Вы находитесь на странице: 1из 22

Notes on the structure of the Lorentz group

(preliminary and unfinished version)


Claudio Bartocci

Contents
1 Generalities

2 The polar decomposition theorem


for the Lorentz group

3 An interlude: the case of O(1, 1)

4 The topological structure of O(p, 1)

10

5 Hyperbolic geometry

14

6 The Poincar disc

17

2014 Claudio Bartocci

The structure of the Lorentz group

1 Generalities
Let (V, g) be a pseudo-Euclidean vector space of signature (p, q); the dimension of
V is n = p + q. As we proved in [2, 8], after suitably choosing an orthonormal basis,
the matrix representing the pseudo-Euclidean product g is a diagonal matrix of the
form:
!

0
Ip
p,q =
0 I q
where I r , r = p, q, is the r r identity matrix.
The group of of g-isometries of V is isomorphic to the matrix group
O(p, q) = { GL(p + q; R) | T p,q = p,q } .
As customary, we denote by SO(p, q) the subgroup of O(p, q) consisting of matrices
whose determinant is equal to 1. When q = 0, we recover the familiar definition of
the orthogonal group:
O(n, 0) O(n) = { GL(n; R) | T I n = I n } .

It is immediate to see that, if O(p, q), then T O(p, q) as well; moreover, since
p,q p,q = I p+q , the inverse of is
1 = p,q T p,q .

(1.1)

When R O(n), the previous formula reduces to the usual one, i.e. R 1 = R T .
From the fact that p,q = q,p , it follows that the groups O(p, q) and O(q, p) are
isomorphic for any p, q; in particular, O(n) ' O(0, n). We shall make use of the standard immersions O(p, q) , O(p +1, q) and O(p, q) , O(p, q +1), which are defined,
respectively, by the assignments

1 0
O(p, q) 7
O(p + 1, q)
0
and
O(p, q) 7

In particular, when R O(p), we have

O(p, q + 1) .

O(p, q + 1).

Theorem 1.1. The matrix group O(p, q) has a natural structure of closed Lie subgroup
of GL(p + q; R); its dimension is
dimO(p, q) =

(p + q)(p + q 1)
.
2

The standard immersions O(p, q) , O(p +1, q) and O(p, q) , O(p, q +1) are embeddings.

Section 2

Proof. Let p+q be the vector space of real (p + q) (p + q) symmetric matrix; its
dimension is

(p+q)(p+q+1)
.
2

Let us introduce the map


H : GL(p + q; R) p+q

defined by H () = p,q . Then O(p, q) = H 1 ( p,q ). The map H is smooth and

its differential at the point is given by ( )

d
dH () =
H ( exp(t )) = T T p,q + T p,q
dt t =0
for any gl(p + q). Now, if satisfies the relation T p,q = p,q , we get
dH () = T p,q + p,q .
This shows that the differential dH : gl(p + q) p+q is surjective: indeed, given
any symmetric matrix S, it suffices to take = (S p,q )/2. By [6, Theorem 1.38], it
follows that O(p, q) = H 1 ( p,q ) has a unique structure of embedded submanifold
of GL(p + q; R) whose dimension is
(p + q)2

(p + q)(p + q + 1) (p + q)(p + q 1)
=
.
2
2

It is easy to conclude (see e.g. [6, Theorem 3.34]) that O(p, q) is a (closed) Lie subgroup of GL(p + q; R). The last claim is straightforward.
N

2 The polar decomposition theorem


for the Lorentz group
From now on, we restrict our attention to the (generalized) Lorentz group O(p, 1).
In order to make notation less cumbersome, the matrix p,1 will be indicated simply by ; vectors in Rp+1 will be denoted by U, V , X , Y , ..., and they will be thought
as column vectors; so, e.g., we shall write X = (X 1 , . . . , X p+1 )T . We fix the immersion Rp , Rp+1 defined by (x 1 , . . . , x p ) 7 (x 1 , . . . , x p , 0); column vectors in Rp will
be denoted by b,
c,
..., u, v, x, y, ...; hence, a vector X can be written also in the
x
block form X =
, with R. The Lorentz group O(p, 1) is the isometry group of

the Minkowski product (defined by ) on Rp+1 , namely the pseudo-Euclidean inner


product defined by
X M Y =

p+1
X
i , j =1

i j X i Y j =

p
X

X i Yi X p+1 Y p+1 .

i =1

p
The usual Euclidean
inner product
between two vectors x, y in R will be denoted
x
y
by x y. So, if X =
and Y =
, one has X M Y = x y .

It is convenient to write a matrix O(p, 1) in the block form

A b
=
,
cT

The structure of the Lorentz group

where A GL(p), b and c are (column) vectors in Rp , and R. Notice that the

I p 0
matrices and =
are elements of O(p, 1); so, if a matrix as above lies
0
1
in O(p, 1), then the matrices

A
A b
b
=
,
=
,
c T
c T
=

cT

c T

lie in O(p, 1) as well. This means that, in our subsequent reasonings, we shall be free
to make the substitutions (b, ) 7 (b, ), (c, ) 7 (c, ), (A, b) 7 (A, b), or
(A, c) 7 (A, c).
The inversion formula (1.1) yields some useful relations between A, b, c, . One
has

!
AT c
T
=
bT
and

So, since

Ip

AT

bT

= =

!
.

, we get
A A T bbT = I p ;

(2.1)

Ac + b = 0 ;

(2.2)

c c + = 1.
Notice that, if b = (b 1 , . . . , b p ) and c = (c 1 , . . . , c p ), then c T c is the scalar c c =

(2.3)
Pp

c
i =1 i

and bbT is the (p p)-matrix whose (i , j ) entry is b i b j . Interchanging the role of


and 1 (i.e. making the substitutions A 7 A T and b 7 c), we obtain the relations
A T A cc T = I p ;
T

A b + c = 0 ;
T

b b+ = 1.

(2.4)
(2.5)
(2.6)

 It comes as no surprise that equations (2.1) (2.6) are invariant under the substitutions (b, ) 7 (b, ), (c, ) 7 (c, ), (A, b) 7 (A, b), (A, c) 7 (A, c).

Equation (2.3) or, equivalently, equation (2.6) entails that 2 1. It is an easy


matter to prove the following elementary, though very useful result.
Lemma 2.1. The following conditions are equivalent:
1) 2 = 1;
2) b = 0;
3) c = 0;
4) A O(p).

Section 2

Proof.

We now wish to apply the polar decomposition theorem for the real general linear group [2, Theorem 8.8] and express any Lorentz matrix as the product of an orthogonal matrix and a symmetric positive definite matrix . To this aim, we slavishly
follow the proof of that theorem: to get the positive definite matrix associated with
O(p, 1) we first compute the product T and then take its square root.
By direct computation making use of equations (2.4), (2.5), (2.6), and (2.3)
one gets

!
!
T
A b
A A + cc T A T b + c
AT c
T
=
=
=
cT
bT
bT A + c T bT b + 2

!
2c
I p + 2cc T
=
.
(2.7)
2c T
c T c + 2
The case 2 = 1 is trivial. Let us assume that 2 > 1; one has the following helpful
identities:
(I p +

cc T
cc T
c(c T c)c T
cc T
)(I p +
) = Ip + 2
+
= I p + cc T ;
1+
1+
1 + (1 + )2

(2.8)

(I p +

cc T
c(c T c)
c(2 1)
)c = c +
=c+
= c .
1+
1+
1+

(2.9)

Of course, equations (2.8) and (2.9) are invariant under the substitution (c, ) 7
(c, ).
The square root of T has to be a positive definite symmetric matrix: in particular, its (p, p) entry must be positive. Thus, we have to distinguish two cases:

!
p
cc T
I p + 1+
c
T
if > 1,
=
,
(2.10)
cT

!
p
cc T
I p + 1
c
T
if < 1,
=
.
(2.11)
c T

Lemma 2.2. Let =

cT

Then, the matrix

be any matrix in O(p, 1). Let =

cc T
I
+
p
S =
1 +
c T

c
=

I p + cc T

||

c T

be the sign of .

(2.12)

2
is positive definite and symmetric and S
= T . If 2 > 1, an orthonormal basis of

v1
v p1
eigenvectors for the matrix (2.12) is given by the vectors
,...,
, U 1 , U 2 , where
0
0

p
{v 1 , . . . , v p1 } is an orthonormal basis for the space c = {w R | c w = 0} Rp and

c
c
p
1p
1
U1 =
U 2 = 2 1 .
2 1 ,
2
2
1
1

The structure of the Lorentz group

p
The corresponding eigenvalues are 1 with multiplicity p 1 and 2 1 .
Proof. By Lemma 2.1, if 2 = 1, then c = 0: so, in this case, S = I p+1 , which is
symmetric and positive definite. Assume now that 2 > 1; therefore, c , 0. It is
obvious that, for any vector v such that c v = c T v = 0, one has

cc
I p + 1+

c T

!
v
v
=
,
0
0


vi
is an eigenvector of eigenvalue 1. On the other hand, taking in
0
mind the identities (2.10) and (2.11), we get
so that each

c
cc T c

c
+
+
c
cc T
p
p

p
2
2

I p + 1 + c 2 1 =
1 (1 T+ ) 1
=

c
c
1
T
+
p
c

2
1

c
c(2 1)
c + ( 1)c + 2 1c
+
+ c
p
p 2
p
2

= 1 (12 + ) 1
=
=
2 1

( 1)
p
2
+
p
1 +
2 1

p c
p
= + 2 1 2 1 .
1

p
It follows that U 1 is an eigenvector for S , having eigenvector is + 2 1 . An
analogous computation
ca be

carried out for U 2 ,pshowing that the corresponding


p
eigenvalue is 2 1 . Notice that, since 2 1 < ||, all eigenvalues are
positive, so that S is positive definite. Equations (2.8) and (2.9) show (
2
= T .
S

!) that
N

The orthogonal matrix R associated with can now be readily obtained (cf. [2,
1
Theorem 8.8]). If we let R = S
, then

T
1 T
1
1 2 1
R
R = S
S
= S
S S = I p+1 .

To get an explicit expression for R , it more convenient to compute R T = R 1 =

Section 2

S 1 :

!
cc T
A T c
c
T Ip +
R =
=
1 +
bT
c T

T T
cc T c
T
T cc A
cb
+ c
c
A +
=
=
1 +
1 +
T
T T
T
2
c A b
c c +

T
c(2 1)
cb
T
T
cb
c
+ c
A +
=
=
1 +
1 +
T
T
2
2
b b
( 1) +

T
cb
0
AT
.
=
1 +
0

(2.13)

Putting all together, we have reached the following fundamental result.


Theorem 2.3 (Polar decomposition theorem). Any =

A
cT

group O(p, 1) can be uniquely factored as = R S , where =


,
||

bc T
A

R =
1 +
0

is a matrix in O(p + 1), =

, and
||
q
S =

I p + cc T

in the Lorentz

c T

is a positive definite symmetric matrix.

 The matrix Q

bc T
lies in O(p), as it can be easily checked ( ) also by
1 +
direct computation using equations (2.1) (2.6). We have the following identity:

= A

Qc = Ac

bc T c
b(2 1)
= b
= b .
1 +
1 +

(2.14)

p
p
For any vector u in Rp , let us define the scalar (u) = uT u + 1 = u u + 1 1
and the matrix

q

uuT
I p + uuT
u
I
+
p
u
=
u =
(2.15)
1 + (u)

.
T
T
u
(u)
u
(u)
Clearly, any matrix S appearing in Theorem 2.3 corresponds to one and only one
u , with u = c (indeed, (c T )(c) = c T c) and = (u)).

The structure of the Lorentz group

Corollary 2.4. The mapping u 7 u yields an embedding Rp O(p + 1). It follows


that the group SO + (p, 1) is not compact.
N
Lemma 2.2 tells us that,
whenever

u, 0, a basis of orthonormal vectors for u is


v1
v p1
provided by the vectors
,...,
, V 1 , V 2 , where {v 1 , . . . , v p1 } is an orthonor0
0
mal basis for the space u Rp and

u
1p
V1 =
(u)2 1 ,
2
1

u
1p
V2 =
(u)2 1 .
2
1

The corresponding eigenvalues are 1 with multiplicity p 1 and (u)


As (u) 1, there is a unique real number a(u) 0 such that

p
(u)2 1.

cosh a(u) = (u) .


So, we have that
q

(u)2 1 = sinh a(u) ,


q
(u) + (u)2 1 = cosh a(u) + sinh a(u) = exp a(u) ,
q
(u) (u)2 1 = cosh a(u) sinh a(u) = exp(a(u)) .

In conclusion, with respect to the given basis of eigenvectors, the matrix u takes
the diagonal form

I p1
0
.

exp a(u)
0
0
0

exp(a(u))

3 An interlude: the case of O(1, 1)


The group O(1, 1), being one-dimensional, is susceptible of an extremely simple description. Nevertheless, the study of this toy example can be instructive to understand the topological structure of O(p, 1) with p > 1.

b
can be factored as the product

a
Theorem 2.3 tells us that any matrix =
c
= R S , where

R =

bc
1 +
0

S =

p
1 + c2
c

c
.

Section 3

p
(u)
u
2
Since = 1 + c = (c) 1, it follows that any S is of the form u =
,
u
(u)
with u = c according to the sign of . By letting a(u) = arcsinh u, one has (u) =
cosh a(u), so that any u can be written in the form

!
cosh a(u) sinh a(u)
u =
.
sinh a(u) cosh a(u)
bc
must be an element of O(1) = {1, 1}, we see that
1 +
R can only be one of the following 4 matrices:

1 0
1 0
1 0
1 0
I=
, =
, =
, I =
.
0 1
0 1
0 1
0 1

On the other hand, since a

More precisely, by using the identities (2.1) (2.6), it is easy to enumerate all possible cases:

if det = 1 and 1 , then R = I ;

if det = 1 and 1 , then R = I , ;

(3.1)

if
det

=
1
and

1
,
then
R

= ;

if det = 1 and 1 , then R = .

 The condition det = 1 is equivalent to either of the following conditions: (i)


a = ; (ii) b = c (!).
Lemma 3.1. The space (1, 1) = {u | u R} is a (closed) Lie subgroup of O(1, 1), which
is isomorphic to R. Thus, (1, 1) is the connected component of the identity in O(1, 1).
Proof. Let a 1 = a(u 1 ) = arcsinh u 1 and a 2 = a(u 2 ) = arcsinh u 2 ; then, one has

!
!
cosh a 1 sinh a 1 cosh a 2 sinh a 2
u1 u2 =
=
sinh a 1 cosh a 1 sinh a 2 cosh a 2
!

cosh a 1 cosh a 2 + sinh a 1 sinh a 2 cosh a 1 sinh a 2 + sinh a 1 cosh a 2


=
=
cosh a 1 sinh a 2 + sinh a 1 cosh a 2 cosh a 1 cosh a 2 + sinh a 1 sinh a 2
!

cosh(a 1 + a 2 ) sinh(a 1 + a 2 )
= sinh(a1 +a2 ) .
=
sinh(a 1 + a 2 ) cosh(a 1 + a 2 )
Therefore, (1, 1) is a Lie subgroup of O(1, 1) and the map R (1, 1) defined by
x 7 sinh x is a Lie group isomorphism. Since R is connected and I (1, 1), we
conclude that (1, 1) is the connected component of the identity in O(1, 1).
N

Caveat lector: when p > 1, the product of two matrices u1 , u2 in O(p, 1) may
even fail to be symmetric!

 We introduce the following notation: (1, 1) = SO (1, 1).


+

As a consequence of Lemma 3.1 and of the enumeration (3.1), we have that SO + (1, 1)
is the subgroup of O(1, 1) consisting of matrices having 1 and determinant equal

10

The structure of the Lorentz group

to 1. Let us now take the disjoint union


O + (1, 1) = SO + (1, 1) SO + (1, 1) ;
it is clear that O(1, 1) is the subgroup of O(1, 1) consisting of matrices having 1.
Also the subgroup SO(1, 1), whose elements are the matrices such that det = 1,
has two connected components, because one has
SO(1, 1) = SO + (1, 1) I SO + (1, 1)

(disjoint union) .

T
Obviously, SO + (1, 1) = O + (1, 1) SO(1, 1).
In conclusion, we have proved the following result.
Corollary 3.2. The group O(1, 1) has 4 connected components, namely
SO + (1, 1) = (1, 1) , I SO + (1, 1) , SO + (1, 1) , SO + (1, 1) .

N
To get a clearer idea of the structure of the group O(1, 1), we can analyze its action
on R2 . Since O(1, 1) is the isometry group of the Minkowski inner product X M Y =
X 1 Y1 X 2 Y2 , its orbits in the (X 1 , X 2 ) plane are contained in the level sets of the
quadratic form X 12 X 22 . Consider the branch + of the hyperbola X 12 X 22 = 1 lying
in the upper half-plane: the matrix

cosh a

sinh a

sinh a

cosh a

takes the point (0, 1) to the point (sinh a, cosh a), which lies on + (1) as well. The
action of SO + (1, 1) on + is transitive (i.e. every point is reached) and without fixed
point, so that it establishes a homeomorphism from SO + (1, 1) onto + (1) (of course,
both spaces are homeomorphic to R). Analogously, one shows that + is homeomorphic to the subspace SO + (1, 1), while the branch of the hyperbola X 12 X 22 =
1 lying in the lower half-plane turns out to be homeomorphic to both I SO + (1, 1)
and SO + (1, 1). Summing up, the group O(1, 1) is a trivial double cover of the disS
S
joint union + ; in other words, it is homeomorphic to the space (+ )O(1).

4 The topological structure of O(p, 1)


x
A vector X =
in the Minkowski space Rp+1 is said to be

space-like if X M X > 0;
time-like if X M X < 0;
light-like (or null) if X M X = 0;
positive (resp. negative) if > 0 (reps. < 0).

Section 4

11

The set of light-like vectors is a two-sheeted cone in Rp+1 , which is called the lightcone. The sheet consisting of positive (resp. negative) light-like vectors is denoted by
C + (resp. C ).
Lemma 4.1. If X and Y are both positive, or both negative time-like vectors, then
X M Y < 0. Moreover, for any t [0, 1], the linear combination (1 t )X + t Y is a timelike vector with the same parity as X and Y .


x
y
Proof. Let us assume X =
and Y =
are two positive time-like vectors. Then,

the inequality X M X < 0 is equivalent to the inequality kxk < ; likewise, we have
kyk < . Now, X M Y = x y . But x y kxkkyk < , so that X M Y < 0. For any
t (0, 1), the vectors (1 t )X and t Y are positive and time-like, and we have
((1 t )X + t Y )M ((1 t )X + t Y ) = (1 t )2 (1 t )X M X + t 2 Y M Y + 2(1 t )t X M Y < 0 .
The case of two negative time-like vectors is readily handled in the same way.

Lemma 4.1 shows that the sets

T + = X Rp+1 X time-like and positive


and T (analogously defined) are convex in particular connected topological
subspaces of Rp+1 ; of course, T + and T are disjoint. The boundary of T + is the
union of the positive light-cone C + and the origin: T + = C + {0}. Likewise, we
have T = C {0}. By contrast, notice that, if p > 1, the space of space-like vectors
is connected.
Every transformation O(p, 1) preserves the likeness of any vector, but not its
parity; in fact, maps positive vectors to negative ones and conversely. Let us define
the subgroup
O + (p, 1) = { O(p, 1) | T + T + } .
It is quite immediate to see that O + (p, 1) has index 2 in O(p, 1): indeed, the action of
O(p, 1) on the space of time-like vectors is continuous and this space has precisely
two connected components, namely T + and T .
Lemma 4.2. Let =
and only if > 0.

cT

be a transformation in O(p, 1); lies in O + (p, 1) if



b
c
and
are time-like.


x
Assume that > 0. Then, for any positive time-like vector X =
, we have X = Y ,



y
c
where Y =
is time-like and = c x+. The vectors
and X are both positive

and time-like, so that, by Lemma 4.1, their Minkowski product is positive and timelike as well: c x < 0. Since > 0, we have also |c x| kckkxk < ; hence,
> 0, and O + (p, 1). Conversly, assume that O + (p, 1). Then, for any positive
Proof. First notice that, by eqq. (2.3) and (2.6), the vectors

12

The structure of the Lorentz group


0
time-like vector X , the vector X = Y is time-like and positive. If we take X =
, we
1

b
obtain Y =
; therefore, it must be > 0.
N

Let us now consider the subgroup SO(p, 1) of O(p, 1) consisting of Lorentz transformations having determinant equal to 1. This subgroup has index 2 in O(p, 1).
Note that SO(p, 1) is not contained in O + (p, 1): in fact, for any Q O(p) with detQ =

0
Q
lies in SO(p, 1) but not in O + (p, 1). It follows that the
1, the matrix
0 1
subgroup
SO + (p, 1) = O + (p, 1) SO(p, 1)

1 0
, we
has index 4 in O(p, 1). More precisely, if we introduce the matrix =
0 I
see that O(p, 1) admits the following disjoint union decomposition:
O(p, 1) = SO + (p, 1) SO + (p, 1) SO + (p, 1) ()SO + (p, 1) .

By Theorem 2.3, any =

cT

in O(p, 1) can be expressed as the product

u ,

, Q O(p) and u is given by eq. (2.15). It is easy to enumerate all


where = ||
possible cases:

SO (p, 1) if and only if Q SO(p) and = 1;

SO + (p, 1) if and only if detQ = 1 and = 1;

SO
(p,
1)
if
and
only
if
Q

SO(p)
and = 1;

+
()SO (p, 1) if and only if detQ = 1 and = 1.

(4.1)

 Using Lemma 4.2 it is immediate to obtain the following disjoint union decomposition

O + (p, 1) = SO + (p, 1) SO + (p, 1) .

(4.2)

In order to extend Corollary 3.2 to the general case of O(p, 1), we have only to show
that SO + (p, 1) is connected. We introduce the space
H p+ = {X T + | X M X = 1} ,
which is obviously diffeomorphic to Rp . A diffeomorphism
(which we will regard as

x
an identification) is provided by the projection X =
Hp 7 x Rp , whose inverse

x
is clearly given by the mapping x 7 p
.
1+xx

Section 5

13

Lemma 4.3. The group SO + (p, 1) acts transitively on H p+ . The isotropy group N of

Q 0
0
with Q SO(p).
the vector N =
is the subgroup of transformations
1
0 1

p
x
Proof. Given any vector X =
H p+ , one has 0 < = 1 + x x = (x). The matrix

x defined according to eq. (2.15), namely


q
x =

I p + xx T
xT

(4.3)

(x)


0
to X . The second claim is now a di1
rect consequence of the polar decomposition theorem 2.3, taking into account the
enumeration (4.1).
N
is an element of SO + (p, 1) and maps N =

 The isotropy group

of any other vector X is a subgroup of SO + (p, 1) conjugate


to N . Precisely, if X = N, then X = 1 N .
X

Theorem 4.4. There is a diffeomorphism


H p+ ' SO + (p, 1)/SO(p) .
As a consequence, the group SO + (p, 1) is connected (therefore, it is the connected component of the identity in O(p, 1)).
Proof. The first statement follows from [3, Chap. 1, Prop. 4.6] and from Lemma 4.3:
since the action of SO + (p, 1) on H p+ is transitive and the isotropy group of any point
can be identified with SO(p), there is a diffeomorphism H + ' SO + (p, 1)/SO(p). As
for the second claim, let us consider the projection : SO + (p, 1) H p+ . Now, H +
is connected and, for any X H p+ , the fibre 1 (X ) is diffeomorphic to SO(p) and
therefore connected. Given any two points 1 , 2 in SO + (p, 1), we can draw paths
connecting 1 , 2 to (1 ), (1 ) and a path connecting (1 ) to (2 ). Thus,
SO + (p, 1) is connected.
N

 Actually, it not hard to show that the SO(p)-principal bundle : SO (p, 1) H


+

is trivial: SO

(p, 1) ' H p+ SO(p).

+
p

Corollary 4.5. The group O(p, 1) has 4 connected components, which are explicitly
described in enumeration (4.1).
N
Using Theorem 4.3 and the decomposition (4.2) we easily get the following result.
Corollary 4.6. The group O + (p, 1) acts transitively on H p+ . The isotropy group of the

Q 0
with Q O(p).
N
vector N is the subgroup of transformations
0 1

14

The structure of the Lorentz group

5 Hyperbolic geometry
In this Section we prove a result of fundamental importance, namely that the group
O + (p, 1) is the isometry group of the p dimensional hyperbolic space.
If X , Y are two vectors in H p+ , their Minkowski product, by Lemma 4.1, is a negative
real number. So, there exists a unique non-negative real number (X , Y ) such that
cosh (X , Y ) = X M Y .
Clearly, we have
(X , X ) = 0 ,
(X , Y ) > 0 for all X , Y ,
(X , Y ) = (Y , X ) .
In order to show that the function : H p+ H p+ R 0 is a (topological) metric it
remains only to prove that it satisfies the triangle inequality.
Lemma 5.1. For all vectors X , Y , Z in H p+ , the triangle inequality
(X , Z) (X , Y ) + (Y , Z)
is satisfied.
Proof. Notice that, by definition, (X , Y ) = (X , Y ) for any O + (p, 1). Since

0
O + (p, 1) acts transitively on H p+ , there is no restriction in assuming that Y =
. Let
1


x
z
X=
and Z =
. We have X M Y = and Y M Z = ; then, cosh (X , Y ) =

and cosh (Y , Z) = . Now, 2 = kxk2 + 1, so that sinh (X , Y ) = kxk; analogously, we


get sinh (Y , Z) = kzk. One has:
X M Z = x z + =
= kxkkzk cos + cosh (X , Y ) cosh (Y , Z)
sinh (X , Y ) sinh (Y , Z) + cosh (X , Y ) cosh (Y , Z) =

= cosh (X , Y ) + (Y , Z) .
Now, cosh (X , Z) = X M Z; since the hyperbolic cosine is an increasing function,
we conclude that
(X , Z) (X , Y ) + (Y , Z) .

 The p-dimensional hyperbolic space is the metric topological space H


A map : Hp Hp is an isometry if it preserves distances, namely if
((X ), (Y )) = (X , Y ) for all X , Y in Hp .

= H p+ , (, ) .

Section 5

15

It is easy to get convinced that any isometry of a metric space is continuous and
injective; as it will be shown in Theorem 5.2, any isometry : Hp Hp happens
also to be surjective (a property that, of course, is not satisfied in the general case).
Theorem 5.2. A map : Hp Hp is an isometry if and only if it is the restriction of a
transformation O + (p, 1).
Proof. It is clear that map : Hp Hp is an isometry if and only if (X )M (Y ) =
X M Y for all X , Y in Hp . Therefore, the restriction of any O + (p, 1) to Hp is an
isometry. Conversely,
let be an isometry, and let us suppose first that (N) = N.
x
Hence, for any X =
, we have:

(X )M N = (X )M (N) = X M N = .

(x)
: Rp Rp preserves
, where the map

the Euclidean product, namely (x)


(y)
= x y for all x, y in Rp . Therefore, by a

well-known basic result in Euclidean geometry, there is Q O(p) such that (x)
= Qx

Q 0
p
X for all X Hp . Let us now suppose
for all x R . Thus, we get (X ) =
0 1
that (N) , N. Since SO + (p, 1) acts transitively
on Hp , there exists a transformation

1 lying in that group such that 1 (N) = N. But then the map 1 : Hp Hp is

Q 0
an isometry fixing the point N; thus, for all X Hp , one has 1 (X ) =
X,
0 1
for some Q O(p). Having in mind the decomposition (4.2), we conclude that
(X ) = 2 X , for some 2 O + (p, 1).
N
It follows that has the form (X ) =

Let us introduce the isometry group of Hp :

I (Hp ) = { : Hp Hp | isometry} .
It is possible to prove that I (Hp ) carries a natural structure of Lie group [5, Chap. 5].
So, Theorem 5.2 can be rephrased in the following way.
Corollary 5.3. There is a Lie group isomorphism I (Hp ) ' O + (p, 1). The subgroup
I + (Hp ) of orientation-preserving isometries is isomorphic to SO + (p, 1).
N

 The space H , as a manifold, can naturally be endowed by restricting the


p

Lorentzian metric on Rp+1 with a Riemannian metric h (therefore, for any point
X Hp , h X is a positive-definite inner product on the tangent space TX Hp ). According to a general procedure valid for all Riemannian metrics, one can associate to
h a topological distance function h such that the distance h (X , Y ) between any
two (sufficiently close) points is equal to the arc length (measured by using h) of the
geodesic segment joining X and Y (see e.g. [4, 1.4]). It turns out that the distance
function distance h coincides with the distance function we have previously defined. As a consequence, the (topological) isometry group I (Hp ) coincides with the
(differentiable) isometry group of the Riemannian manifold (Hp , h).
To get a more perspicuous picture of the hyperbolic space Hp , one can construct

16

The structure of the Lorentz group

Figure 1: The stereographic projection of B p onto Hp from the point (0, 1) maps
2v
1
the point (v, 0) to the point X = 1kvk
2
1+kvk2
its Poincars conformal model. Let B p be the open p-dimensional ball in the Euclidean space Rp+1 . The map:
: Hp B p

x
x
X=
7

1+
is a diffeomorphism. Actually, an easy computation shows that
x 2 1

<1

=
1+
+1
and that the inverse of is the map
1 : B p Hp
v 7

 Let us embed B

1
2v
2 .
1 kvk2 1 + kvk

(5.1)

into R p+1 by mapping the point v to (v, 0). Then, the map
: B Hp is the stereographic projection of B p onto Hp from the point (0, 1),
as shown in Figure 1.
1

We can obviously induce a topological metric on B p by setting

(v, w) = 1 (v), 1 (w)


for any pair (v, w) B p .
It is quite remarkable that the distance (v, w) can be expressed in terms of Euclidean invariants.
Lemma 5.4. For any pair (v, w) of points in B p , one has

cosh (v, w) = 1 + 2

kv wk2
.
(1 kvk2 )(1 kwk2 )

(5.2)

Section 6

17



x
y
Proof. Let v = (X ) and w = (Y ), with X =
and Y =
. By definition we have

cosh (v, w) = cosh (X , Y ) = X M Y .


The following identity is easily obtained from the expression of v in terms of X :
1 kvk2 = 1
similarly, one has 1 kwk2 =

2
kxk2
=
;
(1 + )2 1 +

2
. Then, by direct computation we get:
1+

kv wk2
=
(1 kvk2 )(1 kwk2 )
2

(1 + )(1 + ) 1
2 1
2x y
= 1+
+

=
2
(1 + )2 (1 + )2 (1 + )(1 + )

1
= 1 + 2 + 2 2x y = X M Y .
2

1+2

The isometry group of the space B p , (, ) is called the p-dimensional Mbius


group: this group is obviously isomorphic to the Lorentz group O(p, 1) and provides
an extremely valuable tool to investigate its structure. We refer the reader to Ratcliffes book [5] for a thorough treatment of the general case. Here we limit ourselves,
in the next Section, to analyse the 2-dimensional case, which exhibits some special
features.

6 The Poincar disc


The open ball B 2 is diffeomorphic to open disc D in the complex plane,
D = {z C | kzk < 1} .
As usual, a vector x = (x 1 , x 2 ) is identified with the complex number z = x 1 + i x 2 ;
2
under this identification
q the Euclideanpnorm on B is equal to the Hermitian norm
2
2
Therefore, we can define the hyperbolic
on D, namely kxk = x 1 + x 2 = kzk = z z.
metric on D given by eq. (5.2) in terms of complex quantities only:

(z 1 , z 2 ) = arcosh 1 + 2

kz 1 z 2 k2
(1 kz 1 k2 )(1 kz 2 k2 )

(6.1)

for any pair of points z 1 , z 2 in D. The disc D endowed with such a metric is called the
Poincar disc.
= C taking
The group of homeomorphisms of the extended complex plane C
circles to circles is the group of Mbius transformations, which is generated by the
linear fractional transformations
z 7

az + b
cz + d

a, b, c, d C , ad bc = 1

18

The structure of the Lorentz group

7 (see [1, Chap. 2] for an elementary proof of this


and the by the map z 7 z,
fundamental result). It is not hard to show that any linear fractional transformation
preserving the disc D has either the form
z 7

az + b
,
+ a
bz

(6.2)

z 7

a z + b
,
b z + a

(6.3)

or the form

where a, b C and kak2 kbk2 = 1.

 The group of Mbius transformations of the form (6.2) will be denoted by M

(D).

These transformations preserve the natural orientation of the disk.

Lemma 6.1. The Mbius group M + (D) is isomorphic to the projective special linear
group P SL(2; R).
az + b
Proof. Any transformation m in M + (D) of the form z 7
can be mapped
+ a
bz

a b
to the matrix A m =
, which lies in SL(2; C) because kak2 kbk2 = 1. Clearly,
b a
this correspondence is well-defined only up to the sign of A m ; thus A m has to be
thought of as an element of group P SL(2; C). In this way, since the composition of
two transformations m, m 0 is associated with the product matrix A m A m 0 ( ), we
get an isomorphism M + (D) , where is a subgroup of P SL(2; C). Let us now
consider the matrix

1 i
P=
.
i
1

The product
P 1 A m P =

1 1
2 i

i
1

a
b

b
a

1
i

i
a + b
= Abm =
1
b a

b + a
a b

is a matrix with real entries whose determinant is


(a)2 (b)2 (b)2 + (a)2 = kak2 kbk2 = 1 ;
This shows that the group is conjugate (in PGL(2; C)) to the group P SL(2; R). As
a consequence, we get an isomorphism M + (D) SL(2; R) explicitly given by the
assignment m 7 Abm .
N

 The matrix P = i1

i
is associated with the Mbius transformation
1
z 7

z i
,
i z + 1

which provides a holomorphic diffeomorphism of the upper-half complex plane U


onto the disc D. Thus, Lemma 6.1 implies that M + (U) ' P SL(2; R). We refer the
reader to [1] and [5] for further details.

Section 6

19

A Mbius transformation z 7
composition

az + b
in M + (D) can always be factorized into the
+ a
bz
z 7 e i

where e i =
type

a
a

z +c
,
cz + 1

and c = ba . As kak2 kbk2 = 1, one has kck < 1. A transformation of


z 7 e i z

(6.4)

acts as a rotation in the complex plane and hence it preserves both the Euclidean
distance and the hyperbolic metric (6.1) on D. On the other hand, the transformations of type
z +c
(6.5)
z 7
cz + 1
do not preserve, in general, the Euclidean distance but they do preserve the hyperbolic metric (6.1). Though this fact can be shown through a (somewhat tedious) direct computation, we will follow a different route, which has the advantage of bringing to light the existence of an isomorphism of the groups M + (D) and SO + (2, 1) (see
Theorem 6.4).
: H2 B 2 , which maps the vector X =
We
have introduced earlier an isometry
p
x
x
to the point
; recall that = kxk2 + 1 > 0. By identifying B 2 and D as

1+
above, the map takes the form
: H2 D

x1
1
(x 1 + i x 2 ) .
X = x 2 7 (X ) =
1+

It is immediate to deduce the following identity:


=

1 + k(X )k2
.
1 k(X )k2

(6.6)

Therefore, by eq. (5.1), the inverse map 1 : D H is given by the assignment


z 7 1 (z) =

(z)
,
(z)

where

1
2z
(z) =
,
1 kzk2 2z

(z) =

1 + kzk2
.
1 kzk2

(6.7)

Note that the map : D R2 is one-to-one.


A question that naturally arises at this point is which kind of transformations are
induced on H2 by the Mbius transformations on D through the map 1 . The answer provided by the next two results could hardly be simpler: they are just Lorentz
transformations.

20

The structure of the Lorentz group

Lemma 6.2. Let m be the Mbius transformation given by


z 7 m (z) = e i z .
Then, for any z D, one has
1 (m (z)) = R 1 (z) ,
where

cos
R = sin
0

sin
cos
0

Proof. It is a straightforward computation (

0
0 .
1

!).

Theorem 6.3. Let m c be the Mbius transformation given by


z 7 m c (z) =

z +c
,
cz + 1

with c D. Then, for any z D, one has


1 (m c (z)) = (c) 1 (z) .
Proof. Let us first write down the Lorentz transformation (c) . For the sake of
2
brevity, we write (c) in the form (c) =
c, where c is the 2-dimensional
1 kck2
vector corresponding to the complex number c. An easy computation yields
((c)) =

p
1 + kck2
(c) (c) + 1 =
.
1 kck2

Hence, by eq. (2.15), we have


(c)(c)T
I+
(c) =
1 + ((c))

(c)T

(c)
((c))

2
cc T
1 kck2
2
cT
1 kck2

I+

2
c
1 kck2
1 + kck2
1 kck2

Using the same convention as for (c), we write 1 (z) in the form
1 (z) =

1
2z
2 .
1 kzk2 1 + kzk


y
. It will be enough to compute the spatial component y of

(c) 1 (z) and then to check that is equal to to the spatial component (m c (z))
of 1 (m c (z)); indeed, eqq. (6.6) and (6.7) ensure that the temporal" components
and (m c (z)) will be equal as well. On one hand, we get
We set (c) 1 (z) =

y=

2z
4c
2(1 + kzk2 )c
+
c

z
+
.
1 kzk2 (1 kck2 )(1 kzk2 )
(1 kck2 )(1 kzk2 )

(6.8)

Section 6

21

On the other hand, we find out that the complex number w corresponding to the
vector w = (m c (z)) is
w=

2
1

kz+ck2
2

kcz+1k

2(c z + 1)(z + c)
z +c
=
.
cz + 1 kcz + 1k2 kz + ck2

It is immediate to verify that kcz + 1k2 kz + ck2 = (1 kck2 )(1 kzk2 ); thus, we have
w=

2(ckzk2 + z + c 2 z + c)
=
(1 kck2 )(1 kzk2 )

2(z + c 2 z)
2(1 + kzk2 )c
+
=
=
(1 kck2 )(1 kzk2 ) (1 kck2 )(1 kzk2 )
2(1 + kzk2 )c
2(z + c(c z + cz cz))
=
+
=
(1 kck2 )(1 kzk2 )
(1 kck2 )(1 kzk2 )
2(1 + kzk2 )c
2c(c z + cz)
2z
=
+
+
.
(1 kck2 )(1 kzk2 ) 1 kzk2 (1 kck2 )(1 kzk2 )

The identity kc + zk2 = kck2 + kzk2 + c z + cz shows that c z + cz = 2c z. Therefore we


get
(m c (z)) =

2z
4c
2(1 + kzk2 )c
+
+
c z = y.
(1 kck2 )(1 kzk2 ) 1 kzk2 (1 kck2 )(1 kzk2 )

This concludes the proof.

We are now in a position to establish the main result of this Section, which provides a particularly significant characterization of the Lorentz group O(2, 1).
Theorem 6.4. The group SO + (2, 1) is isomorphic to the Mbius group M + (D); therefore, there is an isomorphism
SO + (2, 1) ' P SL(2; R) .

(6.9)

As consequence, M + (D) is the orientation-preserving isometry group of the Poincar


disc.
Proof. Theorem 2.3, along with the characterization of SO + (2, 1) in (4.1), tells us
that any transformation in SO + (2, 1) is the product of a matrix of type u and an

Q 0
orthogonal matrix of type
, where u R2 and Q SO(2). Thus, because of
0 1
the invertibility of both and , the first claim follows directly from Lemma 6.2 and
Theorem 6.3. The isomorphism (6.9) is then an immediate consequence of Lemma
6.1. Since the map : H2 D is an isometry, the last claim follows directly from
Corollary 5.3.
N

References
[1] A NDERSON , J AMES W., Hyperbolic Geometry, second edition, Springer, New
York 2005.

22

The structure of the Lorentz group

[2] B ARTOCCI , C LAUDIO, Notes in Linear Algebra, 2012 (available online).


[3] B RCKER , T HEODOR & TOM D IECK , TAMMO, Representations of Compact Lie
Groups, Springer-Verlag, New York 1985.
[4] J OST, J RGEN, Riemannian Geometry and Geometric Analysis, Springer-Verlag,
Berlin-Heidelberg 1995.
[5] R ATCLIFFE , J OHN G., Hyperbolic Geometry, second edition, Springer, New York
2006.
[6] WARNER , F RANK W., Foundations of Differentiable Manifolds and Lie Groups,
Springer-Verlag, New York 1983.

Вам также может понравиться