Journal of Geometry and Physics 62 (2012) 2044–2058

Contents lists available at SciVerse ScienceDirect

Journal of Geometry and Physics

journal homepage: www.elsevier.com/locate/jgp

On the universal covering group of the real symplectic group

John Rawnsley
Mathematics Institute, University of Warwick, Coventry CV4 7AL, UK

article info abstract

Article history: A model for the universal covering group of the symplectic group as a Lie group, and some
Received 5 January 2012 calculations based on the model, as well as defining a similar model for the Lagrangian
Accepted 30 May 2012 Grassmannian and relating our construction to the Maslov Index.
Available online 7 June 2012
© 2012 Elsevier B.V. All rights reserved.
MSC:
53D05
20F36
53C29

Keywords:
Symplectic group
Universal cover
Braid group
Maslov index

1. Introduction

The real symplectic group in 2n dimensions, Sp(2n, R), has a number of non-trivial central extensions such as the
metaplectic double covering Mp(2n, R), an extension by Z2 , the circle extension Mpc (2n, R) as the automorphism group
of the Heisenberg group, and the universal covering group Sp  (2n, R) an extension by Z. These extensions do not have
faithful finite dimensional representations, so there are no nice models for them as Lie groups of matrices, and this can
make calculations with them cumbersome. Paths in Sp(2n, R) are not easily lifted to an extension unless they lie in some
subgroup whose inverse image in the extended group is easy to describe.
In [1], an explicit model for the circle extension of the symplectic group Sp(2n, R) (the Mpc group) was used to facilitate
the definition of a theory of symplectic spinors on any symplectic manifold and to extend the Kostant theory of metaplectic
half-forms to this case. The group manifold can be described explicitly as a hypersurface in Sp(2n, R) × C∗ and the group
multiplication given by a single global formula. In [2], these ideas have been used to define Dirac operators on any symplectic
manifold and extend the theory of Habermann [3] to this case.
The method used can be adapted to other central extensions. We do this here for the universal covering group Sp  (2n, R).
We show how to make models for the universal covering manifold using suitable maps to the circle and to write the group
multiplication in terms of an associated cocycle. We then show how to construct a particularly nice explicit circle map using
the ideas from [1] and obtain the formula for its cocycle.
As an application we look at the universal covering group of Sp(2, R) = SL(2, R) and the inverse image of SL(2, Z) in it.
It is a theorem of Milnor that this extension of SL(2, Z) is isomorphic to the braid group on three strands. We use our explicit
formulae for the multiplication to write down generators and show that they satisfy the braid relation.
Finally, we show how the same methods can be used to construct the universal covering manifold of the Lagrangian
Grassmannian and the Maslov Index.

E-mail addresses: [email protected], [email protected].

0393-0440/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.geomphys.2012.05.009
 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058 2045

2. Basic construction

In this section, we show how a smooth circle function on a connected Lie group G with π1 (G, e) = Z can be used to
construct a manifold underlying the covering group.

Definition 1. Let G be a connected Lie group with π1 (G, e) = Z. A circle function on G is a smooth map ϕ : G → S 1 which
induces an isomorphism of the fundamental groups. It is said to be normalised if it satisfies

(i) ϕ(e) = 1;
(ii) ϕ(g −1 ) = ϕ(g )−1 .

We shall regard S 1 as the unit circle in the complex plane, and ϕ as a complex valued function.

Lemma 2. If ϕ is a circle function then there is a unique smooth map 

ϕ : G → S 1 with
ϕ (g )2 = ϕ(g )/ϕ(g −1 ),
 ϕ (e) = 1 ,

ϕ is a normalised circle function.
and 

Proof. By the connectedness of G there can be at most one such map  ϕ.

I claim that the map Ψ : G → S 1 given by Ψ (g ) = ϕ(g )ϕ(g −1 ) is trivial on π1 (G, e). This follows since if γ is a loop at e
in G then Ψ∗ [γ ] = [Ψ ◦ γ ] = [ϕ ◦ γ ] + [ϕ ◦ γ −1 ] = ϕ∗ [γ ] + ϕ∗ [γ −1 ] = ϕ∗ ([γ ] − [γ ]) = 0 since the multiplication in the
fundamental group of a topological group is induced by pointwise multiplication of loops. Moreover, Ψ (e) = ϕ(e)2 thus Ψ
has a unique smooth square root ψ with ψ(e) = ϕ(e). We set  ϕ (g ) = ϕ(g )ψ(g )−1 . Property (i) is immediate in view of the
normalisation of ψ . Further, ϕ (g )2
ϕ (g −1 )2 = ϕ(g )/ϕ(g −1 )ϕ(g −1 )/ϕ(g ) = 1. Thus ϕ (g )
ϕ (g −1 ) = ±1 and (ii) follows since
G is connected and  ϕ (e)
ϕ (e ) = 1. Since Ψ∗ is trivial on π1 so is ψ∗ and hence 
−1
ϕ∗ is an isomorphism. 

Lemma 3. If ϕ is a normalised circle function for G, then there is a unique smooth function η : G × G → R such that

ϕ(g1 g2 ) = ϕ(g1 )ϕ(g2 ) ei η(g1 ,g2 )

with η(e, e) = 0. Further

(i) η(g , e) = η(e, g ) = η(g , g −1 ) = 0, for all g ∈ G;

(ii) η(g1 , g2 ) + η(g1 g2 , g3 ) = η(g1 , g2 g3 ) + η(g2 , g3 ) for all g1 , g2 , g3 ∈ G.

Proof. The existence of η is a consequence of the following. If ϕ : G → H is a continuous map (not necessarily a
homomorphism) of connected topological groups with ϕ(eG ) = eH and Φ : G × G → H is defined by Φ (g1 , g2 ) =
ϕ(g1 g2 )ϕ(g1 )−1 ϕ(g2 )−1 , then Φ∗ : π1 (G × G, (eG , eG )) → π1 (H , eH ) is trivial. From this it follows that Φ has a smooth
logarithm which is unique when its value is specified at one point since G, and hence G × G, is connected.
(i) and (ii) follow by continuity arguments again based on the connectivity of G. For example, exp iη(g , e) =
ϕ(ge)ϕ(g )−1 ϕ(e)−1 = 1 so η(g , e) ∈ 2π Z and thus must be constant. It vanishes for g = e. 

Definition 4. If ϕ is a normalised circle function for the Lie group G set

G = (g , c ) ∈ G × R | ϕ(g ) = ei c .
 


It is easy to see that 1 is a regular value of the map G × R ∋ (g , c ) → ϕ(g )e−i c ∈ S 1 and hence that 
G is a smooth manifold
of the same dimension as G.

Proposition 5. Let 
G be defined as above and set

(g1 , c1 ) · (g2 , c2 ) = (g1 g2 , c1 + c2 + η(g1 , g2 ))

G is a Lie group with identity (eG , 0) and inverse (g , c )−1 = (g −1 , −c ). If σ : 
then  G → G is given by σ (g , c ) = g then σ is a
surjective homomorphism with kernel {(1, 2π k) | k ∈ Z}.

Proof. Associativity and the group properties follow from Lemma 3. σ is obviously smooth and clearly a homomorphism. 

It follows from this proposition that σ is a homomorphism of Lie groups with discrete kernel, and so induces an
isomorphism of Lie algebras.

Theorem 6. σ : 
G → G is the universal covering group of G.
2046 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058

Proof. Let γ : [0, 1] → G be a loop in G, set z (t ) = ϕ(γ (t )) and let k ∈ Z be the winding number of the origin of z (t ) as a
contour in C. This means that we can lift z (t ) = ei c (t ) with c (0) = 0 and then c (1) = 2π k. Since ϕ induces an isomorphism
on the fundamental group, k is also the homotopy class of γ . If (γ , c ) is a loop in  G at (1, 0) then ϕ(γ (t )) = ei c (t ) with
c (1) = c (0) = 0. It follows that k = 0 and hence that γ is contractible in G. So there is a curve of loops δs in G with δ0 = γ
and δ1 ≡ 1. Then ϕ(δs (t )) lifts to R so we can write ϕ(δs (t )) = ei cs (t ) with cs (0) = cs (1) = 0 for all s and c1 (t ) ≡ 0. (δs , cs ) is
then a homotopy of (γ , c ) to a constant loop in 
G. Since (γ , c ) was arbitrary, it follows that  G is simply connected and hence
that it is the universal covering group of G. 

Remark 7. We have a converse to this result. If  G → G is the universal covering group of a Lie group G and π1 (G, eG ) = Z,
G is an extension of G by Z. Take a good cover Uα of G and choose sections sα : Uα → 
then  G. Then on overlaps we have
sβ = sα .nαβ . Viewed as real valued functions, the 1-cocycle nαβ is a coboundary, so there are smooth functions fα on Uα
with fβ − fα = nαβ . If we set ϕ = exp 2π ifα then ϕ is globally defined and smooth on G. Moreover, if we form the cover as
in Definition 4, then it is left as an exercise to show that (g , c ) → sα (g ).(c − fα (g )), when g ∈ Uα , gives a globally defined
isomorphism of Lie groups.

Example 8. As a first example we take G = U(n) then ϕ(g ) = det g is a normalised circle function. In this case η ≡ 0 and

U (n) is a subgroup of U(n) × R.

Lemma 9. U (n) is isomorphic to SU(n) × R.
Proof. Define a map

µ : SU(n) × R → U
 (n) : (k, c ) → (ei c k, nc ),
which is well defined since det(ei c k) = eni c . Moreover µ is a homomorphism since ϕ is a homomorphism and η ≡ 0 giving
µ(k1 , c1 )µ(k2 , c2 ) = (ei c1 k1 , nc1 )(ei c2 k2 , nc2 )
= (ei(c1 +c2 ) k1 k2 , nc1 + nc2 )
= µ(k1 k2 , c1 + c2 ).
The map µ is onto, for if k ∈ U(n) with det k = ei c then det(e−i c /n k) = 1. So the image of (e−i c /n k, c /n) ∈ SU(n) × R under
µ is (k, c ) ∈ U
 (n). The kernel of µ is
(k, c ) ∈ SU(n) × R | ei c k = 1, nc = 0 = {(1, 0)} ,
 

which proves the lemma. 

Example 10. The symplectic group is semisimple, so there are no homomorphisms Sp(2n, R) → S 1 . However, U (n) is a
maximal compact subgroup and so the same fundamental group. Suppose we have a smooth map ϕ : Sp(2n, R) → S 1 such
that ϕ(g −1 ) = ϕ(g )−1 , ϕ(1) = 1, and whose restriction to U (n) is det, then ϕ is a normalised circle function for Sp(2n, R)
and so such a map ϕ will induce an isomorphism on the fundamental group.
In fact we could take any retraction ρ : Sp(2n, R) → U(n) and set ϕ = det ◦ρ to get a circle function and then normalise
it. But the result in this generality might not be all that easy to work with.
In order to have a useable model for Sp  (2n, R) it is necessary to have a nice formula for the multiplication, and this
depends on being able to compute the corresponding cocycle η. We shall see in what follows that there is a good choice of
ϕ which leads to an explicit formula for η. After defining this we shall make some computations with our formulae, to show
that they are indeed ‘useable’.

3. Calculation of a symplectic circle function ϕ

We move to a slightly more abstract setting, taking a general symplectic vector space as our starting point. Let (V , Ω ) be
a 2n-dimensional real symplectic vector space. Fix J ∈ End(V ) such that J 2 = −1, Ω (Ju, J v) = Ω (u, v) for every u, v ∈ V
and Ω (v, J v) > 0 if v ̸= 0. Let
Sp(V , Ω ) = {g ∈ End(V ) | Ω (g v, g w) = Ω (v, w), ∀v, w ∈ V } ;
Gl(V , J ) = {g ∈ Gl(V ) | gJ = Jg } ;
U(V , Ω , J ) = {g ∈ Sp(V , Ω ) | gJ = Jg } = Sp(V , Ω ) ∩ Gl(V , J ).
View V as a complex vector space using J, then Sp(V , Ω ) is isomorphic to Sp(2n, R), U(V , Ω , J ) to U(n) and Gl(V , J ) to
Gl(n, C).
Given g ∈ Sp(V , Ω ) it can be written uniquely as a sum of a J-linear and J-antilinear part, namely,
g = Cg + Dg ,
where Cg = 1
2
(g − JgJ ) and Dg = 12 (g + JgJ ).
 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058 2047

Lemma 11. Cg ∈ Gl(V , J ). If g ∈ U(V , Ω , J ), then Cg = g.

Proof. Suppose that Cg v = 0. Then

0 = 4Ω (Cg v, JCg v) = 2Ω (v, J v) + Ω (g v, Jg v) + Ω (gJ v, JgJ v).
But each term in the right hand side of the above equation is positive definite and so v = 0. 

The definition of the circle function ϕ is now easy, namely,

det Cg
ϕ(g ) = .
| det Cg |

4. Calculation of the cocycle η

We now find a formula for the cocycle η whose existence is guaranteed by Lemma 3. First we make some preliminary
calculations.
Put Zg = Cg−1 Dg . Then g = Cg (1 + Zg ). Zg is C-antilinear. We have g −1 = Cg −1 (1 + Zg −1 ). Equating C-linear and
C-antilinear parts in
1 = g −1 g = Cg −1 (1 + Zg −1 )Cg (1 + Zg )
gives
1 = Cg −1 (Cg + Zg −1 Cg Zg ) and 0 = Cg −1 (Zg −1 Cg + Cg Zg ).
Hence
Zg −1 = −Cg Zg Cg−1 and 1 = Cg −1 Cg (1 − Zg2 ),

since Cg −1 is invertible. Thus 1 − Zg2 is invertible with (1 − Zg2 )−1 = Cg −1 Cg . We can also decompose a product
Cg1 g2 (1 + Zg1 g2 ) = Cg1 (1 + Zg1 )Cg2 (1 + Zg2 )
as
Cg1 g2 = Cg1 (Cg2 + Zg1 Cg2 Zg2 ) = Cg1 (1 − Zg1 Zg −1 )Cg2
2

and
Zg1 g2 = Cg−1 (1 − Zg1 Zg −1 )−1 (Zg1 Cg2 + Cg2 Zg2 )
2 2

= Cg−1 (1 − Zg1 Zg −1 )−1 (Zg1 − Zg −1 )Cg2 .

2 2 2

To define the cocycle η we need to see that the function

det(1 − Zg1 Zg −1 )
(g1 , g2 ) → 2

| det(1 − Zg1 Zg −1 )|
2

has a smooth logarithm. To see this, we determine that the set where Zg lives is essentially the Siegel domain.
Introduce a Hermitian structure on (V , J ) as a C vector space by defining
⟨v, w⟩ = Ω (v, J w) − i Ω (v, w).
Then
⟨J v, w⟩ = i ⟨v, w⟩ = −⟨v, J w⟩ and ⟨v, v⟩ = Ω (v, J v) > 0, for v ̸= 0.
We calculate
2 ⟨Cg∗ v, w⟩ = 2 ⟨v, Cg w⟩
= Ω (v, Jg w) + Ω (v, gJ w) − i Ω (v, g w) + i Ω (v, JgJ w)
= −Ω (g −1 J v, w) + Ω (g −1 v, J w) − i Ω (g −1 v, w) − i Ω (g −1 J v, J w)
= ⟨g −1 v, w⟩ − i ⟨g −1 J v, w⟩ = ⟨g −1 v, w⟩ − ⟨Jg −1 J v, w⟩
= 2 ⟨Cg −1 v, w⟩.
Therefore
Cg∗ = Cg −1 ,
2048 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058

which implies
det Cg −1 det Cg∗
ϕ(g −1 ) = = = ϕ(g ).
| det Cg −1 | | det Cg∗ |
Moreover,
1 − Zg2 = (Cg −1 Cg )−1 = (Cg∗ Cg )−1 ,
which is positive definite. Since Zg is antilinear and ⟨v, w⟩ is antilinear in w , the function (v, w) → ⟨v, Zg w⟩ is complex
bilinear. We claim that it is symmetric.

Lemma 12. For every v, w ∈ V , we have ⟨v, Zg w⟩ = ⟨w, Zg v⟩.

Proof. Now
Ω (Dg v, w) = Ω (v, Dg −1 w) = −Ω (Dg −1 w, v);
and Ω (Dg , J w) = −Ω (Dg −1 w, J v). Hence ⟨Dg v, w⟩ = −⟨Dg −1 w, v⟩ and so

⟨Zg v, Cg∗ w⟩ = ⟨Cg Zg v, w⟩ = −⟨Cg −1 Zg −1 w, v⟩ = ⟨Zg Cg∗ w, v⟩. 

So Zg has the three properties:

(i) Zg is C-antilinear;
(ii) (v, w) → ⟨v, Zg w⟩ is symmetric;
(iii) 1 − Zg2 is self-adjoint and is positive definite.
Let B(V , Ω , J ) be the Siegel domain consisting of Z ∈ EndR (V ) such that
ZJ = −JZ , ⟨v, Z w⟩ = ⟨w, Z v⟩, and 1 − Z2 is positive definite.
Then we have a map
Sp(V , Ω ) → Gl(V , J ) × B(V , Ω , J ) : g → (Cg , Zg ),
whose image is the submanifold (C , Z ) | 1 − Z 2 = (C ∗ C )−1 .
 

Proposition 13. If Z1 , Z2 ∈ B(V , Ω , J ), then 1 − Z1 Z2 is invertible and its real part is positive definite.
Proof. Suppose that v ̸= 0. Then
⟨ (1 − Z1 Z2 ) + (1 − Z1 Z2 )∗ v, v⟩ = ⟨(1 − Z1 Z2 )v, v⟩ + ⟨v, (1 − Z1 Z2 )v⟩
 

= ⟨(1 − Z12 )v, v⟩ + ⟨(1 − Z22 )v, v⟩ + ⟨Z12 v, v⟩

+ ⟨Z22 v, v⟩ − ⟨Z1 Z2 v, v⟩ − ⟨v, Z1 Z2 v⟩
= ⟨(1 − Z12 )v, v⟩ + ⟨(1 − Z22 )v, v⟩ + ⟨Z1 v, Z1 v⟩
+ ⟨Z2 v, Z2 v⟩ − ⟨Z1 v, Z2 v⟩ − ⟨Z2 v, Z1 v⟩
= ⟨(1 − Z12 )v, v⟩ + ⟨(1 − Z22 )v, v⟩ + ∥(Z1 − Z2 )v∥2 .
The first two terms in the last equality are positive, while the last term is nonnegative. Consequently, the kernel of 1 − Z1 Z2
is {0} and its real part is positive definite, which proves the proposition. 
Thus 1 − Z1 Z2 ∈ Gl(V , J ) and it lies in the contractible subset
g ∈ Gl(V , J ) | g + g ∗ is positive definite .
 

It follows that there is a smooth logarithm

a : B(V , Ω , J ) × B(V , Ω , J ) → gl(V , J )
such that 1 − Z1 Z2 = ea(Z1 ,Z2 ) . We can define the cocycle η by
η(g1 , g2 ) = Im Tr (a(Zg1 , Zg −1 )).
2

Therefore
det(1 − Zg1 Zg −1 )
ei η(g1 ,g2 ) = 2
,
| det(1 − Zg1 Zg −1 )|
2

which implies
ϕ(g1 g2 ) = ϕ(g1 )ϕ(g2 ) ei η(g1 ,g2 ) .
 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058 2049

5. The relationship with angle functions

In [4], Milnor is interested in lifting ϕ ◦ σ :  G → R, called an angle function, so that ϕ ◦ σ = ei θ . In

G → S 1 to a map θ : 
our case, this is easy: set θ (g , c ) = c. Then
θ ((g1 , c1 ) · (g2 , c2 )) − θ (g1 , c1 ) − θ (g2 , c2 ) = c1 + c2 + η(g1 , g2 ) − c1 − c2 = η(g1 , g2 ).
Lemma 3 in [4] becomes the n = 1 case of the following.

Lemma 14. |η(g1 , g2 )| < nπ /2.

Proof. The above inequality states that |arg(det(1 − Z1 Z2 ))| < nπ /2 if we choose the branch of arg which is 0 at 1. We
1 1 1 1
know that 1 − Z1 Z2 = X + i Y with X positive definite. So 1 − Z1 Z2 = X 2 (1 + i X − 2 YX − 2 )X 2 which implies

det(1 − Z1 Z2 ) = det X det(1 + i 

Y ),
1 1
Y = X − 2 YX − 2 is self-adjoint. Therefore
where 
n

arg(det(1 − Z1 Z2 )) = arg(det X det(1 + i 
Y )) = arg (1 + i 
yj ),
j =1

where  Y . Since arg(1 + i y) ∈ (−π /2, π /2), we obtain the desired inequality.
yj are the eigenvalues of  

It can be seen that by taking 1 − Z12 and 1 − Z22 positive definite that η takes values in the whole interval (−nπ /2, nπ /2)
so the bound in Lemma 14 is sharp.

 (2, R)
6. Special case: Sp

In this section, we work out the above theory in detail for the universal covering group of Sp(2, R) = SL(2, R). 
0 1
Let V be R2 with standard basis {e1 , e2 }. With respect to this basis the symplectic form Ω has matrix J = −1 0
.
 
a b
Suppose that P is a 2 × 2 real matrix which is J-linear, that is, PJ = JP. Then P = −b a
, which we identify with the
 
1 0
complex number a + i b. Let Σ = 0 −1 , which represents complex conjugation on (V , J ) thought of as a complex vector
α β
 
space. A J-antilinear map Z has a matrix of the form β −α , which can be identified with (α − β J )Σ . In other words,
Z can be identified
 with  the C-antilinear map Z : C → C : z → (α − i β)z.

a b
When g = c d ∈ gl(2, R), a calculation shows that
 
1 a+d b−c
Cg =
2 c−b a+d
and as a real matrix
a2 + c 2 − b2 − d2 2(ab + cd)
 
1
Zg = .
(a + d)2 + (b − c )2 2(ab + cd) −(a2 + c 2 − b2 − d2 )
Using the above identification, this gives
(a2 + c 2 − b2 − d2 ) − 2i (ad + bc )
 
Zg z = αg z =
 z.
(a + d)2 + (b − c )2
For g1 , g2 ∈ Sp(2, R) the value of the cocycle η(g1 , g2 ) defined by

det(1 − Zg1 Zg −1 ) 1 − αg1 α g −1

ei η(g1 ,g2 ) = 2
= 2
(1)
| det(1 − Zg1 Zg −1 )| |1 − αg1 α g −1 |
2 2

where η(g1 , g2 ) is continuous and lies in the open interval (−π /2, π /2).

Example 15. To show how useable these formulae are we calculate with the generators of the integer matrices Sp(2, Z).
This subgroup is generated by
   
1 1 1 0
s1 = and s2 = ,
0 1 −1 1
2050 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058

subject to the relations

s1 s2 s1 = s2 s1 s2 and (s1 s2 s1 )4 = 1.
Using the formula for αg , we obtain the table

g
  αg
1 1
s1 − 15 (1 + 2 i)
0 1
 
1 0
s2 1
5
(1 + 2 i)
−1 1
 
1 −1
s−
1
1 1
5
(−1 + 2 i)
0 1
 
1 0
s−
2
1 1
5
(1 − 2 i)
1 1
 
0 1
s1 s2 1
5
(−1 + 2 i)
−1 1
 
1 −1
(s1 s2 )−1 1
5
(1 + 2 i)
1 0

Next we compute the values of the cocycle η at (s1 , s2 ), (s1 s2 , s1 ), and (s2 , s1 s2 ). From (1) and the above table we obtain

1− 1
(−1 − 2 i) 51 (1 + 2 i)
22 + 4 i 11 + 2 i
ei η(s1 ,s2 ) =  5
 = √ = √ .
1 − (−1 − 2 i) 1 (1 + 2 i)
1
10 5 5 5
5 5

Let θ0 = tan−1 12 . Then

 3
2+i 2 + 11 i
e3θ0 i = √ = √ = i e−η(s1 ,s2 ) i ,
5 5 5
which implies

η(s1 , s2 ) = π /2 − 3θ0 .
Note that η(s1 , s2 ) lies in (−π /2, π /2). Now

1− 1
(−1 + 2 i) 51 (1 + 2 i)
ei η(s1 s2 ,s2 ) =  5
 = 1.
1 − (−1 + 2 i) 1 (1 + 2 i)
1
5 5

Consequently, η(s1 s2 , s2 ) = 0. Similarly,

1− 1
(1 + 2 i) 51 (1 − 2 i)
ei η(s2 ,s1 s2 ) =  5
 =1
1 − (1 + 2 i) 1 (1 − 2 i)
1
5 5

and thus η(s2 , s1 s2 ) = 0.

 (2, R)
7. The braid group B3 lies in Sp

In this section, we show that the three strand braid group B3 , given by the presentation

⟨σ1 , σ2 : σ1 σ2 σ1 = σ2 σ1 σ2 ⟩, (2)

is a discrete subgroup of the universal covering group Sp  (2, R) of Sp(2, R).

σj = (sj , cj ), j = 1, 2 of Sp
First, we look for elements   (2, R) which satisfy the braid relation (2) and thus lie in the image
B3 of B3 under the map


τ : B3 → Sp
 (2, R) : σj → 
σj , for j = 1, 2.

Write

σ1 = (s1 , θ0 + 2π k1 ) and 
 σ2 = (s2 , θ0 + 2π k2 ),
 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058 2051

for some k1 , k2 ∈ Z. Then

σ1 · 
 σ2 = (s1 s2 , 2θ0 + 2π k1 + 2π k2 + η(s1 , s2 ))
= (s1 s2 , 2θ0 + 2π k1 + 2π k2 + π /2 − 3θ0 )
= (s1 s2 , −θ0 + 2π k1 + 2π k2 + π /2).
Consequently,
(
σ1 · 
σ2 ) · 
σ1 = (s1 s2 , −θ0 + 2π k1 + 2π k2 + π /2) · (s1 , θ0 + 2π k1 )
= (s1 s2 s1 , 4π k1 + 2π k2 + π /2 + η(s1 s2 , s1 ))
= (s1 s2 s1 , 4π k1 + 2π k2 + π /2)
and
σ2 · (
 σ1 · 
σ2 ) = (s2 , θ0 + 2π k2 ) · (s1 s2 , −θ0 + 2π k1 + 2π k2 + π /2)
= (s2 s1 s2 , 2π k1 + 4π k2 + π /2 + η(s2 , s1 s2 ))
= (s2 s1 s2 , 2π k1 + 4π k2 + π /2).
Therefore the braid relation
(
σ1 · 
σ2 ) · 
σ1 = 
σ1 · (
σ2 · 
σ1 )
is satisfied if and only if k1 = k2 = k. Hence
σ1 · 
 σ2 · 
σ1 = 
σ1 · 
σ2 · 
σ1 = (J , 6π k + π /2),
for some k ∈ Z. Moreover,
(
σ1 · 
σ2 )3 = (
σ1 · 
σ2 · 
σ1 ) · (
σ2 · 
σ1 · 
σ2 )
= (J , 12π k + π ) = (−1, 12π k + π ).
2

Setting k = 0, we have shown that

σ1 · 
 σ2 · 
σ1 = 
σ1 · 
σ2 · 
σ1 = (J , π /2)
and
(
σ1 · 
σ2 )3 = (−1, π ).
We now show that (
σ1 · 
σ2 )3 generates the centre of Sp
 (2, R). Suppose that
(g1 , a1 ) · (g2 , a2 ) = (g2 , a2 ) · (g1 , a1 )
for every (g2 , a2 ) ∈ Sp
 (2, R). Then
(g1 g2 , a1 + a2 + η(g1 , g2 )) = (g2 g1 , a2 + a1 + η(g2 , g1 ))
for every g2 ∈ Sp(2, R). Hence g1 lies in the centre Z (Sp(2, R)) of Sp(2, R); so in particular, g1 ∈ U(1). Consequently,
η(g1 , g2 ) = η(g2 , g1 ) = 0 and ei a1 = ϕ(g1 ) = det g1 = ±1. In other words, (g1 , a1 ) is either (−1, 2n π + 1) or (1, 2n π ) for
some n ∈ Z. But (−1, π )2n = (1, 2n π ). Thus Z (Sp  (2, R)) is generated by (−1, π ).

Proposition 16.   (2, Z).

B3 = Sp

σ ∈ Sp
Proof. Let   (2, Z) ⊆ Sp  (2, R). Then 
σ = (g , c ), where g ∈ Sp(2, Z). Now g is a word w(s1 , s2 ) in s1 and s2 which is a
product of sj s and s−
k
1
s. Consider

w(
σ1 , 
σ2 ) = (w(s1 , s2 ), c ′ ) = (g , c ′ ).
′
Because σj ∈ Sp (2, R), we have ϕ(g ) = ei c = ei c , which implies that c = c ′ + 2π ℓ for some ℓ ∈ Z. Hence
(1, 2π ℓ) = ( σ2 )6ℓ . But
σ1 · 
σ = (g , c ) = (g , c ′ ) · (1, 2π ℓ),
 because η(1, g ) = 0
= w( σ1 , 
σ2 ) · (
σ1 · σ2 )6ℓ ,
which lies in   (2, Z) ⊆ 
B3 . Thus Sp σ1 and 
B3 , which proves the proposition, since  σ2 generate Sp
 (2, Z). 

We now prove the following lemma.

Lemma 17. B3 is isomorphic to 

B3 .
2052 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058

Proof. Consider
τ σ
B3 →   (2, R) : σj → 
B3 → Sp σj , for j = 1, 2.

Let v(σ1 , σ2 ) be a word in ker τ . Then v(s1 , s2 ) = 1. But s1 s2 s1 (s2 s1 s2 )−1 and (s2 s1 s2 )4 generate the set of relations in Sp(2, Z).
Hence
v(s1 , s2 ) = u(s1 s2 s1 (s2 s1 s2 )−1 , (s2 s1 s2 )4 )
for some word u. Therefore
v(σ1 , σ2 ) = u((σ1 σ2 σ1 )(σ2 σ1 σ2 )−1 , (σ1 σ2 σ2 )4 )
= u(1, (σ1 σ2 σ1 )4 ) = (σ1 σ2 σ1 )4m .
But
(σ1 σ2 σ1 )2 = (σ1 σ2 σ1 )(σ2 σ1 σ2 ) = (σ1 σ2 )3 .
So
(σ1 σ2 σ1 )4m = (σ1 σ2 )6m .
However,
(1, 0) = τ (v(σ1 , σ2 )) = τ ((σ1 σ2 )6m ) = (
σ1
σ2 )6m = ((−1, π ))2m = (1, 2m π ).
Consequently, m = 0, which implies v(σ1 , σ2 ) = 1. Thus the surjective homomorphism τ is injective and hence is an
isomorphism. This completes the proof of the lemma. 
We have thus obtained the following result of Milnor [5].

Theorem 18. The inverse image of Sp(2, Z) in the simply connected covering group of Sp(2, R) is isomorphic to the braid group
on three strands.

8. The Lagrangian Grassmannian

Let (V , Ω , J ) be as above and let V C = V + i V be the complexification of V . Let v = v1 + iv2 → v = v1 − iv2 be complex
conjugation and for a subset W of V C let W = {v | v ∈ W }. A complex subspace W ⊂ V C is said to be real if W = U + iU
for a subspace U ⊂ V and this is the case if and only if W = W .
Extend Ω , Sp(V , Ω ), J, etc. to be C-linear (or bilinear) on V C .

Definition 19. A subspace F of V C is (complex) Lagrangian if

(i) Ω (v, w) = 0 for every v, w ∈ F ;
(ii) dimC F = 21 dimR V .
A Lagrangian subspace F is positive if i Ω (v, v) ≥ 0 for every v ∈ F and strictly positive if i Ω (v, v) > 0 for every v ̸= 0 ∈ F .
A Lagrangian subspace is real if F = W + i W , where W = F ∩ V is a real Lagrangian subspace of V ; F is a complex
structure if F ∩ F = {0}. F + F and F ∩ F are always real subspaces. Real Lagrangian subspaces and complex structures are
the two extremes where F ∩ F is as large or as small as possible.
The first result is the following.

Proposition 20. If F and G are positive Lagrangian subspaces of V C , then F ∩ G is real.

Proof. In fact we show that F ∩ G ⊆ F ∩ F , which implies F ∩ G = F ∩ F ∩ G ∩ G and hence the result.
If v ∈ F ∩ G then v ∈ G, so i Ω (v, v) ≥ 0. Consequently, −i Ω (v, v) ≥ 0. But i Ω (v, v) ≥ 0 since v ∈ F . Hence
Ω (v, v) = 0. But i Ω (v, w) is a positive definite form on F . So by Cauchy–Schwartz, its kernel is all vectors with Ω (v, v) = 0.
But its kernel is F , since F is Lagrangian. Thus v ∈ F and Ω (v, v) = 0, which implies that v ∈ F ∩ F . Hence F ∩ G ⊆ F ∩ F . 

Corollary 21. If F is positive and G is strictly positive, then F ∩ G = {0}.

Proof. If G is strictly positive, then G ∩ G = {0}. So F ∩ G ⊆ G ∩ G = {0}. Hence F ∩ G = {0}. 

Lemma 22. F0 = {(1 + i J )v | v ∈ V } is a strictly positive Lagrangian subspace of V C .

Proof. The lemma follows from
i Ω ((1 + i J )v, (1 − i J )v) = 2Ω (v, J v) > 0,
provided that v ̸= 0. 
 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058 2053

Corollary 23. If F is positive, then F ∩ F0 = {0}.

Since V C = F0 ⊕ F 0 is a direct sum, it follows that the map F → F0 given by v → 1

2
(1 + i J )v is onto. Because
dim F = dim F0 , this map is a C-linear isomorphism. The map V → F0 defined by v → 1
2
(1+i J )v is an R-linear isomorphism.
Hence, given v ∈ V there is a unique v1 + i v2 ∈ F so that

1 1
(1 + i J )(v1 + i v2 ) = (1 + i J )v.
2 2
In other words, v = v1 − J v2 for unique v1 + i v2 ∈ F . Define ZF (v) = v1 + J v2 .

Proposition 24. If F is a positive Lagrangian subspace of V C , then v → ZF v is a J-antilinear map on V , (v, w) → ⟨v, ZF w⟩
is C-symmetric, and 1 − ZF2 is positive. If F is strictly positive, then 1 − ZF2 is positive definite. The correspondence F → ZF is a
bijection from the set of strictly positive Lagrangian subspaces of V C onto the Siegel domain B(V , Ω , J ).

Proof. If v = v1 − J v2 with v1 + i v2 ∈ F , then J v = J v1 + v2 and v2 − i v1 = −i(v1 + i v2 ) ∈ F . So

ZF J v = −J v1 + v2 = −J (v1 − J v2 ) = −JZF v.

Consequently, ZF is J antilinear.
Let v = v1 − J v2 , w = w1 − J w2 where v1 + i v2 , w1 + i w2 ∈ F . Then

⟨v, ZF w⟩ − ⟨w, ZF v⟩ = ⟨v, w1 + J w2 ⟩ − ⟨w, v1 + J v2 ⟩

= Ω (v1 − J v2 , J (w1 + J w2 )) − i Ω (v1 − J v2 , w1 + J w2 )
− Ω (w1 − J w2 , J (v1 + J v2 )) + i Ω (w1 − J w2 , v1 + J v2 )
= −Ω (v1 , w2 ) − Ω (v2 , w1 ) − Ω (v2 , w1 ) − Ω (v1 , w2 )
+ i[−Ω (v1 , w1 ) + Ω (v2 , w2 ) − Ω (v1 , w1 ) + Ω (v2 , w2 )]
= −2i Ω (v1 − i v2 , w1 − i w2 ) = 0,
since F is isotropic. Now

⟨ZF2 v, v⟩ = ⟨ZF v, ZF v⟩ = Ω (v1 + J v2 , J (v1 + J v2 ))

provided that v = v1 − J v2 with v1 + i v2 ∈ F . Then we also have

⟨v, v⟩ = Ω (v1 − J v2 , J (v1 − J v2 )).

Consequently,

⟨(1 − ZF )2 v, v⟩ = 4 Ω (v1 , v2 ).
But

i Ω (v1 + i v2 , v1 − i v2 ) = 2 Ω (v1 , v2 ).

Thus F being positive implies that 1 − ZF2 is positive. F strictly positive implies that 1 − ZF2 is positive definite.
To see that we get a bijection between strictly positive Lagrangian subspaces and B(V , Ω , J ), we define the inverse map.
For Z ∈ B(V , Ω , J ), set
 
1 1
F = (Z + 1)v − i J (Z − 1)v | v ∈ V .
2 2

Since Z + 1 and Z − 1 are invertible, the map v → 12 (Z + 1)v − 12 i J (Z − 1)v has real rank dim V . Hence dimR F = dim V .
But in fact F is a C-subspace. To see this, we find that

i((Z + 1)v − iJ (Z − 1)v) = J (Z − 1)v + i(Z + 1)v

= −(Z + 1)J v − i J (−Z + 1)J v
= (Z + 1)(−J v) − i J (Z − 1)(−J v) ∈ F .
1
Thus F has dimension 2
dim V as a complex space. An easy calculation shows

Ω ((Z + 1)v − iJ (Z − 1)v, (Z + 1)w − i J (Z − 1)w) = 2i(⟨Z v, w⟩ − ⟨Z w, v⟩).

So F is isotropic is equivalent to ⟨Z v, w⟩ = ⟨Z w, v⟩. Positivity is similar. This completes the proof of the proposition. 
2054 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058

Under the correspondence defined in Proposition 24, we see that F is real implies that v1 , v2 ∈ F if v1 + i v2 ∈ F . So
⟨(1 − ZF2 )v, v⟩ = 4 Ω (v1 , v2 ) = 0
and hence ZF2 = 1. Thus the real Lagrangian Grassmannian in our picture is the set of J-antilinear maps Z of V into itself with
Z 2 = 1 and ⟨Z v, w⟩ = ⟨Z w, v⟩.
How does the real symplectic group Sp(V , Ω ) act on Λ(V , Ω )? On Lagrangian subspaces of V it acts by F → g · F for
g ∈ Sp(V , Ω ). What is its action on Z ? In other words, give a formula for Zg ·F in terms of g and ZF . To answer this last question,
let v = v1 − J v2 with v1 + i v2 ∈ F . Then g v1 + i g v2 ∈ g · F . So Zg ·F (g v1 − Jg v2 ) = g v1 + Jg v2 and ZF (v1 − J v2 ) = v1 + J v2 .
Now
g v1 − Jg v2 = Cg (1 + Zg )v1 − JCg (1 + Zg )v2
= Cg (1 + Zg )v1 − Cg J v2 + Cg Zg J v2
= Cg (v1 − J v2 ) + Cg Zg (v1 + J v2 )
= Cg v + Cg Zg ZF v = Cg (1 + Zg ZF )v
and
g v1 + Jg v2 = Cg (1 + Zg )v1 + JCg (1 + Zg )v2
= Cg (v1 + J v2 ) + Cg Zg (v1 − J v2 )
= Cg ZF v + Cg Zg v = Cg (ZF + Zg )v.
Hence
Zg ·F Cg (1 + Zg ZF )v = Cg (ZF + Zg )v.
Since Zg ∈ B(V , Ω , J ), it follows that 1 + Zg ZF is invertible for every positive Lagrangian subspace F . Thus

Zg ·F = Cg (ZF + Zg )(1 + Zg ZF )−1 Cg−1 . (3)

Eq. (3) is a uniform formula for the action of Sp(V , Ω ) on Lagrangian subspaces of V in terms of our parameters.
Formula (3) shows that
g · Z = Cg (Z + Zg )(1 + Zg Z )−1 Cg−1
is a smooth action of Sp(V , Ω ) on the set of antilinear maps Z of V into itself with the properties that ⟨v, Z w⟩ = ⟨w, Z v⟩ for
every v, w ∈ V and 1 − Z 2 is positive definite.

9. The universal covering space of the Lagrangian Grassmannian

In this section, we define a circle function on the Lagrangian Grassmannian and use it to make a construction (4) of its
universal covering space Λ (V , Ω ) by analogy with the universal covering group Sp
 (V , Ω ) of the symplectic group. We show
that our constructions are compatible by giving an explicit action (5) of Sp  (V , Ω ) on Λ
(V , Ω ). Finally, we show how to lift
paths to the universal covers.
Let Λ(V , Ω ) be the real Lagrangian Grassmannian. If λ ∈ Λ(V , Ω ), then λ and J λ are perpendicular with respect to the
Euclidean inner product Re ⟨ , ⟩ on V . In particular, V = λ ⊕ J λ is a direct sum. Define Zλ by Zλ v = v1 − J v2 if v = v1 + J v2
with vi ∈ λ. In this special case, we have Zλ2 = 1 and as before (v, w) → ⟨v, Zλ w⟩ is a C-symmetric bilinear form. Moreover,
λ = {v ∈ V | Zλ v = v}.
To define a circle function ψ : Λ(V , Ω ) → S 1 for Λ(V , Ω ), we pick a base point λ0 ∈ Λ(V , Ω ) and consider the function
λ → Zλ Zλ0 .

Proposition 25. Zλ Zλ0 ∈ U(V , Ω , J ).

Proof. Zλ Zλ0 is J-linear. Moreover,

⟨Zλ Zλ0 u, Zλ Zλ0 v⟩ = ⟨Zλ0 v, Zλ2 Zλ0 u⟩

= ⟨Zλ0 v, Zλ0 u⟩ = ⟨u, Zλ20 v⟩ = ⟨u, v⟩. 

Therefore we can define ψ by λ → detC (Zλ Zλ0 ).

Lemma 26. If U is a C-vector space and A, B : U → U are antilinear maps, then

detC (AB) = detC (BA).

 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058 2055

Proof. Fix a basis to identify U with Cn . Then A becomes z → A1 z and B becomes z → B1 z with A1 , B1 being n × n matrices
with complex entries. So ABz = A1 B1 z and BAz = B1 A1 z. Thus
detC (AB) = detC (A1 B1 ) = detC A1 detC B1
and
detC (BA) = detC (B1 A1 ) = detC B1 detC A1 . 

The above formula allows us to compute ψ(g · λ) using (3) as follows:

Zg ·λ Zλ0 = Cg (Zλ + Zg )(1 + Zg Zλ )−1 Cg−1 Zλ0 .
Consequently,
 
detC (Zg ·λ Zλ0 ) = detC Cg (Zλ + Zg ) (1 + Zg Zλ )Cg−1 Zλ0
    
A B

= detC ((1 + Zg Zλ )−1 Cg−1 Zλ0 Cg (Zλ + Zg ))

= detC (1 + Zg Zλ )−1 detC Cg−1 detC ( Zλ0 Cg (1 + Zg Zλ )Zλ )
   
A B

= detC (1 + Zg Zλ )−1 detC Cg −1

detC (Cg (1 + Zg Zλ )Zλ Zλ0 )

= detC (1 + Zg Zλ )−1 detC Cg−1 · detC Cg detC (1 + Zg Zλ ) detC (Zλ Zλ0 ).

Hence
ψ(g · λ) = ψ(λ)ϕ(g )2 ei ν(g ,λ) ,
where
detC (1 + Zg Zλ )
 2
det Cg i ν(g ,λ)
ϕ(g ) = and e = .
| det Cg | | detC (1 + Zg Zλ )|
The smooth function ν is uniquely determined if we require that ν(e, λ0 ) = 0. This then implies that ν(g , λ) = 0 for every
g ∈ U(V , Ω , J ).

Theorem 27. There is a smooth map ψ : Λ(V , Ω ) → S 1 inducing an isomorphism of π1 (Λ(V , Ω )) with π1 (S 1 ) = Z and
satisfying
ψ(g · λ) = ψ(λ)(detC g )2 ,
if g ∈ U(V , Ω , J ). There is a smooth map ν : Sp(V , Ω ) × Λ(V , Ω ) → R such that
ψ(g · λ) = ψ(λ) ϕ(g )2 ei ν(g ,λ) .

Is the map ψ unique? Obviously ψ can take any value at a given point, but then it is determined. For if ψ1 and ψ2 satisfy
ψi (g · λ) = ψi (λ)(det g )2 for every g ∈ U (V , Ω , J ), then ψ1 /ψ2 is invariant. Since U(V , Ω , J ) acts transitively on Λ(V , Ω ),
it follows that ψ1 /ψ2 is constant. Thus the map ψ is independent of the choice of circle function ϕ . Can any choice of circle
function do? Yes, for if ϕ0 and ν0 satisfy
ψ(g · λ) = ψ(λ) ϕ0 (λ)2 ei ν0 (g ,λ)
then ϕ/ϕ0 is trivial on U(V , Ω , J ). So ϕ(g ) = ϕ0 (g ) ei τ (g ) for some smooth map τ : Sp(V , Ω ) → R with τ (g ) = 0 for every
g ∈ U(V , Ω , J ). Hence
ψ(g · λ) = ψ(λ) ϕ(g )2 ei ν(g ,λ) ,
where ν(g , λ) = 2τ (g ) + ν0 (g , λ).
Define the universal covering space of Λ(V , Ω ) by

(V , Ω ) = (λ, c ) ∈ Λ(V , Ω ) × R | ψ(λ) = ei c

Λ
 
(4)

and the covering projection σ : Λ

(V , Ω ) → Λ(V , Ω ) : (λ, c ) → λ. Also define a map ρ : Λ
(V , Ω ) → R : (λ, c ) → c.
Then
ei ρ(λ,c ) = ei c = ψ(λ) = ψ(σ (λ, c )).
2056 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058

Consequently, ei ρ = ψ ◦ σ . Define an action of Sp

 (V , Ω ) on Λ
(V , Ω ) by
(g , a) · (λ, c ) = (g · λ, c + 2a + ν(g , λ)). (5)

 (V , Ω ) acts transitively on Λ
The above action is smooth. Moreover, Sp (V , Ω ). Combining the above definitions we obtain
ρ( λ) = ρ(
g · λ) + 2 θ (
g ) + ν(σ (
g ), σ (
λ)). (6)
Using (6), we now derive a formula for the cocycle η in terms of the function ν . Applying (6) twice gives
ρ((
g1 · g2 ) · 
λ) = ρ( λ) + 2 θ (
g2 ·  g1 ) + ν(σ (
g1 ), σ (
g2 · λ))
= ρ(λ) + 2 θ (
 g ) + ν(σ (
2 g ), σ (λ)) + 2 θ (
2
 g ) + ν(σ (
1 g ), σ ( 1λ)).
g · 2

But
ρ((
g1 g2 ) · 
λ) = ρ(
λ) + 2 θ (
g1 g2 ) + ν(σ (
g1 g2 ), σ (
λ)).
Consequently,
2 η(g1 , g2 ) = 2(θ (
g1 g2 ) − θ (
g1 ) − θ (
g2 ))
= ν(σ (g2 ), σ (
λ)) + ν(σ (
g1 ), σ ( λ)) − ν(σ (
g2 ·  g1 g2 ), σ (
λ))
= ν(g2 , λ) + ν(g1 , g2 · λ) − ν(g1 · g2 , λ).
Given a continuous closed curve [0, 1] → Λ(V , Ω ) : t → λt with λ0 = λ1 , we lift it to a continuous curve
[0, 1] → Λ (V , Ω ) : t → 
λt = (λt , ct ) with ct a continuous real-valued function of t satisfying eict = ψ(λt ) for all
t ∈ [0, 1]. Then e = ψ(λ1 ) = ψ(λ0 ) = eiλ0 and so (c1 − c0 )/2π is an integer. If (λt , dt ) is a second lift then (ct − dt )/2π
ic1

is integer valued and continuous; hence constant. Thus (d1 − d0 )/2π = (c1 − c0 )/2π and so the integer is independent of
which lift is chosen.

Definition 28. If λt is a continuous closed curve in the real Lagrangian Grassmannian Λ(V , Ω ) and (λt , ct ) is a continuous
(V , Ω ) then the integer (c1 − c0 )/2π constructed as above is called the Maslov index of the curve λt .
lift into Λ

Lift λt to a curve gt ∈ Sp(V , Ω ) with λt = gt · λ0 and set g0 = 1. Then we can lift gt to a curve   (V , Ω ) with
gt in Sp
g0 = (1, 0). We have 
 gt = (gt , at ). Moreover,
 λ0 = (gt , at ) · (λ0 , 0) = (gt · λ0 , 2at + ν(gt , λ0 )) = (λt , 2at + ν(gt , λ0 )).
gt · 
Therefore c1 − c0 = 2a1 + ν(g1 , λ0 ). In other words,

c1 − c0 = 2θ (
g1 ) + ν(g1 , λ0 ). (7)

10. The graph map

The graph of a symplectic linear map of a symplectic vector space is a Lagrangian subspace of the symplectic double so
we get a map of the real symplectic group into a Lagrangian Grassmannian. In this section, we show how our circle function
for the symplectic group is related to the circle function in the Lagrangian Grassmannian of the symplectic double.
If g ∈ Sp(V , Ω ), let
Γg = {(v, g v) | v ∈ V } ⊆ V ⊕ V
be its graph. Then Γg is Lagrangian in (V ⊕ V , Ω ⊕ (−Ω )). Hence we get a map
Γ : Sp(V , Ω ) → Λ(V ⊕ V , Ω ⊕ (−Ω )) : g → Γg ,
called the graph map. We will use Γid = ∆V , the diagonal, as a base point Λ(V ⊕ V , Ω ⊕ (−Ω )).
 in 
J 0
Given J in (V , Ω ) we obtain J1 in (V ⊕ V , Ω ⊕ (−Ω )) defined by 0 −J . We then have the circle functions

det Cg
ϕ : Sp(V , Ω ) → S 1 : g →
| det Cg |
and
Ψ : Λ(V ⊕ V , Ω ⊕ (−Ω )) → S 1 : Γg → det(ZΓg ZΓid ).
 
A B
To aid computations we observe that C D
is J1 -linear if and only if A and D are J-linear and B and C are J-antilinear. We
have the following special cases of Lemma 26.
 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058 2057

Lemma 29.
 
A 0
det = det A det D,
0 D
   
1 B 1 0
det = det = 1.
0 1 C 1

Proposition 30.
Cg−1
   
−Zg 0 1
ZΓg = and ZΓid = .
(Cg∗ )−1 Cg Zg Cg−1 1 0

Proof. ZΓg is defined by

v v1 v2
     
ZΓg = − J1
w w1 w2
if
v v1 v2
     
= + J1
w w1 w2
with (vi , wi )T ∈ Γg . When wi = g vi , this means that
v = v1 + J v2
w = g v1 − Jg v2 = Cg (1 + Zg )v1 − Cg (1 − Zg )J v2 .
Hence
Cg−1 w = v1 − J v2 + Zg (1 − Zg )J v2 .
In other words,
v1 − J v2 = Cg−1 w − Zg v and v1 + J v2 = v.
But
v v1 − J v2 v1 − J v2
     
ZΓg = =
w g v1 + Jg v2 Cg [(v1 + J v2 ) + Zg (v1 − J v2 )]
Cg w − Zg v
−1
−Zg v + Cg−1 w
   
= = .
Cg [v + Zg (Cg−1 w − Zg v)] Cg (1 − Zg2 )v + Cg Zg Cg−1 w

The desired formula for ZΓg follows because 1 − Zg2 = (Cg∗ Cg )−1 . Substituting Zid = 0 and Cid = 1 into the preceding formula
gives the second. 
Thus
Cg−1
 
−Z g
ZΓg ZΓid =
Cg Zg Cg−1 (Cg∗ )−1
(1 − Zg2 )−1
     
1 0 1 −Zg 0 1 0 Cg 0
= .
0 (Cg∗ )−1 0 1 0 1 Zg 1 0 1
Consequently,
det(ZΓg ZΓid ) = det(Cg∗ )−1 × 1 × det(Cg∗ Cg ) × 1 × det Cg−1
det Cg
= = ϕ(g )−2 .
det Cg
Hence, with our normalisation we have proved the following theorem.

Theorem 31. For every g ∈ Sp(V , Ω ) the circle functions are related by
Ψ (Γg ) = ϕ(g )−2 .

Acknowledgement

This work grew out of discussions with Richard Cushman about performing explicit calculations of Maslov indices of
paths. It is a great pleasure to thank him for his help and encouragement.
2058 J. Rawnsley / Journal of Geometry and Physics 62 (2012) 2044–2058

References

[1] P.L. Robinson, J.H. Rawnsley, The Metaplectic Representation, Mpc Structures and Geometric Quantization, in: Memoirs of the AMS, vol. 410, American
Mathematical Society, Providence, RI, 1989.
[2] M. Cahen, S. Gutt, J. Rawnsley, Symplectic Dirac operators and Mpc -structures, Gen. Relativity Gravitation 43 (2011) 3593–3617.
[3] K. Habermann, L. Habermann, Introduction to Symplectic Dirac Operators, in: Lecture Notes in Mathematics, vol. 1887, Springer-Verlag, Berlin, 2006.
[4] J. Milnor, On the existence of a connection with curvature zero, Comment. Math. Helv. 32 (1958) 215–223.
[5] J. Milnor, Introduction to Algebraic K -Theory, in: Annals of Mathematics Studies, vol. 72, Princeton University Press, Princeton, NJ, 1971.