Geodesic completeness of the left-invariant metrics on RH n

Sržan Vukmirovi¢, Tijana ’ukilovi¢

January 14, 2020

Abstract
In this paper we give the full classication of left-invariant metrics of an arbitrary signature on
the Lie group corresponding to the real hyperbolic space. We show that all metrics have constant
sectional curvature and that they are geodesically complete only in the Riemannian case.
It is a well known fact that the real hyperbolic space RH n with the standard Riemannian metric of
constant negative curvature has a structure of a Lie group such that the metric is left-invariant. Milnor
[8] considered a special class of solvable Lie groups with the property that the commutator [x, y] is a
linear combination of x and y for any two elements from the corresponding Lie algebra. Moreover, he
has shown that such an algebra is isomorphic to the Lie algebra of RH n and that every left-invariant
Riemannian metric on such group has a constant negative sectional curvature. In the Lorentz case, Wolf
[13] showed that this group admits a at metric, while Nomizu [9] proved that for an arbitrary K ∈R
there exists a left-invariant metric with sectional curvature K. In [10] Nomizu considered in detail these
metrics that he called Lorentz-Poincaré metrics.
In this paper we classify all left-invariant metrics of arbitrary signature on the Lie group corresponding
to the hyperbolic space. This problem has also been considered in [6]. According to Arnold [1], geodesic
of an arbitrary left invariant metric on a Lie group G can be seen as a motion of a generalized rigid body
with a conguration space G. In the Riemannian case all geodesic are complete, but Gudeiri [4] gave an
example of the Lorentzian metric on four dimensional Lie group with non-complete geodesics. Lauret [7]
has classied all Riemannian left-invariant metrics on the four dimensional nilpotent Lie groups, while the
authors have generalized that result to an arbitrary signature [2]. Calvaruso [3] has classied Lorentzian
left-invariant metrics on the four dimensional Lie groups that are Einstein or Ricci-parallel. In [11] the
Riemannian and Lorentzian left-invariant metrics on the Heisenberg Lie group were classied.
In the preliminary section we introduce a basic notation and give an explanation what does it mean
to classify the left-invariant metrics.
In the Theorem 2.2 of Section 2 we present classication of the arbitrary signature left-invariant
metrics on the Lie group RH n . We show that in the Riemannian case the only metrics are the standard
metrics of constant curvature K < 0 of hyperbolic space. In the Lorentz case, every metric from our
classication is isometric either to the at metric obtained by Wolf, or to the metric of constant curvature
K 6= 0 obtained by Nomizu. Also, in the Theorem 2.3 we prove that every non-at metric on RH n is a
metric of the constant sectional curvature.
The geodesical completeness of metrics is considered in Section 3. We show that all geodesic curves
are complete only in the Riemannian case (Theorems 3.1 and 3.2).
In Section 4 we exhibit isometric imbedding of RH n into the space forms of the curvature K.
2010 Mathematics Subject Classication. Primary 53B30; Secondary 22E25

Key words and phrases : left-invariant metrics, geodesically complete metrics, hyperbolic space.

1
1 Preliminaries

The group structure on a half-space model of a real hyperbolic space RH n = { (x1 , . . . , xn ) ∈ Rn | xn > 0}
is given by:

(x1 , . . . , xn−1 , xn ) · (y1 , . . . , yn−1 , yn ) = (x1 + xn y1 , . . . , xn−1 + xn yn−1 , xn yn ) . (1)

Denote by {e1 , . . . , en } a corresponding basis of the Lie algebra rn = Lie RH n . It is a semi-direct

product Ren n n, where n = L(e1 , . . . , en−1 ) is an abelian ideal and ad(en )|n = id, i.e. the non-null
commutators are:

[en , ek ] = ek , k < n.

Denote by S(rn ) a set of non-equivalent inner products of an algebra rn . With a basis of the Lie
algebra rn xed, the set S(rn ) is identied with symmetric matrices S of an arbitrary signature modulo
the following action of the automorphism group:

S 7→ F T SF, F ∈ Aut(rn ). (2)

Here we denoted by Aut(rn ) the group of automorphisms of the Lie algebra rn , that is dened by:

Aut(rn ) := {F : rn → rn | F linear, bijective, [F x, F y] = F [x, y], x, y ∈ rn }.

It is easy to check the following lemma.

Lemma 1.1. The group Aut(rn ) of automorphisms of Lie algebra rn , in basis {e1 , e2 , . . . , en }, consists
of real matrices of form:
  
A a  A ∈ GL(n − 1, R), a ∈ Rn−1 ∼

Aut(rn ) = = Aff n−1 (R), (3)
0 1 

i.e. it is isomorphic to a group of ane transformations of Rn−1 .

Let x ∈ RH n with x = (x1 , . . . , xn ). For the left translations Lx the dierential dLx in every point
y ∈ RH n is given by:

 
∂ ∂
(dLx ) (y) = xn Lx (y),
∂xk ∂xk

Therefore the left-invariant vector elds X1L , . . . , XnL are given by:

∂
XkL (x) = xn , k ≤ n. (4)
∂xk

2 Left-invariant metrics

Let us denote by Ip,r the diagonal matrix diag(1 , 2 , . . . , p+r ), where k = −1, 1 ≤ k ≤ p and k = 1 ,
p + 1 ≤ k ≤ p + r.
Theorem 2.1. The set S(rn ) of non-equivalent inner products of an arbitrary signature on the algebra
rn is represented by the following matrices:
 
  0 0 1
Ip,r 0
Sλ = , p + r = n − 1, λ 6= 0, S0 =  0 Ip,r 0  , p + r = n − 2.
0 λ
1 0 0

2
Proof. F ∈ Aut(rn ) and denote by S̄ the arbitrary symmetric matrix representing the inner product
Let
q {e1 , . . . , en }. In the same basis F is represented by the matrix (3). We are looking for the
in the basis
T
new basis such that F S̄F has the simplest form.
If we represent S̄ in the following form:

 
S v
S̄ = ,
vT s

where S = ST is (n − 1) × (n − 1) matrix , v ∈ Rn−1 and s ∈ R, then:

 
T AT SA AT (Sa + v)
F S̄F = .
(a S + v T )A
T
a Sa + v T a + aT v + s
T

Now, we distinguish between two cases.

Case 1. S is a regular matrix of signature (p, r). Since S is symmetric there exists A ∈ GL(n − 1, R)
such that AT SA = Ip,r , p + r = n − 1. Setting a = −S −1 v, we get that the corresponding inner product
−1
with λ = s − v S v . Since S̄ is a non-singular matrix, λ 6= 0 must hold.
T
is Sλ ,
Case 2. If S is not regular, without loss of generality, we can assume that S has the form:

 
0 0
,
0 S̃

where S̃ is a regular matrix of signature (p, r), p + r = n − 2. Then there exists a regular matrix A
e such
that:
 1   
w 0 T 0 0
A= and A SA = .
0 A
e 0 Ip,r

For the vector v = (w, v̄ T )T , w 6= 0, v̄ ∈ Rn−2 , we set a = (a1 , āT )T :

v̄ T S̃ −1 v̄ − s
a1 = ∈ R, ā = −S̃ −1 v̄ ∈ Rn−2 ,
2w
to obtain the inner product S0 .
The inner product q on rn gives rise to a left-invariant metric g on the corresponding Lie group. For
global coordinates (x1 , . . . , xn ) on RH n , using the left-invariant vector elds (4), we can nd a coordinate
description of metrics dened in the previous theorem.

Theorem 2.2. Each left-invariant metric on the group RH n , up to an automorphism of RH n , is iso-

metric to one of the following:
1
−dx21 − . . . − dx2p + dx2p+1 + . . . + dx2n−1 + λdx2n , λ 6= 0,


gλ = 2
xn
1
g0 = 2 −dx22 − . . . − dx2p+1 + dx2p+2 + . . . + dx2n−1 + 2dx1 dxn .


xn

Wolf [13] showed that RH n admits at metrics and, in the four dimensional case, according to
classication of Jensen [5], we know that it also admits Einstein metrics. Later, Milnor [8] has shown
that every left-invariant positive denite metric on RH n has a negative sectional curvature, while Nomizu
[9] proved that for every K ∈ R one can nd the left-invariant Lorentz metric on RH n with K as constant
sectional curvature. Yet, we are able to prove more.

Theorem 2.3. All left-invariant metrics of an arbitrary signature on RH n have a constant sectional
curvature.

3
Proof. In order to prove the theorem we need to calculate the curvature tensor. We use the identication
of the left-invariant vector elds XkL XkL (e) = ek .
with their value in the unit element
Recall that the curvature operators R(ei , ej ) belong to the algebra so(q) preserving the inner product
q, i.e.

so(q) := { A ∈ gl(rn ) | AS + SAT = 0} ,

where S denotes the matrix of q. This algebra can be identied with the space Λ 2 rn of bivectors, whose
action on rn is given by:

(x ∧ y)z := q(y, z)x − q(x, z)y, x, y, z ∈ rn .

Using standard calculations, for the metric gλ , we get that the connection is given by the non-zero
expressions:

i
∇ei e i = en , i ∈ {−1, 1}, ∇ei en = −ei , i < n, (5)
λ
and the curvature operators are given by:

1
R(ei , ej ) = − ei ∧ ej .
λ
From the previous is apparent that the sectional curvature is constant K = − λ1 .
For metric g0 all components of curvature tensor R vanish, thus the metric is at and K = 0.

3 Geodesics

Every C1 curve c(t) on the Lie group G, up to the left translations, gives rise to the curve:

γ(t) = L−1
c(t)∗ ċ(t) (6)

on the corresponding Lie algebra g. The curves of g associated to geodesics are solutions of the equations:
ẋ = ad∗x x, (7)

∗
where adx stands for the adjoint of adx relative to the inner product on g.
n
First, let us consider the Einstein metric gλ on RH . Fixing the basis {e1 , . . . , en } ∈ rn , an easy
computation gives us:

k
ad∗ek ek = − en , ad∗en ek = ek , k < n.
λ
Pn
In local coordinates (x1 , . . . , xn ), for γ(t) = xk (t)ek , from the equation (7) we obtain the system:
k=1

n−1
1X
ẋk = xk xn , k < n, ẋn = − j x2j . (8)
λ j=1

n−1
2
= − λ1 k Ck2 .
P
Let C1 , . . . , Cn ∈ C and let us denote by Cn+1 Then the solutions γ(t) = (x1 (t), . . . , xn (t))
k=1
of the system (8) are given by:

Ck 2
xk (t) = , k < n, xn (t) = Cn+1 tan(Cn+1 t + Cn ), for Cn+1 6= 0, 1, (9)
cos(Cn+1 t + Cn )
Ck 1 2
xk (t) = , k < n, xn (t) = − , for Cn+1 = 1, (10)
t + Cn t + Cn
xk (t) = Ck eCn t , k < n, xn (t) = Cn , for Cn+1 = 0. (11)

4
Note that constants C1 , . . . , C n must be real in case of the solutions (10) and (11). For the solution
(9) they can be either real or complex, but they need to satisfy the additional constraints which will be
explained in detail in the proof of the following theorem.

Theorem 3.1. The left-invariant metric gλ on RH n is geodesically complete if and only if it is positive
denite.
Proof. First, note that because of the left-invariance we may consider only the curves γ(t) in the Lie
algebra rn dened by (6).
A geodesic curve whose tangent vector is en corresponds to the solution (11) with Ck = 0, k < n, and
Cn = 1 . Those are the vertical lines ending on the hyperplane xn = 0 and they are complete in every
signature.
Let v = γ(0) 6= en be a tangent vector of a geodesic curve. Notice that from (5) follows that the
two dimensional plane α = L(v, en ) is totally geodesic. Therefore it is enough to discuss the induced
signature in that plane.
|v|2 = −λCn+1
It is not dicult to calculate that for the solution (9)
2 2
, for (10) |v| = 0 and in the last
2
2
case (11) |v| = λCn . If the plane α is non-degenerate then we consider solutions (9) and (10), while the
solution (11) occurs only if α is degenerate.
2
Case 1. Suppose that the plane α is Riemannian. Then Cn+1 < 0, i.e. Cn+1 = iD, D ∈ R. In order to
determine the constant Cn we must consider the Gram determinant associated to the plane α. One can
calculate that

G = −λ2 Cn+1
2
(1 + tan2 Cn ). (12)

In Riemannian case G Cn = iC, C ∈ R. Note that in order to obtain

must be positive, which yields
the real solutions, all the other constants Ck , k < n, must be real. The trigonometric functions in (9)
become hyperbolic functions cosh and − tanh, so the curves are complete. It is easy to check that the
n
corresponding geodesic curves on the Lie group RH are half-ellipses with centers on the hyperplane
xn = 0. The plane α is isometric to the standard hyperbolic plane.
Case 2. Suppose that the plane α is Lorentzian.
2
If |v| 6= 0, then the solution is given by (9) and we distinguish between two cases.
2
When the vectors v and en are of the same character, i.e. of the same signature, then Cn+1 < 0.
Therefore we must take Cn+1 = iD, D ∈ R. Here, the Gram determinant (12) must be negative, thus
Cn = iC + π2 , C ∈ R. Also, all the constants Ck , k < n, must be purely imaginary. The trigonometric
functions in (9) become hyperbolic functions sinh and − coth.
2
When the vectors v and en are of the dierent character, then Cn+1 > 0 and we have the solution (9)
where all the constants are real. In both cases the geodesics are incomplete. The corresponding geodesic
curves on RH n are branches of hyperbolas satisfying xn > 0.
2
If |v| = 0, then the corresponding curves are given by (10). In RH n those are the straight lines
ending on the hyperplane xn = 0 and they are geodesically incomplete.
Case 3. Suppose that the plane α is degenerate. The vector u = v − Cn en is a null vector orthogonal
to all vectors from α. The corresponding solutions are complete geodesics given by (11). These are
parabolas on RH n .
We can conclude that the metric is complete if and only if for every tangent vector v the corresponding
plane α is Riemannian.

Similarly, for the at metric g0 , in local coordinates, we have the system:

n−1
X
ẋ1 = −x1 xn − j x2j , ẋk = xk xn , 1 < k ≤ n.
j=2

The solutions to the system above are given by:

C0 Ck 1
x1 (t) = C1 (t + Cn ) + , xk (t) = , 2 ≤ k < n, xn (t) = − ,
2(t + Cn ) t + Cn t + Cn
x1 (t) = C1 − tC0 , xk (t) = Ck , 2 ≤ k < n, xn (t) = 0,

5
 n−1
k Ck2 and Ck ∈ R, k ≤ n.
P
with C0 =
k=2
Consequently, the following theorem holds.

Theorem 3.2. The pseudo-Riemannian metric g0 on the Lie group RH n is geodesically incomplete.

4 Isometric imbedding into the space forms

p n
Rnp Rn
P P
Denote by the space with the pseudo-Riemannian metric g(X, Y ) = − xk yk + xk yk for
k=1 k=p+1
every X, Y ∈ Rn .
Let Spn ⊆ Rn+1
p be the de Sitter space:

Spn = u = (u0 , . . . , un )| − u20 − . . . − u2p−1 + u2p + . . . + u2n = −λ, λ < 0 .



This is the hypersurface in Rn+1

p with its induced metric of signature (p, n − p) of constant sectional
−1
curvature K = −λ > 0.
n n+1
Similarly, denote by Hp ⊆ Rp+1 the anti-de Sitter space:

Hpn = u = (u0 , . . . , un )| − u20 − . . . − u2p + u2p+1 + . . . + u2n = −λ, λ > 0



with its induced metric of signature (p, n − p) and constant sectional curvature K = −λ−1 < 0.
n
Dene S̃p and H˜pn to be the respective connected, simply connected manifolds corresponding to Spn
n
and Hp .
According to Wolf [12] every complete connected pseudo-Riemannian manifold of signature (p, n − p)
and constant sectional curvature K has an universal pseudo-Riemannian covering S̃pn if K > 0, H˜pn if
K <0 and Rnp if K = 0. Our metrics gλ and g0 have constant sectional curvature and, although they
are not always complete, we are interested in nding a local isometry into the space forms.

Theorem 4.1. (RH n , gλ ) of signature (p, n − p) is isometric to the part of Spn (if λ < 0) and Hpn (if
λ > 0) determined by the condition u0 + un > 0.
Proof. Suppose that λ < 0. Then the metric gλ has the form:

n−1
!
1 1 X
gλ = 2 −dx21 − . . . − dx2p−1 + dx2p + . . . + dx2n−1 + λdx2n = 2 k x2k + λx2n


, λ < 0,
xn xn
k=1

We dene an isometric imbedding f : RH n → Spn by:

f (x) = f (x1 , . . . , xn ) = (u0 , u1 , . . . , un ) = u,

where:
n−1  n−1 
k x2k + λx2n k x2k + λx2n
P P
1+ 1−
k=1 xk k=1
u0 = , uk = , 1 ≤ k < n, un = .
2xn xn 2xn
The image f (RH n ) is an open submanifold:

u = (u0 , . . . , un ) ∈ Spn  u0 + un > 0

 

The proof of the case when λ>0 is similar, only replacing Spn with Hpn .

6
Remark 4.1. The previous theorem has been proven by Nomizu [9] in the Lorentz case. Following the
reasoning from the same paper, one can show that there exists an isomorphism h : RH n → SO+ (p, n − p)
n
such that the mapping f is equivariant, meaning that the following diagram commutes for every g ∈ RH :

Lg
RH n −−−−→ RH n
 
f f
y y
h(g)
Spn −−−−→ Spn

(the same holds if we replace Spn with Hpn ).

Remark 4.2. Note that a geodesic curve c(t) in RH n is incomplete if and only if f (c(t)) reaches the
boundary u0 + un = 0 for a nite value of the ane parameter t. In the Figure 1 we illustrate this with
the example of geodesics on RH 2 .

Figure 1: Geodesics on RH 2 : Riemannian case (left), Lorentz case (right)

Theorem 4.2. (RH n , g0 ) of signature (p, n−p) is isometric to the part of Rnp determined by the condition
y1 + yn > 0.
Proof. We can dene the following change of coordinates, i.e. the map from RH n ⊂ Rnp to Rnp :
 n−1
  n−1

k x2k 2
P P
1 + 2x1 xn + 1 − 2x1 xn + k xk
k=2 xk k=2
y1 = , yk = , 1 < k < n, yn = .
2xn xn 2xn
In this new coordinates, the metric g0 has the form:

n−1
X
g00 = −dy12 + k dyk2 + dyn2
k=2

This is a part of an open half-space of the at space form satisfying the relation y1 + yn > 0.

Acknowledgment

The authors would like to thank the anonymous referee for valuable comments and suggestions.
This research was supported by the Serbian Ministry of Science Project 174012, Geometry, Education
and Visualization with Applications.

7
References

[1] V. I. Arnol'd: Mathematical methods of classical mechanics. Springer Science & Business Media;
2013.

[2] N. Bokan, T. ’ukilovi¢, S. Vukmirovi¢: Lorentz geometry of 4-dimensional nilpotent Lie groups.
Geometriae Dedicata. 2015; 177(1) : 83-102.

[3] G. Calvaruso, A. Zaeim: Four-dimensional Lorentzian Lie groups. Dierential Geometry and its
Applications. 2013; 31(4): 496-509.

[4] M. Guediri: Sur la complétude des pseudo-métriques invariantes à gauche sur les groupes de Lie
nilpotents. Rendiconti del Seminario Matematico Università e Politecnico di Torino. 1994; 52: 371-
376.

[5] G.R. Jensen: Homogeneous Einstein spaces of dimension four. Journal of Dierential Geometry.
1969; 3(3-4): 309-49.

[6] A. Kubo, K. Onda, Y. Taketomi, H. Tamaru: On the moduli spaces of left-invariant pseudo-
Riemannian metrics on Lie groups. Hiroshima Mathematical Journal. 2016; 46(3): 357-74.
[7] J. Lauret: Homogeneous nilmanifolds of dimension 3 and 4. Geometriae Dedicata. 1997; 68: 145-155.

[8] J. Milnor: Curvatures of left-invariant metrics on Lie groups. Advances in mathematics. 1976; 21(3):
293-329.

[9] K. Nomizu: Left-invariant Lorentz metrics on Lie groups. Osaka Journal of Mathematics. 1979;
16(1): 143-50.

[10] K. Nomizu: The Lorentz-Poincaré metric on the upper half-space and its extension. Hokkaido Math-
ematical Journal. 1982; 11: 253-61.

[11] S. Vukmirovi¢: Classication of left-invariant metrics on the Heisenberg group. Journal of Geometry
and Physics. 2015; 94: 72-80.

[12] J.A. Wolf: Homogeneous manifolds of constant curvature. Commentarii Mathematici Helvetici. 1962;
36(1): 112-47.

[13] J.A. Wolf: Isotropic manifolds of indenite metric. Commentarii Mathematici Helvetici. 1964; 39(1):
21-64.

Sržan Vukmirovi¢ Tijana ’ukilovi¢

University of Belgrade, University of Belgrade,
Faculty of Mathematics, Faculty of Mathematics,
Studentski trg 16, Studentski trg 16,
11000 Belgrade, SERBIA 11000 Belgrade, SERBIA

E-mail : vsrdjan@matf.bg.ac.rs E-mail : tijana@matf.bg.ac.rs