If λ = reiφ ∈/ R we consider the symplectic basis {e1 , . . . , ep+1 , f1 , . . .
, fp+1 } of
Eλv ⊕E w1 as above and the conjugate symplectic basis {e1 , . . . , ep+1 , f1 , . . . , fp+1 }
λ
of Eλv ⊕E w1 . Writing ej = √1 (uj −ivj ) and fj = √1 (wj +ixj ) for all 1 ≤ j ≤ p+1
λ
2 2
with the vectors uj , vj , wj , xj in the real vector space V , we get a symplectic
basis {u1 , . . . , up+1 , v1 , . . . , vp+1 , w1 , . . . , wp+1 , x1 , . . . , xp+1 } of the real subspace
of V whose complexification is Eλv ⊕ E w1 ⊕ Eλv ⊕ E w1 . In this basis, the matrix
λ λ
associated to A is
!
JR reiφ , 2(p + 1) 0 τ
−1
0 JR reiφ , 2(p + 1)
where JR (reiφ , 2m) is the 2m × 2m matrix written in terms of 2 × 2 matrices as
R(reiφ ) Id 0 0 ... 0 0
0 R(re iφ
) Id 0 ... 0 0
0 0 R(reiφ ) Id ... 0 0
.. .. .. .. .. .. ..
. . . . . . . (30)
0 0 0 . . . R(re )iφ
Id 0
0 0 0 ... 0 R(re ) iφ
Id
iφ
0 0 0 ... 0 0 R(re )
r cos φ −r sin φ
with R(reiφ ) = . By induction, we get
r sin φ r cos φ
Theorem 23 (Normal form for A|V[λ] for λ ∈ / S 1 .) Let λ ∈
/ S 1 be an eigen-
value of A. Denote k := dimC Ker(A − λ Id) (on V C ) and p the smallest integer
so that (A − λ Id)p+1 is identically zero on the generalized eigenspace Eλ .
If λ 6= ±1 is a real eigenvalue of A, there exists a symplectic basis of V[λ] in
which the matrix associated to the restriction of A to V[λ] is a symplectic direct
sum of k matrices of the form
J(λ, pj + 1) 0 τ
0 J(λ, pj + 1)−1
/ (S 1 ∪ R) is a complex eigenvalue of
with p = p1 ≥ p2 ≥ · · · ≥ pk . If λ = reiφ ∈
A, there exists a symplectic basis of V[λ] in which the matrix associated to the
restriction of A to V[λ] is a symplectic direct sum of k matrices of the form
!
JR reiφ , 2(pj + 1) 0
−1 τ
0 JR reiφ , 2(pj + 1)
with p = p1 ≥ p2 ≥ · · · ≥ pk .
21
3.2.2 Case 2: A|V[λ] for λ = ±1.
In this situation [λ] = {λ} and V[λ] is the generalized real eigenspace of eigen-
value λ, still denoted –with a slight abuse of notation– Eλ . We consider
Qe : Eλ /Ker(A − λ Id)p × Eλ /Ker(A − λ Id)p → R the non degenerate form
e
defined by Q([v], [w]) = Ω (A − λ Id)p v, w . We see directly from equation (27)
that Qe is symmetric if p is odd and antisymmetric if p is even.
If p = 2k − 1 is odd, we choose v ∈ Eλ such that
e
Q([v], [v]) = Ω (A − λ Id)p v, v 6= 0
and consider the smallest subspace Eλv of Eλ stable by A and containing v; it is
generated by
e1 := (A − λ Id)p v, . . . , ei := (A − λ Id)p+1−i v, . . . , ep+1=2k := v .
We have
• Ω(ei , ej ) = 0 if i + j ≤ p + 1(= 2k) by equation (28) because
Ω (A − λ Id)p+1−i v, (A − λ Id)p+1−j v = 0 if p + 1 − j + p + 1 − i ≥ p + 1
• Ω(ei , ep+2−i ) 6= 0; by equation(26) because
Ω (A − λ Id)p+1−i v, (A − λ Id)i−1 v = (−1)i−1 Ω (A − λ Id)p v, v 6= 0 by
the choice of v.
Hence Eλv is a symplecticsubspace because,
in the basis defined by the ei ’s, Ω
0 . ∗
has the triangular form .. and has a non-zero determinant.
∗ ∗
We can choose v so that Ω (A − λ Id)k v, (A − λ Id)k−1 v = d = ±1 by rescaling
the vector. One may further assume that
Ti,j (v) := Ω (A − λ Id)i v, (A − λ Id)j v = 0 for all 0 ≤ i, j ≤ k − 1.
Indeed, by formula (25) we have Ti,j (v) = −λTi+1,j (v)−Ti+1,j−1 (v), Ti,i (v) = 0
and we proceed by decreasing induction (observing that Ti,j (v) = −Tj,i (v)) as
follows:
• if Tk−2,k−1 (v) = α1 , we replace v by v − α2d1 (A − λ Id)2 v; it spans the same
subspace and the quantities Ti,j (v) do not vary for i + j ≥ 2k − 1 but now
Tk−2,k−1 (v) = 0, hence Tk−3,k−1 (v) = −λTk−2,k−1 (v) − Tk−2,k−2 (v) = 0;
• if Tk−3,k−2 (v) = α2 , we replace v by v + α2d2 (A − λ Id)4 v; it spans the
same subspace and the quantities Ti,j (v) do not vary for i + j ≥ 2k − 1, for
k −2 ≤ i ≤ j ≤ k −1 and for i = k −3, j = k −1; but now Tk−3,k−2 (v) = 0,
hence also Tk−4,k−2 (v) = 0, Tk−4,k−1 (v) = 0;
22
