form  

J(λ, k) 0 0 C
 0 J(λ, k) C
0 
 τ 
 0 0 J(λ, k)−1 0 
−1

τ
0 0 0 J(λ, k)
we define a transformation S given in the basis considered by the matrix
 
D 0 0 0
 0 D 0 0 
 
 0 0 D−1 0  where D = diag(1 + ǫ1 , . . . , 1 + ǫk ).
−1
0 0 0 D
It induces a real symplectic transformation of the corresponding subspace of
V[λ] . The transformation AS on that subspace is semisimple because it admits
the quadruples of eigenvalues associated to (1 + ǫ1 )λ, . . . , (1 + ǫk )λ.

On a subspace of W[λ] spanned by a symplectic basis {u1 , . . . , uk , v1 , . . . , vk+1 ,

′ ′
vp+1 , . . . vk+2 , u′p+1 , . . . u′k+1 } with vj = uj for j ≤ k, vj′ = u′j for j ≥ k + 2 and
′
uk+1 = ivk+1 in which the matrix associated to A is of the form


J(λ, k) 0  0  D d
 D 
 0 J(λ, k + 1) 0 
 d ′ 


 −1 τ 
0 0 J(λ, k) 0

−1 τ
0 0 0 J(λ, k + 1)
we define a transformation S given in that basis by the matrix
 
D 0 0 0 0 0
 0 D 0 0 0 0 
 
 0 0 eiǫ 0 0 0 
  where D = diag(1 + ǫ1 , . . . , 1 + ǫk ).
 0 0 0 D−1 0 0 
 −1 
 0 0 0 0 D 0 
0 0 0 0 0 e−iǫ
It induces a real symplectic transformation of the corresponding subspace of
V[λ] .
The transformation AS is semisimple on that subspace because it admits the
quadruples of eigenvalues associated to (1 + ǫ1 )λ, . . . , (1 + ǫk )λ and the double
associated to eiǫ λ. 

4 The set Sp⋆(R2n , Ω0)

We would like to associate an integer to a path in the symplectic group starting
from the identity and ending at an element which does not admit 1 as eigenvalue.
This paragraph is a development of the page 1317 of [12].

31
Definition 27 We denote by Sp⋆ (R2n , Ω0 ) the set of symplectic matrices which
do not have 1 as eigenvalue:

Sp⋆ (R2n , Ω0 ) := A ∈ Sp(R2n , Ω0 )| det(A − Id) 6= 0 .

Remark
 28 The set Sp⋆ (R2 , Ω0 ) has two
 connected components:
A ∈ Sp(R2 , Ω0 )| det(A − Id) > 0 and A ∈ Sp(R2 , Ω0 )| det(A − Id) < 0 .
Indeed Sp(R2 , Ω0 ) = Sl(2, R) and we have:
  !
a + d −b + c
det − Id = (a + d − 1)(a − d − 1) + (b − c)(b + c)
b+c a−d
= 1 + a2 + b2 − c2 − d2 − 2a = 2(1 − a)

thus Sp⋆ (R2 , Ω0 ) is the disjoint union of the two connected components.
  
⋆ 2 a + d −b + c 2 2 2 2
Sp (R , Ω0 ) = | a < 1 and a + b − c − d = 1
b+c a−d
  
a + d −b + c 2 2 2 2
∪ | a > 1 and a + b − c − d = 1
b+c a−d

The property stated above generalizes for all dimensions:

Theorem 29 ([12]) The group Sp(R2n , Ω0 ) is connected. The subset Sp∗ (R2n , Ω0 )
has two connected components defined by

Sp± (R2n , Ω0 ) = A ∈ Sp(R2n , Ω0 ) | ± det(A − Id) > 0

and every loop in Sp± is contractible in Sp(R2n , Ω0 ).

Remark 30 In the connected component where det(A − Id) > 0, we have the
matrix
 !n
+ −1 0
W := − Id and ρ(− Id) = ρ = (−1)n ; (35)
0 −1

In the component where det(A − Id) < 0, we have the diagonal matrix
1
W− := diag(2, −1, . . . , −1, , −1, . . . , −1) (36)
2
   !n−1
2 0 −1 0
and ρ(W − ) = ρ ρ = (−1)n−1 . (37)
0 12 0 −1

Proof of the connectedness: We have seen in paragraph 3.1 that every

element A ∈ Sp(R2n , Ω0 ) may be connected to an element A′ ∈ Sp(R2n , Ω0 ) as
close as we want and with all eigenvalues distinct.

32
 If A ∈ Sp± (R2n , Ω0 ), the element A′ may be chosen in the same set since the
map A 7→ det(A−Id) is a continuous map on Sp(R2n , Ω0 ). We now show that ev-
ery element A′ ∈ Sp+ (R2n , Ω0 ) (and similarly every element A′ ∈ Sp− (R2n , Ω0 ))
with distinct eigenvalues may be connected to an element with all eigenvalues
equal to −1 (respectively to an element with a pair of eigenvalues (2, 12 ) of
multiplicity 1 and all other eigenvalues equals to −1).
Indeed, we consider a basis (of C2n ) in which A′ is diagonal and we modify
′
A on the symplectic subspace

V[λ] ⊗R C = Eλ ⊕ E λ1 ⊕ Eλ̄ ⊕ E 1
λ̄

1 1
by considering λ(t), λ(t) , λ̄(t), λ̄(t) where λ : [0, 1] → C : t 7→ λ(t) is defined by:

λ(t) = (1 − t)λ − t if λ ∈/ S1 ∪ R (38)

1
choosing between λ and so that |λ| < 1
λ
λ(t) = eitπ ei(1−t)ϕ if λ = eiϕ ∈ S 1 \ {±1} (39)
choosing between λ and λ so that 0 < ϕ < π
λ(t) = (1 − t)λ − t if λ is real negative and λ > −1. (40)
 1
For the real positive eigenvalues we consider
n two pairs of eigenvalues
o a, a , b, 1b
2 2
with a > 1 and b > 1. We bring them to a+b a+b
2 , a+b , 2 , a+b without passing
 
t

t 1 t

t 1
trough 1 via 1 − 2 a + 2 b, 1− t a+ t b , 1 − 2 b + 2 a, 1− t b+ t a . Then we
( 2) 2 ( 2) 2
n
 
−iθ  2  iθ o
a+b iθ 2 −iθ a+b
go to e , a+b e , 2 e , a+b e and finally we proceed as
 1 2 1 1
for λ, λ , λ̄, λ̄ with λ ∈ / S ∪ R.
+ 2n
A semisimple element of Sp  (R , Ω0 ) must  have an even number of real
a−1 0
positive eigenvalues since det 1 < 0 if a > 1.
0 a −1
The elements considered above exhaust all possibilities for Sp(R2n , Ω0 ). We
have thus connected every element of Sp+ (R2n , Ω0 ) to a semisimple element with
all eigenvalues equal to −1, i.e. to W + = − Id . This implies that Sp+ (R2n , Ω0 )
is connected.
Note that we have shown also that every matrix in Sp(R2n , Ω0 ) may be con-
nected to − Id since we can connect two eigenvalues 1 on a  symplectic space of di-
cos tπ − sin tπ
mension 2, to two eigenvalues equals to −1 by a rotation .
sin tπ cos tπ
We deduce that Sp(R2n , Ω0 ) is connected.
In a similar way, since a pair of positive real eigenvalues {b, 1b } may be
connected to {2, 12 } without going through 1, we have connected every element
of Sp− (R2n , Ω0 ) to a semisimple element A′′ whose eigenvalues are either 2 and
1
2 (each with multiplicity 1) or −1. We have thus

A′′ = kW − k −1 where k ∈ Sp(R2n , Ω0 ).

33
Since Sp(R2n , Ω0 ) is connected, we can find a path k(t) in the group Sp(R2n , Ω0 )
which links k to Id . So k(t)W − k −1 (t) will link A′′ to W − staying in Sp− (R2n , Ω0 )
as the eigenvalues remained unchanged. 

Remark 31 We have shown above that an element A ∈ Sp∗ (R2n , Ω0 ) is in

Sp+ (R2n , Ω0 ) if and only if the number of real positive eigenvalues is a multiple
of four.

Proof of the contractibility of every loop in Sp∗ (R2n , Ω0 ):

Let γ : [0, 1] → Sp± (R2n , Ω0 ) : t 7→ γ(t) be a loop; i.e a continuous map such
that γ(0) = γ(1).
Recall that a loop is contractible in Sp(R2n , Ω0 ) if and only if its image by
ρ is contractible. Indeed:

ρ : Sp(R2n , Ω0 ) −→ S 1

is a continuous map wich induces an isomorphism on the fundamental groups,

because detC : U (n) → S 1 induces an isomorphism on the fundamental groups
(see section 1.1).
Recall (19) that the value of ρ on a matrix A ∈ Sp(R2n , Ω0 ) may be obtained
in the following way: we consider the eigenvalues λ1 , . . . , λ2n repeated according
to their algebraic multiplicity and we look at the eigenvalues λ1 , . . . , λn of first
kind (thus all eigenvalues such that |λi | < 1, half of those who equal 1, half of
those who equal −1, and r times the value λ = eiϕ 6= ±1 if the quadratic form
Q : Eλ × Eλ −→ C : (v, w) 7−→ Im Ω(v, w̄) has signature (2r, 2s)). Then
Y λi
ρ(A) =
|λi |
λi of first kind

The map

Sp(R2n , Ω0 ) → Λ(n) := Cn /permutations of the elements,

which sends a matrix A on the set (λ1 (A), . . . , λn (A)) of its eigenvalues of the
first kind is continuous. More precisely, given a path γ : [0, 1] → Sp(R2n , Ω0 ), we
can choose continuous functions Λ1 , . . . , Λn : [0, 1] → C such that Λ1 (t), . . . , Λn (t)
are the eigenvalues of first kind of γ(t).
If the path γ takes its values in Sp± (R2n , Ω0 ) there is an even number (re-
Λi
spectively odd) of the |Λ i|
which equal 1 for every value of t. We define functions

Λi (t)
αi : [0, 1] −→ [0, 2π] (i = 1 . . . n) such that eiαi (t) = .
|Λi (t)|
Λi (t)
They are uniquely determined if |Λi (t)| 6= 1. Since there are 2k (resp 2k + 1)
Λi (t)
numbers |Λi (t)| = 1, we define for k (resp k +1) of them the value αi (t) = 2π and

34
for the other k of them the value αi (t) = 0; we can thus make the functions αi
continuous. It is indeed continuous because if a quadruple λ, λ1 , λ, λ1 degenerates
into two pairs of real positive eigenvalues, the two eigenvalues of first kind have
an argument equal to ϕ and to (2π − ϕ) tending one to zero and the other to
2π.
If we start from a loop γ which lies in Sp± (R2n , Ω0 ), its image by each αi is
a loop ⊂ [0, 2π] and is thus [Link]
Hence the map [0, 1] → R : t 7→ i=1 αi
(t) is P
contractible and thus the
n
image by ρ of the loop [0, 1] → S 1 : t 7→ ρ γ(t) = ei i=1 αi (t) is contractible in
S1.
Thus the loop γ is contractible in Sp(R2n , Ω0 ). 

5 Conley-Zehnder index for a path of matrices

in SP(n)
Definition 32 ([11, 12]) We consider the set SP(n) of continuous paths of
matrices in Sp(R2n , Ω0 ) linking the matrix Id to a matrix in Sp⋆ (R2n , Ω0 ) :
( )
2n ψ(0) = Id and
SP(n) := ψ : [0, 1] → Sp(R , Ω0 ) .
1 is not an eigenvalue of ψ(1)

From the decomposition of Sp⋆ (R2n , Ω0 ) into two connected components, every
path ψ ∈ SP(n) may be extended to a continuous path ψe : [0, 2] → Sp(R2n , Ω0 )
such that ψe coincides with ψ on the interval [0, 1] , such that ψ(s)
e ∈ Sp⋆ (R2n , Ω0 )
for all s > 1 and such that the path ends either in ψ(2) e = − Id = W + either
e 1
in ψ(2) = W := diag(2, −1, . . . , −1, 2 , −1, . . . , −1). Remark that ρ(W − ) =
−

(−1)n−1 and ρ(− Id) = (−1)n ; thus ρ2 ◦ ψe : [0, 2] → S1 is a loop.

Definition 33 ([11, 12]) The Conley-Zehnder index of a path ψ ∈ SP(n) is

the integer given by the degree of the map ρ2 ◦ ψ̃ : [0, 2] → S1 :

e
µCZ (ψ) := deg(ρ2 ◦ ψ) (41)

where ψe : [0, 2] → Sp(R2n , Ω0 ) is a continuous extension of the path ψ such that

• ψe coincides with ψ on the interval [0, 1] ;

e ∈ Sp⋆ (R2n , Ω0 ) for all s > 1;
• ψ(s)


e
• ψ(2) e
= − Id =: W + or ψ(2) = W − := diag 2, −1, . . . , −1, 21 , −1, . . . , −1 .

Lemma 34 ([11, 12]) The Conley-Zehnder index of a path ψ ∈ SP(n) does

not depend on the choice of the chosen extension ψ̃.

35