hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
“(ii) ⇒ (i)”: Let α ∈Δ+ satisfy \(\langle \alpha , w \rangle \notin \pi \mathbb {Z}\). Suppose that \(\langle \alpha , w \rangle \notin \frac {\pi }{2} \mathbb {Z}\). Take \(m \in \mathbb {Z}\). Then it follows from Lemma 7.2 (ii) that there exist \(\alpha ^{\prime } \in {\Delta }^{+}\) satisfying \(\langle \alpha ^{\prime }, w \rangle \notin \frac {\pi }{2} \mathbb {Z}\) and \(m^{\prime } \in \mathbb {Z}\) such that
Since \(\langle \alpha , w \rangle \notin \frac {\pi }{2}\mathbb {Z}\) we have \(\alpha ^{\prime } \neq \alpha \). Then \(\alpha ^{\prime }\) is either 2α or \(\frac {1}{2} \alpha \). Suppose that \(\alpha ^{\prime } = 2 \alpha \). Then the multiplicity of the left term is m(α) + m(2α) due to the equality \(\frac {1}{\langle \alpha , w \rangle + m \pi }\alpha = \frac {1}{\langle 2 \alpha , w \rangle + 2 m \pi }2 \alpha \). However, that of the right term is m(2α) since \(\alpha ^{\prime } \neq \alpha \). Thus, we have \(\alpha ^{\prime } \neq 2 \alpha \). Similarly, we have \(\alpha ^{\prime } \neq \frac {1}{2}\alpha \). This is a contradiction. Thus \(\langle \alpha , w \rangle \in \frac {\pi }{2} \mathbb {Z}\) holds for all α ∈Δ+ satisfying \(\langle \alpha , w \rangle \notin \pi \mathbb {Z}\). This shows that the orbit \(H \cdot (\exp w) K\) is totally geodesic and therefore austere. □
To generalize Theorem 8.2, we recall an equivalence relation for involutions: For two involutive automorphisms τ and \(\tau ^{\prime }\) of G, we write \(\tau \sim \tau ^{\prime }\) if there exists c ∈ G such that \(\tau ^{\prime } = \operatorname {Ad}(c) \circ \tau \circ \operatorname {Ad}(c)^{-1}\). If \(\tau \sim \tau ^{\prime }\) and H a symmetric subgroup of G with respect to τ then \(H^{\prime } := \operatorname {Ad}(c)H\) is a symmetric subgroup of G with respect to \(\tau ^{\prime }\). Moreover, the actions of H and \(H^{\prime }\) on M are conjugate, that is, there exists an isomorphism \(\phi : H \rightarrow H^{\prime }\) and an isometry \(\psi : M \rightarrow M\) such that ψ(b ⋅ p) = ϕ(b) ⋅ ψ(p) for b ∈ H and p ∈ M. In fact ϕ := Ad(c) and ψ := Lc satisfy the property. Thus, we can identify \(H^{\prime }\)-orbits with H-orbits via ψ and the theorem is generalized as follows:
Corollary 8.4
Let M, H be as in Theorem 8.2. Suppose that \(\sigma \sim \tau \). Then for \(w \in \mathfrak {g}\) the following conditions are equivalent:
-
(i)
The orbit \(H \cdot (\exp w )K\) through \((\exp w)K \) is an austere submanifold of M,
-
(ii)
The orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
Proof
Let c ∈ G satisfy σ = Ad(c) ∘ τ ∘ Ad(c)− 1. Take g ∈ P(G,G ×{e}) satisfying g(0) = c. Then from (2.3) the diagram
commutes. Since each P(G,H × K)-orbit is the inverse image of an H-orbit under ΦK the assertion follows from Theorem 8.2. □
Next we consider the case σ ∘ τ = τ ∘ σ (Theorem II (ii) in “Introduction”):
Theorem 8.5
Let M = G/K be a symmetric space of compact type and H a symmetric subgroup of G. Suppose that the involutions σ and τ commute. Then if the orbit \(H \cdot (\exp w)K\) through \((\exp w) K\) where \(w \in \mathfrak {g}\) is an austere submanifold of M, the orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
Remark 8.6
The converse of Theorem 8.5 does not hold in general. In the next section, we will show a counterexample of a minimal H-orbit which is not austere but the corresponding minimal P(G,H × K)-orbit is austere.
Proof Proof of Theorem 8.5
Take a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\). We can assume \(w \in \mathfrak {t}\) without loss of generality. Let α ∈Δ+. If the set \(\mathbb {R} \alpha \cap {\Delta }^{+}\) consists of only α then it follows by the same argument as in the proof of Theorem 7.1 that the set
with multiplicities is invariant under the multiplication by (− 1). Let us consider the other cases \(\mathbb {R} \alpha \cap {\Delta }^{+} = \{\alpha , 2 \alpha \}\) or \(\{\alpha , \frac {1}{2} \alpha \}\). It suffices to consider the former case. By Lemma 7.2 the union X ∪ Y of two sets
with multiplicities is invariant under the multiplication by (− 1), and it suffices to show that the union Z ∪ W of two sets
with multiplicities is invariant under the multiplication by (− 1).
Since σ and τ commute, we have 𝜖,δ ∈{± 1}. Thus if \(\langle \alpha , w \rangle \in \frac {\pi }{2} \mathbb {Z}\) then the sets X, Y, Z, and W are empty. Suppose that \(\langle \alpha , w \rangle \notin \frac {\pi }{2} \mathbb {Z}\). Then \(\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon \notin \frac {\pi }{2} \mathbb {Z}\) for all 𝜖 ∈ U(1)α. Thus, \(U(1)_{\alpha }^{*} = U(1)_{\alpha }\). Hence \(m(\alpha ) = {\sum }_{\epsilon \in U(1)_{\alpha }^{*}} m(\alpha , \epsilon )\). This implies that there exists \(\epsilon , \epsilon ^{\prime } \in U(1)_{\alpha }^{*}\) such that
In fact, if this does not hold then for each \(\epsilon \in U(1)_{\alpha }^{*}\) there exists a unique \(\delta (\epsilon ) \in U(1)_{2 \alpha }^{*}\) satisfying
The multiplicity of the left term is m(α,𝜖), or \(m(\alpha , \epsilon ) + m(2 \alpha , \delta ^{\prime })\) if there exists \(\delta ^{\prime } \in U(1)_{2 \alpha }^{*}\) satisfying \(\cot \left (\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon \right ) \alpha = \cot \left (\langle 2\alpha , w \rangle + \frac {1}{2} \arg \delta ^{\prime }\right ) 2 \alpha \). That of the right term is m(2α,δ(𝜖)); due to negation of (8.5), we have \(\cot \left (\langle 2 \alpha , w \rangle + \frac {1}{2} \arg \delta (\epsilon )\right ) \neq \cot \left (\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon ^{\prime }\right )\) for any \(\epsilon ^{\prime } \in U(1)_{\alpha }^{*}\) and thus it is not \(m(2 \alpha , \delta (\epsilon )) + m(\alpha , \epsilon ^{\prime })\) but m(2α,δ(𝜖)). Thus, we get m(α,𝜖) ≤ m(2α,δ(𝜖)). Hence, we obtain
where the last inequality is due to the injective property of the map 𝜖↦δ(𝜖). This contradicts the fact m(α) > m(2α) of Lemma 8.1. Thus, from (8.5), we have
Thus, \(\langle \alpha , w \rangle = - \frac {1}{4} \arg \epsilon - \frac {1}{4} \arg \epsilon ^{\prime }\) mod \(\pi \mathbb {Z}\). Since \(\epsilon , \epsilon ^{\prime } \in \{\pm 1\}\), we obtain \(\langle \alpha , w \rangle \in \frac {\pi }{4} \mathbb {Z}\) and \(\langle 2\alpha , w \rangle \in \frac {\pi }{2} \mathbb {Z}\). Thus
for any 𝜖 ∈ U(1)α and δ ∈ U(1)2α. Thus, the sets Y and W are empty. Hence, the set X with multiplicities is invariant under the multiplication by (− 1). Therefore, by the same argument as in the proof of Theorem 7.1, the set Z with multiplicities is invariant under the multiplication by (− 1). This proves the theorem. □
To generalize Theorem 8.5, we recall an equivalence relation for pairs of involutions introduced by Matsuki [17]. Let (σ,τ) and \((\sigma ^{\prime }, \tau ^{\prime })\) be two pairs of involutive automorphisms of G. We write \((\sigma , \tau ) \sim (\sigma ^{\prime }, \tau ^{\prime })\) if there exist an automorphism ρ of G and an element c ∈ G such that
If K and H are symmetric subgroups of G then \(K^{\prime } := \rho (K)\) and \(H^{\prime } := \operatorname {Ad}(c) \circ \rho (H)\) are symmetric subgroups of G. Moreover, the H-action on G/K and the \(H^{\prime }\)-action on \(G/K^{\prime }\) are conjugate. In fact, these actions are conjugate under the isomorphism \(\phi := \operatorname {Ad}(c) \circ \rho : H \rightarrow H^{\prime }\) and the isometry \(\psi : G/K \rightarrow G/K^{\prime }\) defined by \(\psi (aK) := c \rho (a)K^{\prime }\). Then the theorem is generalized as follows:
Corollary 8.7
Let M, H be as in Theorem 8.5. Suppose that there exists a pair of commuting involutions \((\sigma ^{\prime }, \tau ^{\prime })\) of G satisfying \((\sigma , \tau ) \sim (\sigma ^{\prime } , \tau ^{\prime })\). Then if the orbit \(H \cdot (\exp w)K\) through \((\exp w ) K\) where \(w \in \mathfrak {g}\) is an austere submanifold of M, the orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
Proof
Let ρ ∈ Aut(G) and c ∈ G be as above. Since we equipped an Aut(G)-invariant inner product with \(\mathfrak {g}\) the automorphism ρ is an isometry of G. Thus, it induces an isometry from G/K to \(G/K^{\prime }\), which is still denoted by ρ. The differential \(d \rho : \mathfrak {g} \rightarrow \mathfrak {g}\) induces a linear orthogonal transformation of \(V_{\mathfrak {g}}\) by pointwise operation, which is still denoted by dρ. Note that \(d \rho (g * \hat {0}) = (\rho \circ g) * \hat {0}\) holds for all \(g \in \mathcal {G}\). Take h ∈ P(G,G ×{e}) satisfying h(0) = c. Then from (2.3) the diagram
commutes. Since each P(G,H × K)-orbit is the inverse image of an H-orbit under ΦK the assertion follows from Theorem 8.5. □
Finally, as far as possible, we consider the general case that σ and τ do not necessarily commute. In view of Corollary 8.7, it suffices to consider non-commutative pairs of involutions which are not equivalent to commutative ones. According to the classification result [17] if G is simple then there are three kinds of such non-commutative pairs, and if G is not simple then there are many such non-commutative pairs. For a technical reason, here we focus on the case that G is simple. In this case if (σ,τ) is one of those three pairs then the order of the composition σ ∘ τ is 3 or 4 see also [22, Section 5]. We will use this fact to prove the following theorem (Theorem II (iii) in “Introduction”):
Theorem 8.8
Let M = G/K be a symmetric space of compact type and H a symmetric subgroup of G. Suppose that G is simple. Then if the orbit \(H \cdot (\exp w)K\) through \((\exp w)K\) where \(w \in \mathfrak {g}\) is an austere submanifold of M, the orbit \(P(G, H \times K) * \hat {w}\) through \(\hat {w}\) is an austere PF submanifold of \(V_{\mathfrak {g}}\).
Proof
From the above discussion, it suffices to consider a pair of involutions (σ,τ) where the order l of σ ∘ τ is 3 or 4. Take a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\). We can assume \(w \in \mathfrak {t}\) without loss of generality. Let α ∈Δ+. By the same argument as in the proof of Theorem 8.5, it suffices to consider the case \(\mathbb {R} \alpha \cap {\Delta }^{+} = \{\alpha , 2 \alpha \}\) and to show that the austere property of X ∪ Y implies that of Z ∪ W.
First we show that \(\langle \alpha , w \rangle \in \frac {\pi }{2l} \mathbb {Z}\). If \(U(1)_{\alpha }^{*} \subsetneq U(1)_{\alpha }\) then there exists 𝜖 ∈ U(1)α satisfying \(\langle \alpha , w \rangle + \frac {1}{2} \arg \epsilon \in \frac {\pi }{2}\mathbb {Z}\). This shows \(\langle \alpha , w \rangle \in \frac {\pi }{2l} \mathbb {Z}\). If \(U(1)_{\alpha }^{*} = U(1)_{\alpha }\) then \(m(\alpha ) = {\sum }_{\epsilon \in U(1)_{\alpha }^{*}} m(\alpha , \epsilon )\). Thus, by the same argument as in the proof of Theorem 8.5, there exists \(\epsilon , \epsilon ^{\prime } \in U(1)_{\alpha }^{*}\) such that
From this, we have
Hence, \(\langle \alpha , w \rangle = - \frac {1}{4} \arg \epsilon - \frac {1}{4} \arg \epsilon ^{\prime }\) mod \(\pi \mathbb {Z}\). Therefore, \(\langle \alpha , w \rangle \in \frac {\pi }{2l} \mathbb {Z}\) as claimed.
Since \(\langle \alpha , w \rangle \in \frac {\pi }{2l} \mathbb {Z}\) we have
for any 𝜖 ∈ U(1)α and δ ∈ U(1)2α. Thus if l = 3 then
and if l = 4 then
for \(\epsilon \in U(1)_{\alpha }^{*}\) and \(\delta \in U(1)_{2\alpha }^{*}\). Therefore
for any \(\epsilon \in U(1)_{\alpha }^{*}\) and \(\delta \in U(1)_{2\alpha }^{*}\). This shows that the sets X and Y with multiplicities are respectively invariant under the multiplication by (− 1). Thus, by the similar arguments as in the proof of Theorem 7.1, the sets Z and W with multiplicities are respectively invariant under the multiplication by (− 1). This proves the theorem. □
Remark 8.9
By the same arguments, we can generalize Theorem 8.8 to the case that G is not simple but the order of σ ∘ τ is 3 or 4.
Remark 8.10
In the proofs of Theorems 8.5 and 8.8, we essentially showed that the orbit \(H \cdot (\exp w)\) is an austere submanifold of M if and only if the set
with multiplicities is invariant under the multiplication by (− 1) for each α ∈Δ+.
Example 8.11
Ikawa [9] classified austere orbits of Hermann actions under the assumptions that G is simple and that σ and τ commute. This result was extended by Ohno [22] to the non-commutative case. Thus, applying Theorems 8.5 and 8.8 to those results, we can obtain many examples of homogeneous austere PF submanifolds in Hilbert spaces. Note that so obtained austere PF submanifolds are not totally geodesic due to Corollary 2 in [19].
9 A Counterexample to the Converse
In this section, we show a counterexample to the converse of Theorems 8.5 and 8.8; we show an example of a minimal H-orbit which is not austere but the corresponding minimal P(G,H × K)-orbit is austere. Note that from Theorem 7.1 the root system Δ must be non-reduced. We give such an example by the triple
We shall suppose that p > q.
The involutions σ and τ of G corresponding to K and H respectively are
where Ep denote the unit matrix of order p. Clearly σ and τ commute. The canonical decomposition of \(\mathfrak {g} = \mathfrak {su}(p + q)\) with respect to σ is given by \(\mathfrak {k} = \mathfrak {s}(\mathfrak {u}(p) + \mathfrak {u}(q))\) and
The canonical decomposition of \(\mathfrak {g}\) with respect to τ is given by \(\mathfrak {h} = \mathfrak {so}(p + q)\) and
Thus, we can write
We define a maximal abelian subspace \(\mathfrak {t}\) in \(\mathfrak {m} \cap \mathfrak {p}\) by
Note that \(\mathfrak {t}\) is maximal also in \(\mathfrak {m}\). For each i = 1,⋯,q we set
where Eij denote the square matrix of order q having 1 in the i-th row and j-th column and zeros elsewhere. We set
where
Then, we obtain the root space decomposition
By commutativity of involutions, this decomposition is refined as follows:
We now consider the orbit \(N := H \cdot (\exp w) K\), where \(w \in \mathfrak {t}\) is defined by
Set \(a := \exp w\). Then, from (3.6) and (3.7), the tangent space and the normal space of N are
and for each \(\xi \in \mathfrak {t}\) the principal curvatures of N in the direction of dLa(ξ) are expressed as the inner product of ξ with vectors
whose multiplicities are respectively
Since the set \(\{- (\sqrt {2} + 1)e_{i}, 2 e_{i}, (\sqrt {2} - 1)e_{i}\}\) can not be invariant under the multiplication by (− 1), the orbit N is not an austere submanifold of M. Note that if p − q = 1 then it is a minimal submanifold of M but still not austere.
On the other hand, from Corollary 6.2 (see also Remark 6.5), the principal curvatures of the orbit \(P(G, H \times K) * \hat {w}\) in the direction of \(\hat {\xi }\) are expressed as the inner product of ξ with vectors
whose multiplicities are respectively
Note that
Note also that the sets \(\{- \frac {1}{\pi /2+ m \pi } (e_{i} - e_{j})\}_{m \in \mathbb {Z}}\) and \(\{\frac {1}{n \pi } (e_{i} - e_{j})\}_{n \in \mathbb {Z} \backslash \{0\}}\) with multiplicities are respectively invariant under the multiplication by (− 1). Thus, from the equalities
the orbit \(P(G, H \times K)* \hat {w}\) is austere if and only if p − q = 1. Therefore, we have shown that if p − q = 1 then the orbit \(H \cdot (\exp w) K\) is not austere but the orbit \(P(G, H \times K) * \hat {w}\) is austere. This is the desired counterexample. In this case, the orbit \(H \cdot (\exp w) K\) is a minimal submanifold of M as mentioned above; and thus, the orbit \(P(G, H \times K) * \hat {w}\) is minimal PF submanifold of \(V_{\mathfrak {g}}\) [5, 12].
Finally, we mention further remarks on the converse. As we have seen above, if the root system Δ is non-reduced then there exists a counterexample to the converse of Theorems 8.5 and 8.8. However, even if Δ is non-reduced, the converse holds in some cases. In fact, Theorem 8.2 and Corollary 8.4 are valid in the non-reduced case. Moreover, consider the case σ ∘ τ = τ ∘ σ and set
Suppose that Δ is of type BC and write \({\Delta }^{+} = \{e_{i}, 2 e_{i}\}_{i} \cup \{e_{i} \pm e_{j}\}_{i<j}\). Suppose also that \(\dim \mathfrak {t} \geq 2\) and \({\Delta }_{1}^{+} \cap {\Delta }_{-1}^{+} = \{ e_{i}\}_{i}\). Then, it follows by straightforward calculations that the converse holds (cf. [21]). Note that the counterexample shown in this section satisfies \({\Delta }_{1}^{+} \cap {\Delta }_{-1}^{+} = \{e_{i}\}_{i} \cup \{e_{i} \pm e_{j}\}_{i < j}\) if \(\dim \mathfrak {t} \geq 2\), and \({\Delta }_{1}^{+} \cap {\Delta }_{-1}^{+} = \{e_{1}\}\) if \(\dim \mathfrak {t} = 1\). For the investigation of the triple (Δ,Δ1,Δ− 1) and the corresponding commutative Hermann actions, see Ikawa’s papers [9] and [10].
References
Berndt, J., Vanhecke, L.: Curvature-adapted submanifolds. Nihonkai Math. J 3(2), 177–185 (1992)
Goertsches, O., Thorbergsson, G.: On the geometry of the orbits of Hermann actions. Geom. Dedicata 129, 101–118 (2007)
Gorodski, C., horbergsson, G.: Variationally complete actions on compact symmetric spaces. J. Differential Geom. 62(1), 39–48 (2002)
Harvey, R., Lawson H.B. Jr.: Calibrated geometries. Acta Math. 148, 47–157 (1982)
Heintze, E., Liu, X., Olmos, C.: Isoparametric submanifolds and a Chevalley- type restriction theorem, Integrable systems, geometry, and topology, 151–190, AMS/IP Stud. Adv. Math., 36, Amer. Math. Soc., Providence, RI (2006)
Heintze, E., Palais, R., Terng, C.-L., Thorbergsson, G.: Hyperpolar actions on symmetric spaces, Geometry, topology, & physics, 214–245, Conf. Proc. Lecture Notes Geom. Topology, IV, Int. Press, Cambridge, MA (1995)
Hermann, R.: Variational completeness for compact symmetric spaces. Proc. Amer. Math. Soc 11, 544–546 (1960)
Hermann, R.: Totally geodesic orbits of groups of isometries. Nederl. Akad. Wetensch. Proc. Ser. A 65 = Indag. Math 24, 291–298 (1962)
Ikawa, O.: The geometry of symmetric triad and orbit spaces of Hermann actions. J. Math. Soc. Japan 63(1), 79–136 (2011)
Ikawa, O.: A note on symmetric triad and Hermann action, Differential geometry of submanifolds and its related topics, 220–229, World Sci. Publ., Hackensack, NJ (2014)
Ikawa, O., Sakai, T., Tasaki, H.: Weakly reflective submanifolds and austere submanifolds. J. Math. Soc. Japan 61(2), 437–481 (2009)
King, C., Terng, C.-L.: Minimal submanifolds in path space, Global analysis in modern mathematics, 253–281, Publish or Perish, Houston, TX (1993)
Koike, N.: On proper Fredholm submanifolds in a Hilbert space arising from submanifolds in a symmetric space. Japan J. Math. (N.S.) 28(1), 61–80 (2002)
Koike, N.: Collapse of the mean curvature flow for equifocal submanifolds. Asian J. Math 15(1), 101–127 (2011)
Kollross, A.: A classification of hyperpolar and cohomogeneity one actions. Trans. Amer. Math. Soc 354(2), 571–612 (2002)
Kollross, A.: Hyperpolar actions on reducible symmetric spaces. Transform Groups 22(1), 207–228 (2017)
Matsuki, T.: Classification of two involutions on compact semisimple Lie groups and root systems. J. Lie Theory 12(1), 41–68 (2002)
Morimoto, M.: Austere and arid properties for PF submanifolds in Hilbert spaces. Differential Geom. Appl. 69(101613), 24 (2020)
Morimoto, M.: On weakly reflective PF submanifolds in Hilbert spaces. Tokyo J. Math 44(1), 103–124 (2021)
Morimoto, M.: On weakly reflective submanifolds in compact isotropy irreducible Riemannian homogeneous spaces. Tokyo J. Math 44(2), 467–476 (2021)
Morimoto, M.: On the converse problem for austere orbits of path group actions induced by Hermann actions. In: Proceedings of The 23rd international differential geometry workshop on submanifolds in homogeneous spaces & related topics, vol. 23, pp 157–165 (2021)
Ohno, S.: Geometric properties of orbits of Hermann actions. arXiv:2101.00765
Palais, R.S.: Morse theory on Hilbert manifolds. Topology 2, 299–340 (1963)
Palais, R.S., Terng, C.-L.: Critical Point Theory and Submanifold Geometry. In: Lecture Notes in Mathematics, vol. 1353. Springer, Berlin (1988)
Pinkall, U., Thorbergsson, G.: Examples of infinite dimensional isoparametric submanifolds. Math. Z 205(2), 279–286 (1990)
Smale, S.: Morse theory and a non-linear generalization of the Dirichlet problem. Ann. Math. 80(2), 382–396 (1964)
Smale, S.: An infinite dimensional version of Sard’s theorem. Amer. J. Math 87, 861–866 (1965)
Tasaki, H.: Certain minimal or homologically volume minimizing submanifolds in compact symmetric spaces Tsukuba. J. Math 9(1), 117–131 (1985)
Terng, C.-L.: Proper Fredholm submanifolds of Hilbert space. J. Differential Geom 29(1), 9–47 (1989)
Terng, C.-L.: Polar actions on Hilbert space. J. Geom. Anal 5(1), 129–150 (1995)
Terng, C.-L., Thorbergsson, G.: Submanifold geometry in symmetric spaces. J. Differential Geom 42(3), 665–718 (1995)
Acknowledgements
The author would like to thank Professor Yoshihiro Ohnita for useful discussions and valuable suggestions. The author is also grateful to Professors Shinji Ohno, Hiroyuki Tasaki, and Hiroshi Tamaru for their interests in this work and useful comments. Thanks are also due to Professors Naoyuki Koike and Takashi Sakai for their encouragements. Finally, the author would like to thank the referees of their careful readings of this paper and valuable comments.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of Interest
The author declares no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supported by the Grant-in-Aid for Research Activity Start-up (No. 20K22309) and by Osaka City University Advanced Mathematical Institute (MEXT Joint Usage/Research Center on Mathematics and Theoretical Physics JPMXP0619217849).
Rights and permissions
About this article
Cite this article
Morimoto, M. Curvatures and Austere Property of Orbits of Path Group Actions Induced by Hermann Actions. Transformation Groups (2022). https://doi.org/10.1007/s00031-022-09732-w
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00031-022-09732-w
Keywords
- Hermann action
- Hyperpolar action
- Proper Fredholm action
- Principal curvature
- Austere submanifold
- Proper Fredholm submanifold