Symmetric Space

Bowen Liu1

1
[email protected]
 1

Contents

Part 1. Riemannian symmetric space 3

0. Geometric viewpoints 3
0.1. Basic definitions and properties 3
0.2. Transvection 5
0.3. Holonomy group 6
0.4. Symmetric space, locally symmetric space and homogeneous
space 6
1. Algebraic viewpoints 9
1.1. Riemannian symmetric space as a Lie group quotient 9
1.2. Riemannian symmetric pair 11
1.3. Examples of Riemannian symmetric pair 12
2. Curvature of Riemannian symmetric space 14
2.1. Formulas 14
2.2. Computations 17

Part 2. Classifications of Riemannian symmetric space 20

3. Decompositions 20
3.1. Orthogonal symmetric Lie algebra 20
3.2. Decomposition into pieces of different types 21
4. Irreducibility 24
4.1. Irreducible orthogonal symmetric Lie algebra 24
4.2. Decomposition into irreducible pieces 25
5. Duality 26
6. Classifications of Riemannian symmetric space 28
6.1. Classifications of irreducible orthogonal symmetric Lie algebra 28
6.2. Relations between different viewpoints 28
6.3. Summary 29
7. More properties of Non-compact type 30

Part 3. Hermitian symmetric space 31

8. Hermitian symmetric space 31
9. Bounded symmetric domains 32
9.1. The Bergman metrics 32
9.2. Classical bounded symmetric domains 32
9.3. Curvatures of classical bounded symmetric domains 32

Part 4. Appendix 33
Appendix A. Remarks 33
A.1. Effectivity 33
Appendix B. Lie group and Lie algebra 34
B.1. Fundamental theorems 34
B.2. Adjoint action 34
B.3. Semisimple Lie algebras 34
2

Appendix C. Basic facts in Riemannian geometry 35

C.1. Killing fields 35
C.2. Hopf theorem 37
C.3. Other basic facts 38
References 39
 3

Part 1. Riemannian symmetric space

0. Geometric viewpoints
0.1. Basic definitions and properties.

0.1.1. Riemannian symmetric space.

Definition 0.1.1 (Riemannian symmetric space). A Riemannian manifold

(M, g) is called a Riemannian symmetric space if for each p ∈ M there
exists an isometry φ : M → M , which is called a symmetry at p, such that
φ(p) = p and (dφ)p = − id.

Remark 0.1.1. Theorem C.3.1 implies if symmetry at point p exists, then

it’s unique.

Proposition 0.1.1. The following statements are equivalent:

(1) (M, g) is a Riemannian symmetric space.
(2) For each p ∈ M , there exists an isometry φ : M → M such that φ2 = id
and p is an isolated fixed point of φ.

Proof. From (1) to (2). Let φ be a symmetry at p ∈ M . Since (dφ2 )p =

(dφ)p ◦ (dφ)p = id and φ2 (p) = p, one has φ2 = id by Theorem C.3.1. If p is
not an isolated fixed point, then there exists a sequence {pi }∞i=1 converging
to p such that φ(pi ) = pi . For 0 < δ < inj(p), there exists sufficiently large
k such that pk ∈ B(p, δ), and we denote v = exp−1 p (pk ). Since φ is an
isometry, one has φ(expp (tv)) and expp (tv) are two geodesics connecting p
and pk , and thus
φ(expp (tv)) = expp (tv)
by uniqueness. In particular, one has v = (dφ)p v, which is a contradiction.
From (2) to (1). From φ2 = id we have (dφ)2p = id, so only possible
eigenvalues of (dφ)p are ±1. Now it suffices to show all eigenvalues of (dφ)p
are −1. Otherwise if it has an eigenvalue 1, there exists some non-zero
v ∈ Tp M such that (dφ)p v = v. Since φ is an isometry, one has φ(expp (tv))
and expp (tv) are geodesics with the same direction at p. Thus

φ(expp (tv)) = expp (tv)

for 0 < t < inj(p). In particular, p is not an isolated fixed point, which is a
contradiction. □

Proposition 0.1.2. The fundamental group of a Riemannian symmetric

space is abelian.

Corollary 0.1. A surface of genus g ≥ 2 does not admit a Riemannian

metric with respect to which it is a symmetric space.
4

0.1.2. Locally Riemannian symmetric space.

Definition 0.1.2 (locally Riemannian symmetric space). A Riemannian
manifold (M, g) is called a locally Riemannian symmetric space if each p ∈
M has a neighborhood U such that there exists an isometry φ : U → U such
that φ(p) = p and (dφ)p = − id.
Theorem 0.1.1. Let (M, g) be a Riemannian manifold. Then the following
statements are equivalent:
(1) (M, g) is a locally Riemannian symmetric space.
(2) ∇R = 0.
Proof. From (1) to (2). If φ is the symmetry at point p ∈ M , then it’s an
isometry such that (dφ)p = − id, and thus for u, v, w, z ∈ Tp M , one has
−∇u R(v, w)z = (dφ)p (∇u R(v, w)z)
= ∇(dφ)p u ((dφ)p )v, (dφ)p w)(dφ)p z
= ∇u R(v, w)z
This shows (∇R)p = 0, and thus ∇R = 0 since p is arbitrary.
From (2) to (1). For arbitrary p ∈ M , it suffices to show
φ = expp ◦Φ0 ◦ exp−1
p : B(p, δ) → B(p, δ)

is an isometry, where 0 < δ < inj(p) and Φ0 = − id : Tp M → Tp M . For

e(t) = expp (tΦ0 (v)), if we define
v ∈ Tp M with |v| < δ and γ(t) = expp (tv), γ
Φt = P0,t;eγ ◦ Φ0 ◦ Pt,0;γ
then direct computation shows
Φ∗t Rγe(t) = (Pt,0;γ )∗ ◦ Φ∗0 ◦ (P0,t;eγ )∗ Rγe(t)
(a)
= (Pt,0;γ )∗ ◦ Φ∗0 Rγe(0)
(b)
= (Pt,0;γ )∗ Rγ(0)
(c)
= Rγ(t)
where
(a) and (c) holds from Proposition C.3.2.
e(0) = γ(0) and R is a (0, 4)-tensor.
(b) holds from γ
Then by Theorem C.3.2, that is Cartan-Ambrose-Hicks’s theorem, φ is an
isometry, which completes the proof. □
Remark 0.1.2. The proof for locally Riemannian symmetric space has par-
allel curvature tensor can be applied to other situations. For example, one
can easy show if a p-form ω is invariant under isometries, that is φ∗ ω = ω
for arbitrary isometry, then dω = 0, and in Section 8 we will use this idea
to show any almost Hermitian symmetric space is Kähler.
 5

0.2. Transvection.

Definition 0.2.1 (transvection). Let (M, g) be a Riemannian symmetric

space and γ be a geodesic. The transvection along γ is defined as

Tt = sγ( t ) ◦ sγ(0) ,
2

where sp is the symmetry at point p.

Proposition 0.2.1. Let (M, g) be a Riemannian symmetric space and Tt

be the transvection along geodesic γ. Then
(1) For any a, t ∈ R, sγ(a) (γ(t)) = γ(2a − t).
(2) Tt translates the geodesic γ, that is Tt (γ(s)) = γ(t + s).
(3) (dTt )γ(s) : Tγ(s) M → Tγ(t+s) M is the parallel transport Ps,t+s;γ .
(4) Tt is one-parameter subgroup of Iso(M, g).

Proof. For (1). It follows from the uniqueness of geodesics with given initial
value.
For (2). By (1) one has

Tt (γ(s)) = sγ( t ) ◦ sγ(0) (γ(s))

2

= sγ( t ) (γ(−s))
2

= γ(t + s).

For (3). Let X be a parallel vector field along γ. By uniqueness of parallel

vector fields with given initial data, we have (dsγ(0) )γ(s) Xγ(s) = −Xγ(−s) for
all s, since (dsγ(0) )γ(0) Xγ(0) = −Xγ(0) . Thus

(dTt )γ(s) Xγ(s) = (dsγ( t ) )γ(−s) (−Xγ(−s) )

2
= Xγ(t+s) .

This shows (dTt )γ(s) = Ps,t+s;γ .

For (4). In order to show Tt+s = Tt ◦ Ts , it suffices to check they’re same
at some point, so do their derivatives, since isometry can be determined by
these two values. Note that
Tt+s (γ(0)) = γ(t + s)
= Tt ◦ Ts (γ(0)),
(dTt+s )γ(0) = P0,t+s;γ
= Ps,t+s;γ ◦ P0,s;γ
= (dTt )γ(s) ◦ (dTs )γ(0)
= (d(Tt ◦ Ts ))γ(0) .

This completes the proof. □

6

0.3. Holonomy group.

Definition 0.3.1 (holonomy). Let (M, g) be a Riemannian manifold and γ
be a piecewise smooth loop centered at p ∈ M . Then the parallel along γ
gives an isometry on Tp M , and the set of all such isometries forms a group
called holonomy group, denoted by Holp (M, g).
Remark 0.3.1. Note that if q is another base point, and γ is a path from p to
q, then Holq = Pγ Holp Pγ−1 , and thus they are isomorphic, so for convenience
we just denote it by Hol.
Theorem 0.3.1. Let (M, g) be a Riemannian manifold. Then
(1) Hol is a Lie group and its identity component Hol0 is compact.
(2) Hol0 is given by parallel transport along null homotopic loops. As a
consequence, if M is simply-connected, then Hol = Hol0 .
Proposition 0.3.1. Let (M, g) be a Riemannian symmetric space with
G = Iso(M, g) and K = Gp for some p ∈ M . Then Holp ⊆ K.
Proof. Note that holonomy group is the group of parallel transports along
all piecewise smooth loops centered at p, and such a loop γ be written as
a limit of geodesic polygons γi . The parallel transport along any edge of
the polygon is given by applying a transvection along that edge, and so
the parallel transport along the full polygon is a composition of isometries
which sends p back to itself, hence it is an element of the isotropy group
K. Since K is compact, the sequence of parallel transports along geodesic
polygons approximating the given loop has a convergent subsequence, and
thus Holp ⊆ K. □
0.4. Symmetric space, locally symmetric space and homogeneous
space. In this section, we will show any complete locally Riemannian sym-
metric space is a quotient of Riemannian symmetric space (Corollary 0.2),
and any Riemannian symmetric space is a Riemannian homogeneous space
(Corollary 0.3).
0.4.1. Riemannian symmetric space and locally Riemannian symmetric space.
Theorem 0.4.1. Let (M, g) be a complete, simply-connected locally Rie-
mannian symmetric space. Then (M, g) is a Riemannian symmetric space.
Proof. For p ∈ M and 0 < δ < inj(p), suppose φ : B(p, δ) → B(p, δ) is an
isometry such that φ(p) = p and (dφ)p = − id. For arbitrary q ∈ M , we
use Ωp,q to denote all curves γ with γ(0) = p, γ(1) = q, and for c ∈ Ωp,q we
choose2 a covering {B(pi , δi )}ki=0 of c such that
(1) 0 < δi < inj(pi ).
(2) B(p0 , δ0 ) = B(p, δ) and pk = q.
(3) pi+1 ∈ B(pi , δi ).
2Since injective radius is a continuous function, it has a positive minimum on curve c,
so such covering exists.
 7

If we set φ = φ0 , then we can define isometries φi : B(pi , δi ) → M such

that φi (pi ) = φi−1 (pi ) and (dφi )pi = (dφi−1 )pi by using Cartan-Ambrose-
Hicks’s theorem successively, and by Theorem C.3.1 one has φi and φi+1
coincide on B(pi , δi ) ∩ B(pi+1 , δi ). The covering together with isometries we
construct is denoted by A = {B(pi , δi ), φi }ki=0 . For arbitrary x ∈ [0, 1], if
c(x) ∈ B(pm , δm ), we may define
φA (c(x)) := φm (c(x)),
(dφA )c(x) := (dφm )c(x) .
e pi , δei ), φ
In particular, φA (q) := φk (q). If B = {B(e ei }li=0 is another covering
of c, let’s show φA (q) = φB (q). Consider
I = {x ∈ [0, 1] | φA (c(x)) = φB (c(x)), (dφA )c(x) = (dφB )c(x) }.
It’s clear I 6= ∅, since 0 ∈ I. Now it suffices to show it’s both open and
closed to conclude 1 ∈ I.
e pn , δen ), that is
(a) It’s open: For x ∈ I, we assume c(x) ∈ B(pm , δm ) ∩ B(e
en (c(x)),
φm (c(x)) = φ
en )c(x) .
(dφm )c(x) = (dφ
Then one has
φm ◦ expc(x) (v) = expφm (c(x)) ◦(dφm )c(x) (v)
= expφen (c(x)) ◦(dφ
en )c(x) (v)
en ◦ expc(x) (v).
=φ
Since expc(x) maps onto a neighborhood of c(x), it follows that some
neighborhood of x also lies in I, and thus I is open.
(b) It’s closed: Let {xi }∞
i=1 ⊆ I be a sequence converging to x. Without
lose of generality we may assume {xi }∞ e pn , δen ), then
i=1 ⊆ B(pm , δm ) ∩ B(e
one has
en (c(xi )),
φm (c(xi )) = φ
en )c(xi ) .
(dφm )c(xi ) = (dφ
By taking limit we obtain the desired results.
Since φA (q) is independent of the choice of coverings, we use φ(q) to denote
it for convenience, and as a consequence we obtain the following map
F : Ωp,q → M
c 7→ φ(q).
Note that F (c) is locally constant, and thus it’s independent of the choice
of homotopy classes of c. Since M is simply-connected, one has F : Ωp,q →
M is constant, so we obtain a local isometry φ : M → M which extends
φ : B(p, δ) → B(p, δ). By Proposition C.3.1 φ is a Riemannian covering
map since M is complete, and thus φ is a diffeomorphism since M is simply-
connected, which implies φ is an isometry. □
8

Corollary 0.2. Let (M, g) be a complete locally Riemannian symmetric

f/Γ, ge) where (M
space. Then it’s isometric to (M f, ge) is a Riemannian sym-
metric space and Γ ∼ = π1 (M ) is a discrete Lie group acting on M f freely,
properly and isometrically.
f, ge) be the universal covering of (M, g) with pullback met-
Proof. Let (M
f, ge) is a simply-connected Riemannian manifold with parallel
ric. Then (M
curvature tensor. Moreover, by Proposition C.3.3 it’s complete, hence it is
symmetric. □
0.4.2. Riemannian symmetric space and Riemannian homogeneous space.
Definition 0.4.1 (Riemannian homogeneous space). A Riemannian mani-
fold (M, g) is called a Riemannian homogeneous space, if Iso(M, g) acts on
M transitively.
Proposition 0.4.1. Let (M, g) be a Riemannian homogeneous space. If
there exists a symmetry at some point p ∈ M , then (M, g) is a Riemannian
symmetric space.
Proof. Let φ be a symmetry at p ∈ M . For arbitrary q ∈ M , there exists
an isometry ψ : M → M such that ψ(p) = q since (M, g) is a Riemannian
homogeneous space. Then
φq := ψ ◦ φ ◦ ψ −1
is the desired symmetry at q. □
Theorem 0.4.2. Let (M, g) be a Riemannian symmetric space. Then
(1) (M, g) is complete.
(2) the identity component of isometry group acts transitively on M .
Proof. For (1). For arbitrary geodesic γ : [0, 1] → M with γ(0) = p, γ ′ (0) =
v, the curve β(t) = φ(γ(t)) : [0, 1] → M is also a geodesic with β(0) = p and
β ′ (0) = −v. Now we obtain a smooth extension γ ′ : [0, 2] → M of γ, given
by (
′ β(1 − t), t ∈ [0, 1]
γ (t) =
γ(t − 1), t ∈ [1, 2].
Repeat above process to extend γ to a geodesic defined on R, which shows
completeness.
For (2). For p, q ∈ M , let γ be a geodesic connecting p, q. Then the
transvection along γ gives an isometry which maps p to q. Since the transvec-
tion lies in the identity component of isometry group, one has the identity
component of isometry group acts transitively on M . □
Corollary 0.3. The Riemannian symmetric space (M, g) is a Riemannian
homogeneous space.
 9

1. Algebraic viewpoints
1.1. Riemannian symmetric space as a Lie group quotient.
Definition 1.1.1 (involution). An automorphism σ of a Lie group G is
called an involution if σ 2 = idG .
Definition 1.1.2 (Cartan decomposition). Let G be a Lie group and σ be
an involution of G. The eigen-decomposition of g given by (dσ)e is called
Cartan decomposition, that is,
g = k ⊕ m,
where
k = {X ∈ g | (dσ)e (X) = X},
m = {X ∈ g | (dσ)e (X) = −X}.
Proposition 1.1.1. Let g = k ⊕ m be the Cartan decomposition given by
σ. Then
[k, k] ⊆ k, [k, m] ⊆ m, [m, m] ⊆ k.
Proof. Since σ is a Lie group homomorphism, (dσ)e gives a Lie algebra
homomorphism, and thus
(dσ)e ([X, Y ]) = [(dσ)e (X), (dσ)e (Y )],
where X, Y ∈ g. □
Lemma 1.1.1. Let G be a Lie group and K ⊆ G be a closed subgroup.
A left invariant metric on G which is also right invariant under K gives a
left-invariant metric on G/K.
Theorem 1.1.1. Let (M, g) be a Riemannian symmetric space and G =
(Iso(M, g))0 . For p ∈ M , K denotes the isotropic group of Gp .
(1) The mapping σ : G → G, given by σ(g) = sp gsp is an involution auto-
morphism of G.
(2) If Gσ is the set of fixed points of σ in G, then (Gσ )0 ⊆ K ⊆ Gσ .
(3) If g = k ⊕ m is the Cartan decomposition given by σ, then k is the Lie
algebra of K, and thus m ∼ = Tp M as vector spaces.
(4) There is a left invariant metric on G/K such that G/K with this metric
is isometric to (M, g).
Proof. For (1). It’s clear σ preserves G, and it’s an involution since for
arbitrary g ∈ G, one has σ 2 (g) = σ(sp gsp ) = s2p gs2p = g.
For (2). It follows from the following two steps:
(a) To show K ⊆ Gσ . For any k ∈ K, in order to show k = sp ksp , it suffices
to show they and their differentials agree at some point by Theorem
C.3.1, since both of them are isometries, and p is exactly the point we
desired.
10

(b) To see (Gσ )0 ⊆ K. Suppose exp(tX) ⊆ (Gσ )0 is a one-parameter sub-

group. Since σ(exp(tX)) = exp(tX), one has
exp(tX)(p) = sp exp(tX)sp (p) = sp exp(tX)(p).

But p is an isolated fixed point of sp , which implies exp(tX)(p) = p for all

t. This shows the one-parameter subgroup lies in K. Since exponential
map of Lie group is a diffeomorphism in a small neighborhood of identity
element e and (Gσ )0 can be generated by a neighborhood of e, which
implies the whole (Gσ )0 ⊆ K.
For (3). Note that (Gσ )0 ⊆ K ⊆ Gσ , it suffices to show k ∼ = Lie Gσ . For
X ∈ k, we claim γ2 (t) = σ(exp(tX)) : R → G is a one-parameter subgroup.
Indeed, note that
γ2 (t) · γ2 (s) = sp exp(tX)sp · sp exp(sX)sp
= σ(exp(tX + sX))
= γ2 (t + s).

Moreover, γ2 (t) = σ(exp(tX)) and γ1 (t) = exp(tX) are two one-parameter

subgroups of G such that γ1 (0) = γ2 (0) and γ2′ (0) = (dσ)e (X) = X = γ1′ (0).
Then γ1 (t) = γ2 (t), and thus exp(tX) ∈ Gσ for all t ∈ R. This shows
k ⊆ Lie Gσ , and the converse inclusion is clear, so one has k = Lie Gσ .
For (4). Let π : G → M be the natural projection given by π(g) = gp.
Then for k ∈ K and X ∈ g one has
 
d −1
(dπ)e (Ad(k)X) = (dπ)e k exp(tX)k
dt t=0
d
= π(k exp(tX)k −1 )
dt t=0
d
= k exp(tX)k −1 · p
dt t=0
d
= k exp(tX) · p
dt t=0
= (dLk )p (dπ)e (X).

By using the equivalent isomorphism (dπ)e |m : m → Tp M , one has an Ad(K)-

invariant metric on m, and then we can extend it to an Ad(K)-invariant
metric on g = k⊕m by choosing3 arbitrary Ad(K)-invariant metric on k such
that m ⊥ k. This shows one has a left-invariant metric on G which is also
right invariant with respect to K, and by Lemma 1.1.1 it gives a left-invariant
metric on G/K. Now it suffices to show G/K with this metric is isometric to
(M, g). For any gK ∈ G/K, consider the following communicative diagram

3Such metric exists since K is compact.

 11

(dπ)e |m
m = TeK G/K Tp M
dLg dLg

TgK G/K Tgp M

Since both (dπ)e |m and (dLg ) are linear isometries, one has TgK G/K is
isometric to Tgp M , and thus G/K with this metric is isometric to (M, g). □

1.2. Riemannian symmetric pair. In Theorem 1.1.1 one can see that if
(M, g) is a symmetric space, then it gives a pair of Lie groups (G, K) with
an involution σ on G such that
(Gσ )0 ⊆ K ⊆ Gσ .
Moreover, there exists a left-invariant metric on G/K such that G/K with
this metric is isometric to (M, g). This motivates us a useful way to con-
struct Riemannian symmetric spaces from a pair of Lie groups with certain
properties, and such a pair is called a Riemannian symmetric pair.
Definition 1.2.1 (Riemannian symmetric pair). Let G be a connected Lie
group and K ⊆ G be a closed subgroup. The pair (G, K) is called a sym-
metric pair if there exists an involution σ : G → G with (Gσ )0 ⊆ K ⊆ Gσ .
If, in addition, the group Ad(K) ⊆ GL(g) is compact, then (G, K) is said
to be a Riemannian symmetric pair.
Remark 1.2.1. The first condition of above definition means K is compact up
to the center of G since the kernel of Ad is the center of G. By Theorem 1.1.1
every Riemannian symmetric space gives a Riemannian symmetric pair.
Definition 1.2.2 (associated). If (M, g) is a Riemannian symmetric space,
G = (Iso(M, g))0 and K is the isotropy group Gp of some point p ∈ M , then
(G, K) is a Riemannian symmetric pair. In this case (G, K) is called the
Riemannian symmetric pair associated to (M, g).
Proposition 1.2.1. Let (G, K) be a symmetric pair given by σ. Then there
is an isomorphism as Lie algebras
k∼= Lie K,
and an isomorphism as vector spaces
m∼
= TeK G/K
Proof. It’s the same as proof of (3) in Theorem 1.1.1. □
Corollary 1.1. Let σe : G/K → G/K be the automorphism given by σ
e(gK) =
σ(g)K. Then (deσ )eK = − idG/K .
Proof. σe is well-defined since K ⊆ Gσ , and by construction one has (de
σ )eK =
(dσ)e |m . Then (deσ )eK = − idG/K since m = {X ∈ g | (dσ)e X = −X}. □
12

Theorem 1.2.1. Let (G, K) be a Riemannian symmetric pair given by σ.

Then there exists a left-invariant metric on M = G/K making it to be a
Riemannian symmetric space.
Proof. Since Ad(K) ⊆ GL(g) is a compact subgroup, by averaging trick
there exists an inner product on g which is also Ad(K)-invariant, and thus
it gives a left-invariant metric on M by Lemma 1.1.1. Moreover, by Corollary
1.1 one has (de σ )eK = − idM .
Now it suffices to show for any gK ∈ M , (de σ )gK : TgK M → Tσ(g)K M is
an isometry. Note that σ e(ghK) = σ(g)σ(h)K = σ(g)e σ (hK) holds for all
h ∈ G. This shows σ e ◦ Lg = Lσ(g) ◦ σ
e, where Lg : M → M is given by
Lg (hK) = ghK. By taking differential one has the following communicative
diagram
(de
σ )eK
TeK M TeK M
(dLg )eK (dLσ(g) )eK
(de
σ )gK
TgK M Tσ(g)K M

Since (dLg )eK , (dLσ(g) )eK , (de

σ )eK are isometries, one has (de
σ )gK is also an
isometry as desired. □
Remark 1.2.2. In Theorem 2.1.1 we will see the curvature tensor of G/K is
independent of the choice of the left-invariant metric on it, so here we only
care about existence, which is guaranteed by Ad(K) is compact.
1.3. Examples of Riemannian symmetric pair.
Example 1.3.1. G = SL(n, R) together with K = SO(n) gives a Riemann-
ian symmetric pair, where σ is defined by
σ : SL(n, R) → SL(n, R)
g 7→ (g −1 )T .
Indeed, note that
(SL(n, R))σ = SO(n).
Thus SL(n, R)/ SO(n) is a Riemannian symmetric space, and it can be
viewed as a generalization of hyperbolic plane H2 , since SL(2, R)/ SO(2) ∼
=
H2 .
Example 1.3.2. G = SO(n + 1) together with K = SO(n) gives a Rie-
mannian symmetric pair, where σ is defined by
σ : SO(n + 1) → SO(n + 1)
−1
a 7→ I1,n aI1,n ,
where I1,n = diag{−1, 1, . . . , 1}. Indeed, a direct computation shows
  
±1 0
SO(n+1)σ = {a ∈ SO(n+1) | I1,n a = aI1,n } = ∈ SO(n + 1) | b ∈ O(n) ,
0 b
 13

which implies (SO(n + 1)σ )0 = SO(n) ⊆ SO(n + 1). Thus S n ∼

= SO(n +
1)/ SO(n) is a Riemannian symmetric space.
Example 1.3.3 (compact Grassmannian). Consider the Grassmannian of
oriented k-planes in Rk+l , denoted by M = Gr c k (Rk+l ). It’s clear that
SO(k + l) acts on M transitively with isotropy group SO(k) × SO(l), and
thus M ∼= SO(k + l)/ SO(k) × SO(l). Consider the involution
σ : SO(k + l) → SO(k + l)
−1
a 7→ Ik,l aIk,l ,
where Ik,l = diag{−1, . . . , −1, 1, . . . , 1}. A direct computation shows
| {z } | {z }
k times l times
SO(k + l)σ = S(O(k) × O(l)).
Then (SO(k + l)σ )0 = SO(k) × SO(l) ⊆ SO(k + l)σ , and thus M is a Rie-
mannian symmetric space, called compact Grassmannian. In particular,
c 1 (Rn+1 ).
S n = Gr
Example 1.3.4 (hyperbolic Grassmannian). In Rk,l with k ≥ 2, l ≥ 1, let’s
consider the following quadratic form
  X
k X
k+l
t Ik 0
t
v Ik,l w = v w= vi wi − vj wj .
0 −Il
i=1 j=k+1
The group of linear transformation X that preserves this quadratic form is
denoted by O(k, l), that is
XIk,l X t = Ik,l ,
and SO(k, l) are those with positive determinant. Now consider set con-
sisting of those oriented k-dimensional subspaces of Rk,l on which quadratic
form Ik,l are positive definite. This space is called the hyperbolic Grass-
mannian M = Gr c k (Rk,l ), which is also an open subset of Gr
c k (Rk+l ). It’s
clear G = SO(k, l) acting transitively on M with isotropy group Gp =
SO(k) × SO(l). As in Example 1.3.3 one can also construct an involution σ
to show Grc k (Rk,l ) is a Riemannian symmetric space.

Example 1.3.5. Suppose K is a compact connected Lie group. Then (K ×

K, ∆K) is a Riemannian symmetric pair given by σ, where σ : K × K →
K × K is given by (x, y) 7→ (y, x), since
(K × K)σ = {(a, a) | a ∈ K} = ∆K.
Then any compact Lie group is a Riemannian symmetric space.
14

2. Curvature of Riemannian symmetric space

2.1. Formulas. Let (M, g) be a Riemannian symmetric space with isometry
group G and isotropy group Gp . On one hand, there is a Cartan decompo-
sition of Lie algebra g given by involution σ : G → G, that is
g = k ⊕ m,
where m ∼= Tp M as vector spaces, and k is the Lie algebra of isotropy group
Gp . On the other hand, by Corollary C.3 there is another decomposition of
g given by
g = k′ ⊕ m′ ,
where
k′ = {X ∈ g | Xp = 0},
m′ = {X ∈ g | (∇X)p = 0}.
In fact, for any complete Riemannian manifold, the following proposition
shows k ∼= k′ , and thus above two Cartan decompositions are exactly the
same.
Proposition 2.1.1. Let (M, g) be a complete Riemannian manifold with
isometry group G and isotropy group Gp . Then the Lie algebra of Gp is
{X ∈ g | Xp = 0}.
Proof. Let X ∈ g with Xp = 0 and φt : M → M be the flow of X. If we
denote γp (t) = φt (p), then it suffices to show γp (t) ≡ p. For any smooth
function f : M → R, one has
d
γp′ (s)f = f ◦ γp (t)
dt t=s
d
= f ◦ γp (s + t)
dt t=0
d
= (f ◦ φs )(γp (t))
dt t=0
= Xp (f ◦ φs )
=0
□
Proposition 2.1.2. Let (M, g) be a Riemannian symmetric space and G =
Iso(M, g) with Lie algebra g. For any p ∈ M , one has Cartan decomposition
g = k ⊕ m. Then for any S ∈ k, one has
B(S, S) ≤ 0,
where B is the Killing form of g. Moreover, the identity holds if and only if
S = 0.
 15

Proof. Since a Killing field is determined by Xp and (∇X)p , one has elements
in k are determined by (∇X)p , and note that ∇X is a skew-symmetric
matrice, so
k∼ = {(∇X)p ∈ so(Tp M ) | X ∈ k}.
By using this identification, there is a natural inner product on k given by
hS1 , S2 i = tr(S1 S2T ) = − tr(S1 S2 ).
By adding inner product on m obtained from m ∼ = Tp M and the one on k
constructed as above together, one can construct an inner product on g such
that g = k ⊕ m is orthogonal. For any S ∈ k, we claim with respect to this
metric, ad(S) : g → g is skew-symmetric. Indeed, for X1 , X2 ∈ k, one has
had(S)X1 , X2 i = − tr((ad(S)X1 )X2 )
= − tr((SX1 − X1 S)X2 )
= tr(X1 (SX2 − X2 S))
= −hX1 , ad(S)X2 i.
For Y1 , Y2 ∈ m, since Sp = 0 and (∇S)p is skew-symmetric, one has
had(S)Y1 , Y2 i = h∇S Y1 − ∇Y1 S, Y2 i
= −h∇Y1 S, Y2 i
= h∇Y2 S, Y1 i
= −hY1 , ∇S Y2 − ∇Y2 Si
= −hY1 , ad(S)Y2 i.
If X ∈ k and Y ∈ m, since [k, k] ⊆ k and [k, m] ⊆ m, one has
had(S)X, Y i = 0,
hX, ad(S)Y i = 0.
Similarly one has
had(S)Y, Xi = 0,
hY, ad(S)Xi = 0.
This completes the proof of our claim. Then one has
X X
B(S, S) = tr(ad(S)◦ad(S)) = had(S)◦ad(S)ei , ei i = − had(S)ei , ad(S)ei i ≤ 0.
i i
Moreover, if B(S, S) = 0, then ad(S) = 0 and for any X ∈ g, one has
0 = ad(S)X = ∇S X − ∇X S = −∇X S,
since Sp = 0. This implies (∇S)p = 0, and thus S = 0. □
Remark 2.1.1. For S ∈ k, the most important part of the proof of B(S, S) = 0
if and only if S = 0 is ad(S) = 0 if and only if S = 0. In other words,
k ∩ z = {0}, where z is the center of Lie algebra g.
Theorem 2.1.1. Let (M, g) be a Riemannian symmetric space and G =
Iso(M, g). For any p ∈ M , g = k ⊕ m with m ∼
= Tp M .
16

(1) For any X, Y, Z ∈ m, there holds

R(X, Y )Z = −[Z, [Y, X]],
1
Ric(Y, Z) = − B(Y, Z).
2
(2) If Ric(g) = λg, then for X, Y ∈ m, one has
2λR(X, Y, Y, X) = −B([X, Y ], [X, Y ]).
Proof. For (1). For any X, Y, Z ∈ m, direct computation shows
(a)
R(X, Y )Z = R(X, Z)Y − R(Y, Z)X
(b)
= ∇Z ∇Y X − ∇ ∇Z Y X − ∇ Z ∇X Y + ∇∇Z X Y
(c)
= −∇Z [X, Y ]
(d)
= −[Z[X, Y ]],
where
(a) holds from the first Bianchi identity.
(b) holds from (2) of Proposition C.1.1.
(c) holds from X, Y ∈ m, and thus (∇X)p = (∇Y )p = 0.
(d) holds from
∇Z [X, Y ] − ∇[X,Y ] Z = [Z, [X, Y ]],

and (∇Z)p = 0.
To see Ricci curvature, note that for Y ∈ m,
ad(Y ) : k → m, ad(Y ) : m → k.
Thus if Y, Z ∈ m, one has ad(Z) ◦ ad(Y ) preserves the decomposition g =
k ⊕ m. Then
tr(ad(Z) ◦ ad(Y )|m ) = tr(ad(Z)|k ◦ ad(Y )|m )
= tr(ad(Y )|m ◦ ad(Z)|k )
= tr(ad(Y ) ◦ ad(Z)|k ).
Hence we obtain
B(Y, Y ) = tr(ad(Y ) ◦ ad Y |k ) + tr(ad(Y ) ◦ ad Y |m ) = 2 tr(ad(Y ) ◦ ad(Y )|m ).
Since Ricci tensor is trace of curvature tensor, and thus
1
Ric(Y, Y ) = − tr(ad(Y ) ◦ ad(Y )|m ) = − B(Y, Y ).
2
Then by using polarization identity, one has Ric(Y, Z) = −B(Y, Z)/2.
 17

For (2). If Ric(g) = λg, then

2λg(R(X, Y )Y, X) = −2λg(ad(Y ) ◦ ad(Y )X, X)

= −2 Ric(ad(Y ) ◦ ad(Y )X, X)
= B(ad(Y ) ◦ ad(Y )X, X)
= −B(ad(Y )X, ad(Y )X)
= −B([X, Y ], [X, Y ]).

Corollary 2.1. Let (M, g) be a Riemannian symmetric space which is an

Einstein manifold with Einstein constant λ. Then
(1) If λ > 0, then (M, g) has non-negative sectional curvature.
(2) If λ < 0, then (M, g) has non-positive sectional curvature.
(3) If λ = 0, then (M, g) is flat.

Proof. By Theorem 2.1.1 one has

2λR(X, Y, Y, X) = −B([X, Y ], [X, Y ]) ≥ 0,

since [X, Y ] ∈ [m, m] ⊆ k and B is negative definite on k. This shows (1)

and (2). If λ = 0, one has B([X, Y ], [X, Y ]) ≡ 0 for arbitrary X, Y . Then
by Proposition 2.1.2 one has [X, Y ] ≡ 0 for arbitrary X, Y , and thus (M, g)
is flat. □

2.2. Computations.

Example 2.2.1. In Example 1.3.1 we have already shown that M = SL(n, R)/ SO(n)
is a Riemannian symmetric space. Consider its Cartan decomposition

sl(n) = so(n) ⊕ m,

where m consists of symmetric matrices and m ∼ = Tp M for p ∈ M . On m we

can put the usual Euclidean metric, that is for X, Y ∈ m, we define

1
hX, Y i = tr(XY T ) = tr(XY ) = B(X, Y ),
2n
where B is the Killing form of sl(n). By Theorem 2.1.1 the corresponding
metric on M has the curvature formulas
B
Ric(g) = − = −ng,
2
B([X, Y ], [X, Y ])
R(X, Y, Y, X) = ≤ 0.
2n
18

Hence it has non-positive sectional curvatures. One can also show its sec-
tional curvature is non-positive by computing curvature tensor as follows
R(X, Y, Z, W ) = tr([Z, [X, Y ]]W )
= tr(Z[X, Y ]W − [X, Y ]ZW )
= tr(W Z[X, Y ] − [X, Y ]ZW )
= tr([X, Y ][Z, W ])
= − tr([X, Y ][Z, W ]T )
= −h[X, Y ], [Z, W ]i.
Example 2.2.2 (compact Grassmannian). In Example 1.3.3 we have al-
c k (Rk+l ) is a Riemannian symmetric space with
ready shown that M = Gr
Cartan decomposition
so(k + l) = so(k) ⊕ so(l) ⊕ m,
where m =∼ Tp M for p ∈ M . Note that one has the block decomposition of
matrices in so(k + l) as follows
  
X1 B
so(k + l) = | X1 ∈ so(k), X2 ∈ so(l), B ∈ Mk×l (R) .
−B T X2
  
∼ 0 B
Then one has m = | B ∈ Mk×l (R) . If we put the usual
−B T 0
Euclidean metric on m, that is
      T !
0 A 0 B 0 A 0 B
, = tr
−AT 0 −B T 0 −AT 0 −B T 0
  
0 A 0 B
= − tr
−AT 0 BT 0
   
1 0 A 0 B
=− B , ,
k+l−2 −AT 0 −B T 0
where B is the Killing form of so(n). Then the corresponding metric on M
has the curvature formulas
B k+l−2
Ric(g) = − = g,
2 2
B([X, Y ], [X, Y ])
R(X, Y, Y, X) = − ≥ 0,
k+l−2
where X, Y ∈ m. This shows the compact Grassmannian has the non-
negative sectional curvature.
Example 2.2.3 (hyperbolic Grassmannian). In Example 1.3.4 we have al-
c k (Rk,l ) is a Riemannian symmetric space with
ready shown that M = Gr
Cartan decomposition
so(k, l) = so(k) ⊕ so(l) ⊕ m,
 19

where m ∼ = Tp M for p ∈ M . Note that one has the block decomposition of

matrices in so(k, l) as follows
  
X1 B
so(k, l) = | X1 ∈ so(k), X2 ∈ so(l), B ∈ Mk×l (R) .
B T X2
  
∼ 0 B
Then one has m = | B ∈ Mk×l (R) . If we put the usual Eu-
BT 0
clidean metric on m, then
       
0 A 0 B 1 0 A 0 B
, = B , ,
AT 0 BT 0 k+l−2 AT 0 BT 0
where B is the Killing form of so(k, l). Then the corresponding metric on
M has the curvature formulas
B k+l−2
Ric(g) = − = − g,
2 2
B([X, Y ], [X, Y ])
R(X, Y, Y, X) = ≤ 0,
k+l−2
where X, Y ∈ m. This shows the hyperbolic Grassmannian has non-positive
sectional curvature.
Remark 2.2.1. Later we will see compact Grassmannian and hyperbolic
Grassmannian are dual to each other in Example 5.0.2.
Example 2.2.4. In Example 1.3.5 one has a compact connected Lie group
G∼= G×G/G∆ is a Riemannian symmetric space with Cartan decomposition
g ⊕ g = g∆ ⊕ g⊥ , where
g∆ = {(X, X) | X ∈ g},
g⊥ = {(X, −X) | X ∈ g}.
Then one has m ∼
= g⊥ , and thus curvature tensor can be computed as follows
R(X, Y )Z = R((X, −X), (Y, −Y ))(Z, −Z)
= [(Z, −Z), [(X, −X), (Y, −Y )]]
= ([Z, [X, Y ]], −[Z, [X, Y ]]).
Hence, we arrive at that the formula
R(X, Y )Z = [Z, [X, Y ]].
Remark 2.2.2. If one computes the curvature tensor in the standard way
using bi-invariant metric, then the formula has a factor 1/4 on it.
20

Part 2. Classifications of Riemannian symmetric space

3. Decompositions
So far, we have seen that any Riemannian symmetric space (M, g) gives a
Riemannian symmetric pair (G, K) with involution σ, and any Riemannian
symmetric pair gives a pair (g, s) of Lie algebra g and involution s of g
with Cartan decomposition g = k ⊕ m. In this section, we will study such
pairs of Lie algebras and prove decomposition theorems, which will give
decomposition theorems for symmetric spaces.

3.1. Orthogonal symmetric Lie algebra.

Definition 3.1.1 (orthogonal symmetric Lie algebra). An orthogonal sym-
metric Lie algebra is a pair (g, s) consisting of a real Lie algebra g and an
involution s 6= id of g such that k is a compactly embedded subalgebra4,
where k is given by Cartan decomposition g = k ⊕ m.
Remark 3.1.1. For an orthogonal symmetric Lie algebra (g, s), the term
”orthogonal” is motivated by the fact that Cartan decomposition g = k ⊕ m
is an orthogonal direct sum with respect to the Killing form of g.
Definition 3.1.2 (isomorphism). Two orthogonal symmetric Lie algebra
(g1 , s1 ), (g2 , s2 ) are called isomorphic to each other, if there exists a Lie
algebra isomorphism ρ : g1 → g2 such that s2 ◦ ρ = ρ ◦ s1 .
Definition 3.1.3 (effective). Let (g, s) be an orthogonal symmetric Lie
algebra with Cartan decomposition g = k⊕m. It’s called effective if k∩z = 0,
where z is the center of g.
Example 3.1.1. Let (G, K) be a Riemannian symmetric pair given by invo-
lution σ. Then it gives an orthogonal symmetric pair (g, s), where g = Lie G
and s = (dσ)e . Moreover, if (G, K) is given by a Riemannian symmetric
space, then (g, s) is effective.
Proposition 3.1.1. Let (g, s) be an effective orthogonal symmetric Lie
algebra with Cartan decomposition g = k ⊕ m. Then the Killing form of g
is negative definite on k.
Proof. Let B be the Killing form of g and K ⊆ GL(g) be the compact Lie
group such that Lie K = adg (k). Without lose of generality we may assume
K ≤ SO(n), and thus adg (k) consisting of skew-symmetric matrices. Hence
for S ∈ k,
X X
B(S, S) = tr(ad(S)◦ad(S)) = had(S)◦ad(S)ei , ei i = − had(S)ei , ad(S)ei i ≤ 0,
i i

and the equality holds if and only if S ∈ z ∩ k = 0. □

4See Definition B.2.1

 21

Theorem 3.1.1. Every orthogonal symmetric Lie algebra (g, s) with Cartan
decomposition g = k ⊕ m gives a Riemannian symmetric pair (G, K) with
G/K simply-connected.
Proof. By Theorem B.1.1 there exists a unique connected and simply-connected
e such that Lie G
Lie group G e = g and there also exists a unique connected Lie
e e
subgroup K ⊆ G with Lie algebra k by Theorem B.1.2. Moreover, by The-
e→G
orem B.1.3 there exists a unique involution σ : G e such that (dσ)e = s,
e e
and K is the identity component of Gσ . Then (G, K)e e is the Riemannian
symmetric pair given by σ. To see M = G/ e K e is a simply-connected Rie-
mannian symmetric space, we consider the exact sequence
0→K e →G e → M → 0,
which gives a long exact sequence of homotopy groups as
e → π1 (M ) → π0 (K)
· · · → π1 (G) e → ···
Since Ke is connected and G
e is simply-connected, M is simply-connected as
desired. □
Remark 3.1.2. In above case, G-action on M = G/K may not be effective,
and it’s almost effective5 if and only if (g, s) is effective.
3.2. Decomposition into pieces of different types.
3.2.1. Types.
Definition 3.2.1 (types). Let (g, s) be an effective orthogonal symmetric
Lie algebra with Cartan decomposition g = k ⊕ m and Killing form B. Then
(g, s) is of
(1) of compact type if B|m < 0;
(2) of non-compact type if B|m > 0;
(3) of Euclidean type if B|m = 0;
(4) of semisimple type if g is semisimple, or equivalently, B is non-degenerate.
Remark 3.2.1. It’s clear is an effective orthogonal symmetric Lie algebra is
of compact type or non-compact type, then it’s of semisimple type, since by
Proposition 3.1.1 B|k < 0, and g = k ⊕ m is an orthogonal direct sum with
respect to B.
Definition 3.2.2 (types).
(1) A Riemannian symmetric pair is of one of above types if its correspond-
ing orthogonal symmetric Lie algebra is.
(2) A Riemannian symmetric space is of one of above types if its corre-
sponding Riemannian symmetric pair is.
Proposition 3.2.1. Let (g, s) be an effective orthogonal symmetric Lie
algebra with Cartan decomposition g = k ⊕ m. It’s of Euclidean type if and
only if [m, m] = 0.
5A group acting on a set almost effectively if only finite many elements act trivially.
22

Proof. If (g, s) is of Euclidean type, then B(k, m) = 0 and B|k < 0 implies
m is the kernel of Killing form B, and thus m is an ideal. Then
[m, m] ⊆ m ∩ k = 0.
Conversely, if [m, m] = 0, then by definition of Killing form it’s clear B|m =
0. □
Proposition 3.2.2. Let (G, K) be a Riemannian symmetric pair and M =
G/K.
(1) If (G, K) is of compact type, then M has non-negative sectional curva-
ture.
(2) If (G, K) is of non-compact type, then M has non-positive sectional
curvatures.
(3) If (G, K) is of Euclidean type, then M is flat6. In particular, if M is
simply-connected, then it’s isometric to Rn .
Proof. If (G, K) is of compact type, we may assume Ad(K)-invariant inner
product on m is given by −B|m , and thus by 2.1.1 one has
1
Ric = − B.
2
This shows M is Einstein with Einstein constant 1/2, and thus by Corollary
2.1 one has M has non-negative sectional curvature. Similarly one can
show if (G, K) is of non-compact type, then M has non-positive sectional
curvatures, and (G, K) is of Euclidean type, then M is flat. □

3.2.2. Decomposition of effective orthogonal symmetric Lie algebra.

Theorem 3.2.1. Let (g, s) be an effective orthogonal symmetric Lie algebra
and B be the Killing form of g. Then there exists ideals g0 , g− and g+ with
the following properties:
(1) g = g0 ⊕ g− ⊕ g+ .
(2) g0 , g− and g+ are invariant under s and orthogonal with respect to
Killing form B of g.
(3) Let s0 , s− , s+ be the restrictions of s to g0 , g− and g+ . The pairs
(g0 , s0 ), (g− , s− ) and (g+ , s+ ) are effective orthogonal symmetric Lie
algebras of the Euclidean type, compact type and non-compact type,
respectively.
Proof. See Theorem 1.1 in Chapter V of [Hel78]. □

3.2.3. Decomposition of Riemannian symmetric space.

Theorem 3.2.2. Let (M, g) be a simply-connected symmetric space. Then
M = M0 × M+ × M− is the Riemannian product of symmetric space of
Euclidean, non-compact and compact types respectively.
6A Riemannian manifold is called flat, if its sectional curvatures are zero.
 23

Proof. Let (G, K) with involution σ be the Riemannian symmetric pair given
by (M, g) and (g, s) be the corresponding effective orthogonal symmetric
e → G be the universal covering and K
Lie algebra. Let p : G e be the identity
−1
component of p (K). Then it induces a covering map of p : G/ e Ke → G/K
e e
by g K → p(g)K. Since M is simply-connected, M = G/K. e e
By Theorem 3.2.1, we obtain a decomposition g = g0 ⊕ g− ⊕ g+ . By
Theorem B.1.1, there exists simply-connected Lie groups G0 , G− and G+
e = G0 ×
with Lie algebras g0 , g− and g+ . Then it gives a decomposition G
G− × G+ . If K e = K0 × K− × K+ is the corresponding decomposition,
then the spaces M0 = G0 /K0 , M− = G− /K− and M+ = G+ /K+ gives the
desired decomposition. □
24

4. Irreducibility
4.1. Irreducible orthogonal symmetric Lie algebra.
Definition 4.1.1 (irreducible). Suppose (g, s) is an orthogonal symmetric
Lie algebra with Cartan decomposition g = k ⊕ m. Then (g, s) is called
irreducible if
(1) g is semisimple and k contains no ideal of g;
(2) the Lie algebra ad(k) acts irreducibly on m.
Remark 4.1.1. Any irreducible orthogonal symmetric Lie algebra (g, s) is
effective, since z ∩ k is an ideal in k, and thus vanishes.
Definition 4.1.2 (irreducible).
(1) A Riemannian symmetric pair is called irreducible if its corresponding
orthogonal symmetric Lie algebra is.
(2) A Riemannian symmetric space is called irreducible if its corresponding
Riemannian symmetric pair is.
Lemma 4.1.1 (Schur lemma). Let B1 , B2 be two symmetric bilinear forms
on a vector space V such that B1 is positive definite. If a group K acts
irreducibly on V such that B1 and B2 are invariant under K, then B2 = λB1
for some constant λ.
Proof. Since B1 is positive definite, there exists an endomorphism L : V →
V such that
B2 (u, v) = B1 (Lu, v),
where u, v ∈ V . Since B1 , B2 are invariant under K, one has for any k ∈ K
B1 (kLu, v) = B1 (Lu, k −1 v) = B2 (u, k −1 v) = B2 (ku, v) = B1 (Lku, v),
holds for arbitrary u, v ∈ V , which implies Lk = kL for all k ∈ K. On the
other hand, the symmetry of B1 , B2 implies
B1 (Lu, v) = B2 (u, v) = B2 (v, u) = B1 (Lv, u) = B1 (u, Lv).
Hence L is symmetric with respect to B1 , and thus the eigenvalues of L are
real. If 0 6= E ⊆ V is an eigenspace with eigenvalue λ, the fact kL = Lk
implies E is invariant under K. Since K acts irreducibly on V , one has
E = V , that is L = λI, which implies B2 = λB1 . □
Proposition 4.1.1. Let (G, K) be an irreducible Riemannian symmetric
pair given by σ. Then there is up to scaling a unique left-invariant metric
on M = G/K.
Proof. It suffices to show there is up to scaling a unique Ad(K)-invariant
inner product on m. Since (G, K) is an irreducible Riemannian symmetric
pair, then K acts on m irreducibly by adjoint representation, and thus by
Lemma 4.1.1 any two Ad(K)-invariant inner product on m are scalar multi-
ples of each other. In particular, −B|m and B|m give such an inner product
in compact and non-compact cases respectively. □
 25

4.2. Decomposition into irreducible pieces.

Theorem 4.2.1. Let (g, s) be an effective orthogonal symmetric Lie algebra
with Cartan decomposition g = k ⊕ m such that g is semisimple and k does
not contain an ideal of g. Then there are ideals (gi )i∈I of g such that
L
(1) g = i gi .
(2) The ideals gi are mutually orthogonal with respect to Killing form B of
g, and they are invariant under s.
(3) Denoting by si the restriction if s to gi , each (gi , si ) is an irreducible
orthogonal symmetric Lie algebra.
Proof. See Proposition 5.2 in Chapter VIII of [Hel78]. □
As Theorem 3.2.2, this decomposition of effective orthogonal symmetric
Lie algebra gives a decomposition of Riemannian symmetric space as follows.
Theorem 4.2.2. Let (M, g) be a simply-connected Riemannian symmetric
space. Then M is a product
(M, g) ∼
= (M0 , g0 ) × (M1 , g1 ) × · · · × (Mn , gn ),
where (M0 , g0 ) is a Riemannian symmetric space of Euclidean type and for
i ≥ 1, the factors (Mi , gi ) are irreducible Riemannian symmetric spaces.
Proof. See Proposition 5.5 in Chapter VIII of [Hel78]. □
26

5. Duality
Definition 5.0.1 (complexification). Let g be a (real) Lie algebra. Then
its complexification gC = g ⊗ C is a complex Lie algebra, with Lie bracket
√ √ √
[X1 + −1Y1 , X2 + −1Y2 ] := [X1 , X2 ]−[Y1 , Y2 ]+ −1 ([Y1 , X2 ] + [X1 , Y2 ]) .

Suppose (g, s) is an orthogonal symmetric Lie algebra with Cartan de-

composition g = k ⊕ m. Then there are following bracketing relations:
(1) √ ⊆ k.
[k, k] √ √
(2) √ −1m]
[k, √ = −1[k, m] ⊆ −1m.
(3) [ −1m, −1m] = −[m, m] ⊆ k.
√
In particular, g∗ := k ⊕ −1m is a real Lie subalgebra of gC . Let sC be the
C-linear extension of s to gC and s∗ be the restriction of sC to g∗ . Then
(g∗ , s∗ ) is also an orthogonal symmetric Lie algebra, which is defined to be
the dual of (g, s).

Theorem 5.0.1. Let (g, s) be an orthogonal symmetric Lie algebra with

dual (g∗ , s∗ ).
(1) If (g, s) is of compact type, then (g∗ , s∗ ) is of non-compact type, and
vice versa.
(2) If (g, s) is of Euclidean type, then (g∗ , s∗ ) is of Euclidean type.
(3) (g, s) is irreducible if and only if (g∗ , s∗ ) is irreducible.

Proof. For (1) and (2). It suffices to establish a relation between the re-
spective Killing forms. Note that there √
is an isomorphism of vector spaces
∗
Ψ : g → g given by X + Y 7→ X + −1Y . For Z1 , Z2 ∈ m, a direct
computation shows
√ √ √ √ √ √ 
adg∗ ( −1Z1 ) adg∗ ( −1Z2 )(X + −1Y ) = −1Z1 , [ −1Z2 , X + −1Y ]
√
= − [Z1 , [Z2 , X]] − −1 [Z1 , [Z2 , Y ]]
= −Ψ([Z1 , [Z2 , X + Y ]])
= −Ψ(adg (Z1 ) adg (Z2 )(X + Y )).
√ √
Therefore Bg∗ ( −1Z1 , −1Z2 ) = −Bg (Z1 , Z2 ). As a consequence, Bg |m >
0 if and only if Bg∗ |√−1m < 0 and vice versa.
For (3). Note that g is semisimple if and only if its Killing form is non-
degenerate, so g is semisimple if and only if g∗ is, and thus (g, s) is irreducible
if and only if (g∗ , s∗ ) is irreducible. □

Example 5.0.1. Consider the orthogonal symmetric Lie algebra (sl(n, R), s),
where s : X 7→ −X T . Its Cartan decomposition is given by

k = {X ∈ sl(n, R) | X T + X = 0},
m = {X ∈ sl(n, R) | X T = X}.
 27

Then sl(n, R)C = sl(n, C) and

√ √
k + −1m = {Z ∈ sl(n, C) | Z = X + −1Y, X T + X = 0, Y T = Y }
T
= {Z ∈ sl(n, C) | Z + Z = 0}
= su(n).
As a consequence, the Riemannian symmetric space SL(n, R)/ SO(n) and
SU(n)/ SO(n) are dual to each other. For n = 2, one has H2 is dual to S 2 ,
since SU(2) is the universal covering of SO(3).
Example 5.0.2. Consider the orthogonal symmetric Lie algebra (so(n), s),
where s is given by
s : so(n) → so(n)
X 7→ Ik,l XIk,l
where k + l = n. Its Cartan decomposition is given by
so(n) = so(k) ⊕ so(l) ⊕ m,
where   
0 B
m= | B ∈ Mk×l (R) .
−B T 0
Then
 √  
√X1 −1B
g∗ = | X1 ∈ so(k), X2 ∈ so(l), B ∈ Mk×l (R) .
− −1B T X2
It’s easy to verify the mapping
 √   
√X1 T −1B X1 B
=
− −1B X2 BT X2
is a Lie algebra isomorphism of g∗ to so(p, q). This shows compact Grass-
mannian and hyperbolic Grassmannian are dual to each other.
28

6. Classifications of Riemannian symmetric space

6.1. Classifications of irreducible orthogonal symmetric Lie alge-
bra.
Theorem 6.1.1. Let (g, s) be an effective semisimple orthogonal symmetric
Lie algebra. Then it must be isomorphic to one of the following four cases:
CI g is a compact simple Lie algebra and s is an involution of g;
CII g = g1 ⊕ g2 , where g1 , g2 are compact simple Lie algebra and s(X, Y ) =
(Y, X);
NI g is a non-compact simple Lie algebra such that its complexification gC
is a complex simple Lie algebra, and s is an involution of g.
NII g is a non-compact simple Lie algebra such that its complexification gC
is not a complex simple Lie algebra, and s is an involution of g.
Moreover, CI and NI are dual to each other, while CII and NII are dual to
each other.
Theorem 6.1.2. Let (g, s) be an effective semisimple orthogonal symmetric
Lie algebra of non-compact type. Then Riemannian symmetric space arisen
from (g, s) is unique up to isometry, the center of (Iso(M, g))0 = {e} and M
is simply-connected.
6.2. Relations between different viewpoints. Along the lecture note,
we have encountered three categories listed as follows:
(1) Riemannian symmetric space.
(2) Riemannian symmetric pair.
(3) Orthogonal symmetric Lie algebra.
And the relations are shown in the following diagram
Theorem 1.2.1 Theorem 3.1.1

(1) Theorem 1.1.1

(2) Example 3.1.1
(3)

6.2.1. (1) and (2). Since any Riemannian symmetric space (M, g) gives a
Riemannian symmetric pair (G, K), where G = (Iso(M, g))0 and K = Gp
for some p ∈ G. In general, if a Riemannian symmetric pair (G, K) gives
the Riemannian symmetric space (M, g), G = (Iso(M, g))0 may be false.
Example 6.2.1. G = Rn and K = {0}, then σ : x 7→ −x makes (G, K) to
be a Riemannian symmetric pair, which gives the Riemannian symmetric
space Rn , but the identity component of the isometry group is larger than
G.
However, the following theorem shows it’s essentially the only exception.
Theorem 6.2.1. Let (G, K) be a Riemannian symmetric pair of semisimple
type and G acts effectively on M = G/K. Then G = (Iso(M, g))0 .
Proof. See Theorem 4.1 in Chapter V of [Hel78]. □
 29

6.2.2. (1) and (3). Given a Riemannian symmetric space (M, g), there is
a Riemannian symmetric pair (G, K), and thus we obtain an orthogonal
f = G/
symmetric Lie algebra (g, s). If we use M e K e to denote the Riemannian
symmetric space given by (g, s), a natural question is what’s the relationship
between M and M f? Since G e is simply-connected and has the same Lie
algebra as G, there exists a covering map p : G e → G. Moreover, since K e
e
and K also have the same Lie algebra, p(K) = K and thus p induces a
covering map p : Mf → M , which gives an isomorphism TpeM f∼ =m∼ = Tp M .
If we endow M f with Riemannian metric obtained from p, then p is a local
isometry, and thus p is a Riemannian covering since M f is complete. In
f.
particular, if M is simply-connected, then it’s isometric to M
Another thing is that we can show the correspondence between (1) and (3)
is bijective under some hypothesis, which can help us to classify irreducible
Riemannian symmetric space.
Theorem 6.2.2. Let (g, s) be a non-compact irreducible effective orthogo-
nal symmetric Lie algebra. Then Riemannian symmetric space (M, g) which
can give (g, s) is unique up to isometry. Moreover, the center of (Iso(M, g))0
is trivial and M is simply-connected.
6.3. Summary.
30

7. More properties of Non-compact type

Lemma 7.0.1. Let (G, K) be a non-compact Riemannian symmetric pair
given by σ. Then B ∗ (-, -) = −B(σ(-), -) on g has the following properties:
(1) B ∗ is positive definite.
(2) If X ∈ k, then ad(X) : g → g is skew-symmetric.
(3) If X ∈ m, then ad(X) : g → g is symmetric.
Proposition 7.0.1. Let (G, K) be a non-compact Riemannian symmetric
pair given by σ. Then
(1) G is non-compact and K is connected.
(2) K is a maximal compact subgroup of G.
(3) Z(G) ⊆ K.
(4) G is diffeomorphic to K × Rn , and G/K is diffeomorphic to Rn .
Proposition 7.0.2. Let G be a non-compact semisimple Lie group with
finite center. Then there exists a maximal compact subgroup K, unique up
to conjugacy, such that G is diffeomorphic to K × Rn . Moreover,
 31

Part 3. Hermitian symmetric space

8. Hermitian symmetric space
Definition 8.0.1 (Hermitian symmetric space). Let (M, g) be a Riemann-
ian symmetric space. (M, g) is said to be a Hermitian symmetric manifold
if (M, g) is a Hermitian manifold and the symmetric at each point is a holo-
morphic isometry.
Lemma 8.0.1. Any almost Hermitian structure on a Riemannian symmet-
ric space (M, g) is integrable, and any Hermitian symmetric space is Kähler.
Proof. Suppose φ is the symmetry at point p ∈ M and J is an almost
Hermitian structure of (M, g). Since φ is a holomorphic isometry one has
(dφ)p ◦ J = J ◦ (dφ)p , and thus
−NJ (X, Y ) = (dφ)p NJ (X, Y )
= (dφ)p ([JX, JY ] − J[JX, Y ] − J[X, JY ] − [X, Y ])
= [JX, JY ] − J[JX, Y ] − J[X, JY ] − [X, Y ]
= NJ (X, Y ).
This shows NJ = 0 at point p, and since p is arbitrary one has NJ ≡ 0,
which implies J is integrable. By the same argument one can show ∇J = 0,
and thus (M, g) is Kähler. □
Proposition 8.0.1. Let (G, K) be a symmetric pair with Cartan decom-
position g = k ⊕ m. If J : m → m satisfies
(1) J is orthogonal and J 2 = − id.
(2) J ◦ Ad(k) = Ad(k) ◦ J for all k ∈ K.
Then M = G/K is a Hermitian symmetric space, and thus Kähler.
Corollary 8.1. Let (G, K) be a symmetric pair. Then
(1) (G, K) is Hermitian symmetric if and only if its dual is Hermitian sym-
metric.
(2) If (G, K) is irreducible and Hermitian symmetric, then it’s Kähler-
Einstein.
Proposition 8.0.2. Let (G, K) be an irreducible symmetric pair.
(1) If (G, K) is of compact type, then it’s Hermitian symmetric if and only
if H 2 (M, R) 6= 0.
(2) (G, K) is Hermitian symmetric if and only if K is not semisimple.
(3) The complex structure J is unique up to a sign.
Proof. For (1). It’s clear if (G, K) is Hermitian symmetric, then H 2 (M, R) 6=
0 since its Kähler form lies in it; Conversely, for 0 6= ω ∈ H 2 (M, R), we may
construct a new 2-form ω e by
ˆ
ωep := ωgp dg.
G
e is invariant under isometries.
It’s clear ω □
32

9. Bounded symmetric domains

9.1. The Bergman metrics.
9.2. Classical bounded symmetric domains.
9.3. Curvatures of classical bounded symmetric domains.
 33

Part 4. Appendix
Appendix A. Remarks
A.1. Effectivity. In this section we try to explain the motivation of effec-
tivity of orthogonal symmetric Lie algebra.
Remark A.1.1. Let (G, K) be a Riemannian symmetric pair associated to
a Riemannian symmetric space (M, g). Note that G acts on M effectively,
and thus we claim K contains no non-zero subgroup of G. Otherwise if N is
a normal subgroup of G contained in K, it suffices to show for any n ∈ N ,
it fixes every point of M since G acts on M effectively. For any q ∈ M ,
suppose q = gp for some g ∈ G and hence
nq = ngp = g(g −1 ng)p = gp = q.
In particular, Z(G)∩K = {e}, and thus k∩z = 0. As a consequence, if (G, K)
is a Riemannian symmetric pair associated to a Riemannian symmetric space
and (g, s) is the orthogonal symmetric Lie algebra given by (G, K), then it’s
effective.
34

Appendix B. Lie group and Lie algebra

B.1. Fundamental theorems.
Theorem B.1.1. Every finite-dimensional (real) Lie algebra is the Lie al-
gebra of some simply-connected Lie group.
Theorem B.1.2. If G is a Lie group and h ⊆ Lie G is a Lie subalgebra,
then there exists a unique connected Lie subgroup H ⊆ G with Lie H = h.
Theorem B.1.3. If Φ : Lie G → Lie H is a Lie group homomorphism and
G is simply-connected, then there exists a unique Lie group homomorphism
φ : G → H such that Φ = (dφ)e .
Lemma B.1.1. Suppose G, H are connected Lie groups with Lie algebras
g, h and φ : G → H is a Lie group homomorphism. If (dφ)e : g → h is
bijective, then φ is a covering map.
Corollary B.1. If G,e G are connected Lie groups having isomorphic Lie
e e is the universal covering of G.
algebra and G is simply-connected, then G
Corollary B.2. If connected and simply-connected Lie groups G, H have
isomorphic Lie algebra, then G and H are isomorphic.
B.2. Adjoint action.
Definition B.2.1 (compactly embedded). Let g be a Lie algebra. A subal-
gebra k ≤ g is compactly embedded if ad(k) is the Lie algebra of a compact
subgroup of GL(g).
B.3. Semisimple Lie algebras.
Definition B.3.1 (semisimple). A Lie algebra g is called semisimple if the
Killing form B of g is non-degenerate.
Definition B.3.2 (simple). A Lie algebra g is called simple if it’s semisimple
and has no ideals except {0} and g.
Definition B.3.3. A Lie group is called semisimple (simple) if its Lie alge-
bra is semisimple (simple).
Proposition B.3.1. A semisimple Lie algebra has center {0}.
Proposition B.3.2. A semisimple Lie algebra g is the direct sum
g = g1 ⊕ . . . gr .
where gi (1 ≤ i ≤ r) are all the simple ideals in g. Every ideal a of g is the
direct sum of certain gi .
Proposition B.3.3.
(1) Let g be a semisimple Lie algebra. Then g is compact if and only if the
Killing form of g is negative definite.
(2) Every compact Lie algebra g is a direct sum g = z ⊕ [g, g], where z is the
center of g and the ideal [g, g] is semisimple and compact.
 35

Appendix C. Basic facts in Riemannian geometry

C.1. Killing fields.
C.1.1. Basic properties.
Proposition C.1.1. Let (M, g) be a Riemannian manifold and X be a
Killing field.
(1) If γ is a geodesic, then J(t) = X(γ(t)) is a Jacobi field.
(2) For any two vector fields Y, Z,
∇Y ∇Z X − ∇∇Y Z X + R(X, Y )Z = 0
Proof. For (1). Suppose φs is the flow generated by X. Then we obtain a
variation α(s, t) = φs (γ(t)) consisting of geodesics, and thus
∂φs (γ(t))
X(γ(t)) =
∂s s=0
is a Jacobi field.
For (2). It’s an equation of tensors, so we check it pointwisely and use
normal coordinate {xi } centered at p. Moreover, we assume X = X i ∂x∂
i,Y =
∂ ∂
∂xj
, Z = ∂xk
. Then
∂
∇Y ∇Z X − ∇∇Y Z X + R(X, Y )Z = ∇j ∇k X + X i Rijk
l
∂xl
l
∂2X l i ∂Γki ∂
=( j k
+ X j
+ X i Rijk
l
) l
∂x ∂x ∂x ∂x
l
∂2X l ∂Γ jk ∂
= ( j k + Xi )
∂x ∂x ∂xi ∂xl
∂Γljk ∂Γlik ∂2X l
l
since Rijk = ∂xi
− ∂xj
+ Γsjk Γlis − Γsik Γljs . Now it suffices to show ∂xj ∂xk
+
∂Γl
≡ 0. In order to show this, for arbitrary p ∈ M , consider a geodesic
X i ∂xjki
γ starting at p and consider Jacobi field J(t) = X(γ(t)). Direct computation
shows
∂X i dγ k dγ k ∂
J ′ (t) = ( k + X i Γlki )
∂x dt dt ∂xl γ(t)
l
∂ 2 X l dγ j dγ k j
i ∂Γki dγ dγ
k ∂
J ′′ (0) = ( j k
+ X j
) l
∂x ∂x dt dt ∂x dt dt ∂x p
∂Γl dγ j dγ k ∂
∂2X l
= ( j k + X i ki )
∂x ∂x ∂xj dt dt ∂xl p

∂2X l ∂Γljk ∂Γl ∂Γljk dγ j dγ k ∂

=( + Xi + X i ki − Xi )
∂xj ∂xk ∂xi ∂xj ∂xi dt dt ∂xl
p

∂2X l ∂Γljk dγ j dγ k ∂
=( + Xi ) − R(X, γ ′ )γ ′
∂xj ∂xk ∂xi dt dt ∂xl
p
36

which implies
l
∂2X l i
∂Γjk
+ X =0
∂xj ∂xk ∂xi
holds at point p, and since p is arbitrary, this completes the proof. □
Corollary C.1. Let (M, g) be a complete Riemannian manifold and p ∈
M . Then a Killing field X is determined by the values Xp and (∇X)p for
arbitrary p ∈ M .
Proof. The equation LX g ≡ 0 is linear in X, so the space of Killing fields is
a vector space. Therefore, it suffices to show if Xp = 0 and (∇X)p = 0, then
X ≡ 0. For arbitrary q ∈ M , let γ : [0, 1] → M be a geodesic connecting p
and q with γ ′ (0) = v. Since J(t) = X(γ(t)) is a Jacobi field, and a direct
computation shows
(∇v X)p = J ′ (0)
Thus J(t) ≡ 0, since Jacobi field is determined by two initial values. In
particular, Xq = J(1) = 0, and since q is arbitrary, one has X ≡ 0. □
Corollary C.2. The dimension of vector space consisting of Killing fields
≤ n(n + 1)/2.
Proof. Note that ∇X is skew-symmetric and the dimension of skew-symmetric
matrices is n(n − 1)/2. Thus the dimension of vector space consisting of
Killing fields ≤ n + n(n − 1)/2 = n(n + 1)/2. □
C.1.2. Killing field as the Lie algebra of isometry group.
Lemma C.1.1. Killing field on a complete Riemannian manifold (M, g) is
complete.
Proof. For a Killing field X, we need to show the flow φt : M → M generated
by X is defined for t ∈ R. Otherwise, we assume φt is defined on (a, b). Note
that for each p ∈ M , curve φt (p) is a curve defined on (a, b) having finite
constant speed, since φt is isometry. Then we have φt (p) can be extended
to the one defined on R, since M is complete. □
Theorem C.1.1. Let (M, g) be a complete Riemannian manifold and g
the space of Killing fields. Then g is isomorphic to the Lie algebra of G =
Iso(M, g).
Proof. It’s clear g is a Lie algebra since [LX , LY ] = L[X,Y ] . Now let’s see it’s
isomorphic to Lie algebra consisting of Killing field as Lie algebra.
(1) Given a Killing field X, by Lemma C.1.1, one deduces that the flow
φ : R ×M → M generated by X is a one parameter subgroup γ : R → G,
and γ ′ (0) ∈ Te G.
(2) Given v ∈ Te G, consider the one-parameter subgroup γ(t) = exp(tv) : R →
G which gives a flow by
φ : R ×M → M
(t, p) 7→ exp(tv) · p
 37

Then the vector field X generated by this flow is a Killing field.

This gives a one to one correspondence between Killing fields and Lie algebra
of G, and it’s a Lie algebra isomorphism. □

Corollary C.3 (Cartan decomposition). Let (M, g) be a complete Rie-

mannian manifold and G = Iso(M, g) with Lie algebra g. The Lie algebra g
of G has a decomposition as vector spaces

g=k⊕m

where
k = {X ∈ g | Xp = 0}
m = {X ∈ g | (∇X)p = 0}
and they satisfy
[k, k] ⊆ k, [m, m] ⊆ k, [k, m] ⊆ m

Proof. The decomposition follows from Corollary C.1 and Theorem C.1.1,
and it’s easy to see
[k, k] ⊆ k, [m, m] ⊆ k
For arbitrary X ∈ k, Y ∈ m and v ∈ Tp M , one has
∇v [X, Y ] = ∇v ∇X Y − ∇v ∇Y X
= −R(Y, v)X + ∇∇v X Y + R(X, v)Y − ∇∇v Y X
=0

since Xp = 0 and (∇Y )p = 0. This shows [k, m] ⊆ m. □

C.2. Hopf theorem. The argument about analytic continuation in Theo-

rem 0.4.1 can be used to give a proof of Hopf’s theorem.

Theorem C.2.1 (Hopf). Let (M, g) be a complete, simply-connected Rie-

mannian manifold with constant sectional curvature K. Then (M, g) is
isometric to
 n 1
 √
(S ( K ), gcan ) K>0
f n
(M , gcan ) = (R , gcan ) K=0

 n √1
(H ( −K ), gcan ) K < 0

Proof. For p ∈ M, pe ∈ M f and δ < min{inj(p), inj(ep)}. By Cartan-Ambrose-

Hicks’s theorem, there exists an isometry φ : B(p, δ) → B(e p, δ) such that
φ(p) = pe and (dφ)p equals to a given linear isometry, since both (M, g) and
f, ge) have constant sectional curvature K. By the same argument in proof
(M
of Theorem 0.4.1, there is an isometry φ : (M, g) → (M f, ge) which extends
φ : B(p, δ) → B(ep, δ). In particular, this completes the proof. □
38

C.3. Other basic facts.

Theorem C.3.1. Let φ, ψ : (M, gM ) → (N, gN ) be two local isometries
between Riemannian manifolds, and M is connected. If there exists p ∈ M
such that
φ(p) = ψ(p)
(dφ)p = (dψ)p
then φ = ψ.
Theorem C.3.2 (Cartan-Ambrose-Hicks). Let (M, g) and (M f, ge) be two
Riemannian manifolds and Φ0 : Tp M → TpeM f be a linear isometry, where p ∈
f. For 0 < δ < min{injp (M ), injpe(M
M, pe ∈ M f)}, The following statements
are equivalent.
(1) There exists an isometry φ : B(p, δ) → B(e p, δ) such that φ(p) = pe and
(dφ)p = Φ0 .
(2) For v ∈ Tp M, |v| < δ, γ(t) = expp (tv), γ
e(t) = exppe(tΦ0 (v)), if we define
f
Φt = P0,t;eγ ◦ Φ0 ◦ Pt,0;γ : Tγ(t) M → Tγe(t) M
then Φt preserves curvature, that is (Φt )∗ R = R.
Proposition C.3.1. Let (M, gM ), (N, gN ) be complete Riemannian mani-
folds and f : M → N be a local diffeomorphism such that for all p ∈ M and
for all v ∈ Tp M , one has |(df )p v| ≥ |v|. Then f is a Riemannian covering
map.
Theorem C.3.3 (Myers-Steenrod). Let (M, g) be a Riemannian manifold
and G = Iso(M, g). Then
(1) G is a Lie group with respect to compact-open topology.
(2) for each p ∈ M , the isotropy group Gp is compact.
(3) G is compact if M is compact.
Proposition C.3.2. Let (M, g) be a Riemannian manifold, γ : I → M a
smooth curve and Ps,t;γ : Tγ(s) M → Tγ(t) M is the parallel transport along
γ. For any s ∈ I with v = γ ′ (s) and tensor T , one has
d
∇v T = (Ps,t;γ )∗ Tγ(t)
dt t=s
In particular, if ∇T = 0 then
(Ps,t;γ )∗ Tγ(t) = Tγ(s)
holds for arbitrary t, s ∈ I.
Proposition C.3.3. If π : (Mf, ge) → (M, g) is a Riemannian covering, then
f is.
M is complete if and only if M
 39

References
[Hel78] Sigurdur Helgason. Differential geometry, Lie groups, and symmetric spaces, vol-
ume 80 of Pure and Applied Mathematics. Academic Press, Inc. [Harcourt Brace
Jovanovich, Publishers], New York-London, 1978.