To show that SU (n) is simply connected, we use the action of SU (n) on the

sphere S 2n−1 = { z ∈ Cn | |z|2 = 1 }. This action is clearly transitive and the

isotropy group at (1, 0, . . . , 0) is isomorphic to SU (n− 1). This yields a fibration
SU (n)/SU (n − 1) = S 2n−1 .
The long exact sequence in homotopy corresponding to this fibration gives



. . . → π2 (S 2n−1 ) → π1 SU (n − 1) → π1 SU (n) → π1 (S 2n−1 ).
If n ≥ 2, we have π2 (S 2n−1 ) = 0 and π1 (S 2n−1 ) = 0 thus





π1 SU (n) ≃ π1 SU (n − 1) ≃ . . . ≃ π1 SU (1) = π1 {1} = 0.


Corollary 9 The symplectic group Sp(R2n , Ω0 ) is homeomorphic to the topo-

logical product of the circle S 1 , of the group SU (n) and of the vector space of
real symplectic 2n × 2n matrices in the symplectic Lie algebra sp(R2n , Ω0 ). The
fundamental group of Sp(R2n , Ω0 ) is isomorphic to Z. Every continuous map
from Sp(R2n , Ω0 ) to S 1 which coincides with the determinant on U (n) induces
an isomorphism of the fundamental groups.
Those results follow directly from the two previous lemmas and from the fact
that π1 (S 1 ) = Z. We shall construct in the next section a continuous map ρ from
Sp(R2n , Ω0 ) to S 1 which coincides with the determinant on U (n) (inducing thus
an isomorphism of the fundamental groups).

2 The rotation map ρ

Theorem 10 ([12]) There exists a unique family of continuous maps
ρ : Sp(R2n , Ω0 ) → S1
(one for each integer n ≥ 1) with the following properties:
1. [determinant] ρ coincides with detC on the unitary subgroup
ρ(A) = detC A if A ∈ Sp(2n) ∩ O(2n) = U(n);
2. [invariance] ρ is invariant under conjugation :
ρ(kAk −1 ) = ρ(A) ∀k ∈ Sp(R2n , Ω0 );

3. [normalisation] ρ(A) = ±1 for matrices which have no eigenvalue on the

unit circle;
4. [multiplicativity] ρ behaves multiplicatively with respect to direct 
sums :
′
A 0
if (R2n , Ω0 ) = (R2m , Ω0 ) ⊕ (R2(n−m) , Ω0 ), and if A = with
0 A′′
A′ ∈ Sp(R2m , Ω0 ) and A′′ ∈ Sp(R2(n−m) , Ω0 ), then
ρ(A) = ρ(A′ )ρ(A′′ ).

8
2.1 Construction of ρ in dimension 2
We use as before the identification R2 ∼
= C; the matrix of the rotation in R2 by
an angle ϕ is identified with the multiplication by eiϕ in C :
   !
cos ϕ − sin ϕ iϕ cos ϕ − sin ϕ
↔e thus ρ = eiϕ .
sin ϕ cos ϕ sin ϕ cos ϕ

This gives in particular: ρ (Id) = 1 and ρ (− Id) = −1.

In dimension 2 we have
  
2 a + d −b + c 2 2 2 2
Sp(R , Ω0 ) = Sl(2, R) = | a +b −c −d =1 .
b+c a−d
The eigenvalues of A ∈ Sp(R2 , Ω0 ) are the roots of det(A√− λ Id), thus the
solutions of λ2 − 2aλ + 1 = 0; so the eigenvalues are λ = a ± a2 − 1. There are
three possible cases:
1. a2 > 1 : In this case the two eigenvalues are real and distinct (their
product is equal to 1but they differ from 1 and −1); so ∃k ∈ Sl(2, R)
λ 1 0
such that kAk −1 = . By the invariance condition ρ(A) =
0 λ11
ρ(kAk −1 ). By continuity of ρ and by the normalisation condition, since
we have seen that ρ (Id) = 1 and ρ (− Id) = −1, we have
ρ(A) = 1 if a > 1 and ρ(A) = −1 if a < −1.

2. a2 < 1 : There are no real eigenvalues; in the complexified space C2 , the

two eigenvalues are complex conjugate and their product is 1; denote them
by eiϕ , e−iϕ .
If z = v − iw (z1 = v1 − iw1 , z2 = v2 − iw2 ) is an eigenvector for A in
C2 of eigenvalue eiϕ , the vector z = v + iw = (z1 , z2 ) is an eigenvector of
−iϕ
eigenvalue
 ′ e :   iϕ   ′    −iϕ 
′
a b z1 e z1 a b′ z1 e z1
= , = .
c′ d′ z2 eiϕ z2 c′ d′ z2 e−iϕ z2
Hence we have


Av = A 12 (z + z)
= 12 (cos ϕ + i sin ϕ)(v − iw) + 21 (cos ϕ − i sin ϕ)(v + iw)
= cos ϕv + sin ϕw


Aw = A 2i (z − z)
i
= 2 (cos ϕ+ i sin ϕ)(v − iw) − 2i (cos ϕ − i sin ϕ)(v + iw)
= − sin ϕv + cos ϕw.
so that in the basis {v, w} of R2 the matrix associated to A is given by
 
cos ϕ − sin ϕ
.
sin ϕ cos ϕ

9
 Remark 11 Denote by λ ( and call Krein positive) the eigenvalue eiϕ or
e−iϕ for which Ω0 (vλ , wλ ) > 0 if zλ = vλ − iwλ is an eigenvector of A in C2
of eigenvalue λ. Another eigenvector corresponding to the same eigenvalue
λ has the form z ′ = αzλ = (ã+ib̃)(vλ −iwλ ) = (ãvλ + b̃wλ )−i(−b̃vλ +ãwλ )
so that Ω0 (ãvλ + b̃wλ , −b̃vλ + ãwλ ) = (ã2 + b̃2 )Ω0 (vλ , wλ ) and the sign
of Ω0 (v, w) does not depend on the choice of the eigenvector z = v − iw.
On the other hand z = v − (−iw) is an eigenvector corresponding to
the conjugate eigenvalue and will correspond clearly to the opposite sign.
Remark also that Ω0 (z, z) = 2iΩ0 (v, w) so that Im Ω0 (zλ , zλ ) > 0.
   
2 0 1 v1 w1
In R , Ω0 is the matrix . We have det = Ω(v, w).
−1 0 v2 w2
Defining
 
1 v1 w1
k= p , where (v1 , v2 )+i(w1 , w2 ) = vλ +iwλ = zλ ,
Ω0 (vλ , wλ ) v2 w2

2
the
 matrix
 k is of determinant  1, thus k ∈ Sp(R , Ω0 ), and we have
1 0
k =√ 1 v and k = √ 1 w so that
0 Ω0 (vλ ,wλ ) 1 Ω0 (vλ ,wλ )

 
cos ϕ − sin ϕ
(k −1 Ak) = .
sin ϕ cos ϕ

Thus, if a2 < 1 and if eiϕ is an eigenvalue such that Az = eiϕ z where

z = v − iw with Ω0 (v, w) > 0, we get, using the invariance and the
determinant conditions for ρ,

ρ(A) = eiϕ .

3. a2 = 1 : We already know that ρ(Id) = 1 and ρ(− Id) = −1.

Let v be an eigenvector of A of eigenvalue a and suppose that there doesn’t
exist another eigenvector linearly independent from v. For any w linearly
independent from v, we have

Av = av
Aw = c̃v + aw.
 
v1 w1
Take w such that Ω0 (v, w) = 1. Then k = ∈ Sp(R2 , Ω0 ) and
v2 w2
 
a c̃
k −1 Ak = .
0 a

This matrix
 tis the t limit,
 for t tending to 0, of the path of sympletic
e a e c̃
matrices . For t 6= 0, these matrices have two distinct
0 e−t a

10
 real eigenvalues equal to et a and e−t a; so their image under ρ is equal to
a (cf case 1). By invariance and continuity of ρ, we have

ρ(A) = a.

To summarize:

Proposition 12 Conditions (1), (2) et (3) determine a unique continuous map

ρ : Sl(2, R) → S1 : A 7→ ρ(A)

defined by
 1

 1 ifTr(A) =: a > 1
2



 i.e if the eigenvalues of A are real positive;

−1 if a 6 −1
ρ(A) =

 i.e if the eigenvalues of A are real negative;

 iϕ iϕ

 e if e is an eigenvalue of A such that Ω0 (v, w) > 0

when z = v − iw is a corresponding eigenvector.

The condition (4) is of course empty in dimension 2.

2.2 Construction of ρ in any dimension.

We extend Ω0 C-linearly to V ⊗R C and represent it by the same matrix.
We also extend any A ∈ Sp(V = R2n , Ω0 ) C-linearly to V ⊗R C. If vλ de-
notes an eigenvector of A ∈ Sp(V = R2n , Ω0 ) in V ⊗R C of eigenvalue λ
then Ω0 (Avλ , Avµ ) = Ω0 (λvλ , µvµ ) = λµΩ0 (vλ , vµ ), thus Ω0 (vλ , vµ ) = 0 un-
less µ = λ1 . Hence the eigenvalues of A arise in “quadruples”
 
1 1
[λ] := λ, , λ, . (5)
λ λ

2.2.1 ρ(A) for a semisimple element A

Definition 13 An element A is semisimple if V ⊗R C is the direct sum of its
eigenspaces.
Denote by Eλ the eigenspace corresponding to the eigenvalue λ in V ⊗R C.
Remark that if v = u + iu′ is in Eλ with u and u′ in V then v := u − iu′ is in
Eλ so that Eλ ⊕ Eλ is the complexification of a real subspace of V . The space

W[λ] := Eλ ⊕ E λ1 ⊕ Eλ ⊕ E 1 (6)
λ

is the complexification of a real symplectic subspace V[λ] and

R2n = V[λ1 ] ⊕ V[λ2 ] ⊕ . . . ⊕ V[λK ] (7)

11
where the direct sum is symplectic orthogonal and where [λ1 ], . . . [λK ] are the
distinct quadruples exhausting the eigenvalues of A. Hence, by multiplicativity
of ρ
ρ(A) = ρ(A[λ1 ] ) · ρ(A[λ2 ] ) · . . . · ρ(A[λK ] ) (8)
where A[λi ] is the restriction of A to V[λi ] .
• If λ = ±1, we have A[λ] = ± Id; thus by the determinant condition, we
have
    12 dim V[λ]
1 if λ = 1 λ
ρ(A[λ] ) = ρ(± Id) = 1 = .
−1 2 dim V[−1] if λ = −1 |λ|
(9)
• If λ ∈ R \ {±1}, no eigenvalue is on S 1 ; by the normalisation condition
and continuity, we have
    21 dim V[λ]
1 if λ > 0 λ
ρ(A[λ] ) = 1 = . (10)
−1 2 dim V[λ] if λ < 0 |λ|

• If λ ∈ C \ S 1 ∪ R, no eigenvalue is on S 1 ; by the normalisation condition

and continuity, we have
ρ(A[λ] ) = 1. (11)
Indeed, we can bring continuously λ, λ1 , λ, λ1 simultaneously all to 1 (or all
to −1) and in both cases ρ(A[λ] ) = 1 because dim V[λ] = 4 dimC Eλ .
• If λ = eiϕ ∈ S 1 \ {±1}, W[λ] := Eλ ⊕ Eλ ; we define

Q : Eλ × Eλ → R : (z, z ′ ) 7→ Q(z, z ′) := Im Ω0 (z, z ′ ); (12)

where Im a denotes the imaginary part of a complex number a. It is a

nondegenerate symmetric 2-form on the vector space Eλ viewed as a real
vector space. It is indeed symmetric because

Im Ω0 (z, z ′ ) = −Im Ω0 (z ′ , z) = Im Ω0 (z ′ , z) = Im Ω0 (z ′ , z)

and it is nondegenerate because

1

Q(z, z ′ ) = Im Ω0 (z, z ′ ) = Ω0 (z, z ′ ) − Ω0 (z, z ′ )
2i
so that Q(z, z ′) = 0 ∀z ∈ Eλ iff Ω0 (z, z ′ ) − Ω0 (z, z ′ ) = 0 ∀z ∈ Eλ .
Replacing z by iz this implies iΩ0 (z, z ′ ) + iΩ0 (z, z ′ ) = 0 ∀z ∈ Eλ hence
Ω0 (z, z ′ ) = 0 ∀z ∈ Eλ and this implies z ′ = 0.
We can thus find a vector z1 ∈ Eλ such that Q(z1 , z1 ) = 2a1 6= 0. Writing
z1 = u1 − iv1 , the subspace of V[λ] generated by u1 and v1 is symplectic

12
 and stable by A; we have Ω0 (u1 , v1 ) = a1 and the restriction A1 of A to
this subspace in the basis {u1 , v1 } has the form
 
cos ϕ − sin ϕ
A1 = .
sin ϕ cos ϕ

If a1 > 0, we have as before (cf the case where a2 < 1), ρ(A1 ) = eiϕ . If
a1 < 0, we permute the vectors u1 and v1 and we have ρ(A1 ) = e−iϕ .
Indeed v1 − iu1 = −i(u1 + iv1 ) = −i(u1 − iv1 ) which is of eigenvalue λ =
e−iϕ . In conclusion as V[λ] is the direct sum of hu1 , v1 i and its symplectic
orthogonal, we have
i
ρ(A[λ] ) = e 2 ϕ Sign(Q) (13)
where Sign(Q) denotes the signature of Q (the number of positive eigenval-
ues minus the number of negative eigenvalues of Q). The 12 factor comes
from the fact that Q(iz, iz ′) = Q(z, z ′) and the vectors z and iz define the
same real vector space Span{u1 , v1 } of V[λ] .
The map ρ is continuous as can be viewed by considering possible variations
of the “quadruples” of eigenvalues; for example, in a continuous variation of
the matrix A, a quadruple of eigenvalues corresponding to λ ∈ / S 1 ∪ R can
only degenerate into a pair of real eigenvalues counted twice or into a pair of
eigenvalues on
n the circleo counted twice but with opposite signs. In this last
1 1
case, [λ] = λ, λ , λ, λ degenerates into {eiϕ , e−iϕ , e−iϕ , eiϕ } and the image
by ρ of the corresponding matrix is 1. Indeed, if λ = reiϕ ∈ / S 1 ∪ R is an
eigenvalue and if z = u − iv is an eigenvector of A of eigenvalue λ and z ′ =
u′ +iv ′ an eigenvector of eigenvalue 1/λ such that Ω0 (z, z ′ ) = 2, then Ω0 (z, z ′ ) =
0; Ω0 (z, z) = 0; Ω0 (z, z ′ ) = 0; Ω0 (z, z ′ ) = 2; Ω0 (z ′ , z ′ ) = 0, so that the 2-form Ω
and the matrix A are written, in the basis {u, v, u′ , v ′ },
   
0 0 1 0 r cos ϕ −r sin ϕ 0 0
 0 0 0 1   r sin ϕ r cos ϕ 0 0 
Ω0 =   −1 0 0 0  A = 
  1 1
.

0 0 r cos ϕ − r sin ϕ
1 1
0 −1 0 0 0 0 r sin ϕ r cos ϕ

and the limit when r equals 1 give ρ(A) = eiϕ e−iϕ = 1 because
 in the basis
0 0 0 −2
 0 0 2 0 
{z, z ′, iz, iz ′} of Eeiϕ the matrix of Q is  
 0 2 0 0 .
−2 0 0 0

2.2.2 ρ(A) for any A

Lemma 14 Consider A ∈ Sp(V, Ω) and let λ be an eigenvalue of A in V ⊗R C.
Then Ker(A − λ Id)j is the symplectic orthogonal of Im(A − λ1 Id)j .

13
Proof:


Ω (A − λ Id)u, Av = Ω(Au, Av) − λΩ(u, Av) = Ω(u, v) − λΩ(u, Av)


= −λΩ u, A − λ1 Id v

and by induction


j 
1
Ω (A − λ Id)j u, Aj v = (−λ)j Ω u, A − λ Id v . (14)

The result follows from the fact that A is invertible. 

Corollary
 15 If Eλ denotes the generalized eigenspace of eigenvalue λ, i.e
Eλ := v ∈ V ⊗R C | (A − λ Id)j v = 0 for an integer j > 0 , we have

Ω(Eλ , Eµ ) = 0 when λµ 6= 1.

Indeed the symplectic orthogonal of Eλ = ∪j Ker(A − λ Id)j is the intersec-

tion of the Im(A − λ1 Id)j . By Jordan normal form, this intersection is the sum
of the generalized eigenspaces corresponding to the eigenvalues which are not λ1 .

Lemma 14 will be used in the next section to establish normal forms of sym-
plectic matrices. Its corollary is well-known and usually proven by induction.

Recall that dimC Eλ = mλ is the algebraic multiplicity of λ, that is the

exponent of t − λ in the factorisation of det(A − t Id).
Note that if v = u+iu′ is in Ker(A−λ Id)j with u and u′ in V then v = u−iu′
is in Ker(A − λ Id)j so that Eλ ⊕ Eλ is the complexification of a real subspace
of V . From this remark and lemma 15 the space

W[λ] := Eλ ⊕ E λ1 ⊕ Eλ ⊕ E 1 (15)
λ

is again the complexification of a real and symplectic subspace V[λ] and

R2n = V[λ1 ] ⊕ V[λ2 ] ⊕ . . . ⊕ V[λK ] . (16)

We have again
ρ(A) = ρ(A[λ1 ] ) · ρ(A[λ2 ] ) · . . . · ρ(A[λK ] ) (17)
where A[λi ] is the restriction of A to V[λi ] .
Every symplectic matrix can be approached as closely as we want by a
semisimple symplectic matrix. To be complete we give a proof of this property
in the next section. By the continuity hypothesis, the map ρ is thus necessarilly
defined as follows:

Theorem 16 ([12, 3]) Let A ∈ Sp(R2n , Ω). We consider the eigenvalues {λi }
of A. For an eigenvalue λ = eiϕ ∈ S 1 \ {±1}, we consider the number m+ (λ) of

14
positive eigenvalues of the symmetric non degenerate 2-form Q defined on the
generalized eigenspace Eλ by
Q : Eλ × Eλ → R : (z, z ′ ) 7→ Q(z, z ′ ) := Im Ω0 (z, z ′ ).
Then Y +
1 − 1
(λ)
ρ(A) = (−1) 2 m λ2m (18)
λ∈S 1 \{±1}

where m− is the sum of the algebraic multiplicities mλ = dimC Eλ of the real

negative eigenvalues.
Proof: We have seen that it is necessarilly of that form. The first term in
1 −
formula (18), (−1) 2 m , corresponds to the conditions (9), (10) and (11). The
second term comes from the fact that if λ = eiϕ and of A[λ] is the restriction
1
of A to V[λ] then ρ(A[λ] ) = e 2 iϕ Sign Q . In formula (18), we count one term for
λ and one term for λ. Remark that Im Ω0 (z, z ′ ) = −Im Ω0 (z, z ′ ) so that
the number of negative eigenvalues of Q on Eλ × Eλ is equal to the number of
positive eigenvalues of Q on Eλ × Eλ . Thus the signature of Q on Eλ × Eλ is
1 1 +
1
m+ (λ)
equal to m+ (λ) − m+ (λ). Hence e 2 iϕ Sign Q = λ 2 m (λ) λ 2 .
This map ρ satisfies clearly the hypotheses of continuity, invariance, normal-
ization and multiplicativity. It is an extension of the map determinant on U (n)
because every element of U (n) is semisimple. 

Definition 17 Let λ be an eigenvalue of A ∈ Sp(R2n , Ω0 ) with |λ| = 1. When

λ = 1( or − 1), the corresponding generalized eigenspace is of even dimension
2m1 (or 2m−1 ) and we count m1 times the eigenvalue 1 (or m−1 times the
eigenvalue −1) as Krein positive. For the pair of eigenvalues λ = eiϕ et λ̄ = e−iϕ ,
when the quadratic form
Q : Eλ × Eλ −→ C : (v, w) 7−→ Im Ω(v, w̄)
is of signature (2r, 2s) we count r times the eigenvalue eiϕ and s times the
eigenvalue e−iϕ as Krein positive.

Remark 18 ([12]) Another expression of ρ on a matrix A ∈ Sp(R2n , Ω0 ) is

obtained as follows: we consider the eigenvalues of A, λ1 , . . . , λ2n , repeated
accordingly to their algebraic multiplicities. We say that an eigenvalue λi is of
the first kind if |λi | < 1 or if |λi | = 1 and if it is “positive in the sense of Krein”.
Then
Y λi
ρ(A) = . (19)
|λi |
λi of first kind

Proposition 19 The map ρ is not a group homomorphism but we always have

N
ρ(AN ) = ρ(A) ∀A ∈ Sp(R2n , Ω0 ) and ∀N ∈ Z. (20)
This results directly from the construction of ρ and from the fact that the
eigenvalues of AN are equal to the N th powers of the eigenvalues of A.

15