56 Classical Mechanics - Mazhe

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Chapter 56

Classical mechanics

56.1 Some symplectic and Poisson geometry


56.1.1 Symplectic manifold
A symplectic structure on a vector space V is a skew-symmetric, nondegenerate bilinear
2-form : V ˆ V Ñ . We define the symplectic group SPp q as the group of linear operators
A : V Ñ V such that pAu, Avq “ pu, vq for every u, v P V . It is easy to see that elements of
SPpV q satisfy
At A “ . (56.1)

The Lie algebra of SPp q is denoted by spp q. Taking the derivative of equation (56.1) with
respect to A, one finds the following condition for B P spp q:

B ` Bt “ 0. (56.2)

If W is a subspace of the symplectic vector space V , the symplectic complement is

W Ê “ tx P V tel que Êpx, yq “ 0 @y P W u. (56.3)

We say that W is a Lagrangian of V when W “ W Ê . Typically, when one consider the canonical
structure ppi , qj q “ ”ij on 2n , the spaces Spantpi tel que i “ 1, . . . , nu and Spantqi tel que i “
1, ¨ ¨ ¨ , nu are Lagrangian.
A symplectic manifold is the data of a smooth manifold M and a symplectic structure Êx
on each tangent space Tx M . The map x fiÑ Êx is required to be a smooth section of the 2-tensor
bundle.

Définition 56.1.
A symplectic Lie algebra is a Lie algebra s endowed with a symplectic structure Ê such that @x,
y, z P s,
Êprx, ys, zq ` Êpry, zs, xq ` Êprz, xs, yq “ 0. (56.4)

56.1.2 Poisson manifold


Let M be a smooth manifold. A Poisson structure, or a Poisson bracket on M is a map
t., .u : C 8 pM q ˆ C 8 pM q Ñ C 8 pM q such that
` ˘
(1) C 8 pM q, t., .u is a Lie algebra,
(2) for each f P C 8 pM q, the map tf, .u is a derivation of the algebra C 8 pM q:

tf, ghu “ tf, guh ` gtf, hu.

2399
2400 CHAPTER 56. CLASSICAL MECHANICS

56.1.3 Hamiltonian action


Let pM1 , Ê1 q and pM2 , Ê2 q be symplectic manifolds. A symplectomorphism from M1 to M2
is a diffeomorphism Ï : M1 Ñ M2 such that Ï˚ Ê2 “ Ê1 .
For any function f P C 8 pM q, we define the Hamiltonian field Xf P XpM q associated with f
by
ipXf qÊ “ df. (56.5)
Existence of such a Xf for each f is assured because Ê is nondegenerate. A symplectic structure
induces a Poisson bracket by defining

tf, gu “ ´ÊpXf , Xg q “ ´Xg pf q “ Xf pgq. (56.6)

In local coordinates, one can write Ê “ 12 Êij dxi ^dxj and Xf “ Ê ij Bi f Bj , where pÊ ij q is the inverse
matrix of pÊij q. The Poisson tensor defined by

tf, gu “ P kl Bk f Bl g, (56.7)

is nothing else than P “ Ê ´1 .

Théorème 56.2.
If Ï : M Ñ M 1 is a diffeomorphism between two Poisson manifolds pM, P q and pM 1 , P 1 q, then the
following are equivalent:
(1) Ï˚ pXf ˝Ï q “ Xf1 ,
(2) tu ˝ Ï, v ˝ Ïu “ tu, vu1 ˝ Ï,
(3) Ï˚ P “ P 1 ,
If moreover the Poisson structures P and P 1 come from symplectic forms Ê and Ê 1 , Ï˚ Ê 1 “ Ê.

Définition 56.3.
Let · : GˆM Ñ M be a symplectic action of a Lie group G on M (i.e. ·g : M Ñ M is a symplectic
transformation of M for each g P G). That action is Hamiltonian if, for every X P G, there
exists a map ⁄X P C 8 pM, q such that

ipX ˚ qÊ “ d⁄X , (56.8a)


t⁄X , ⁄Y u “ ⁄rX,Y s (56.8b)

where X ˚ is the fundamental vector field associated with the vector X.

Définition 56.4.
The map ⁄ : G Ñ C 8 pM q which satisfies (56.8) is the dual momentum map while the momentum
map is J : M Ñ G ˚ defined by
⁄X pxq “ xJpxq, Xy (56.9)
for all X P G.

Using the musical isomorphism (see 49.3.3), equation (56.8a) can be written as pd⁄X q7 “ X ˚
or d⁄X “ pX ˚ q5 .
Since ipX⁄Y qÊ “ d⁄Y “ ipY ˚ qÊ, we have:

X⁄Y “ Y ˚ . (56.10)

The symplectic gradient is


sgrad u “ pduq7 .

Remarque 56.5.
If the action only fulfils (56.8a), it is said to be weakly Hamiltonian. But one can find some
literature in which the first condition is called a Hamiltonian action and the second, a strongly
Hamiltonian one.
56.1. SOME SYMPLECTIC AND POISSON GEOMETRY 2401

Remarque 56.6.
Be careful with some choices of signs. We made at least three choices: the definition of ⁄X in
(56.8a) can get a minus sign; the definition of the Poisson bracket (56.6) is modulo a sign; and the
definition of the fundamental fields that we take is
d” ı
Xx˚ “ ·e´tX x . (56.11)
dt t“0

It can be written without the minus sign in the exponential. Each of these choice is free, by
condition (56.8b) is not! If one make “wrong” choices, this can become

t⁄X , ⁄Y u “ ´⁄rX,Y s , (56.12)


ˇ
with all the consequences in the computations. For example, if one defines Xx˚ “ dt ·exp tX x t“0 ,
d ˇ
one finds an extra minus sign in equations (80.39), (80.41) and (80.43).

56.1.4 Coadjoint orbits


Let G be a Lie group and G its Lie algebra. We know that G acts on the dual G ˚ by

g · › “ › ˝ Adpg ´1 q “ Adpgq˚ › (56.13)

for g P G and › P G ˚ . The second equality defines the coadjoint action Ad˚ : G ˆ G ˚ Ñ G ˚ . In
other words, for all X P G,

pg · ›qpXq “ x›, Adpg ´1 qXy “ xAdpgq˚ p›q, Xy.

This representation is also called the contragredient representation. In this context, the notion
of fundamental fields is given bu X›˚ “ › ˝ adpXq. Let ◊› “ tg · ›|g P Gu, the orbit of › in G ˚ . It
can be shown that
Ễx pXx˚ , Yx˚ q “ xx, rX, Y sy (56.14)
defines a symplectic form on ◊› , the coadjoint orbit of ›.

Proposition 56.7.
The coadjoint action is Hamiltonian.

Proposition 56.8.
The diffeomorphism
Ïpgq : ◊› Ñ ◊›
(56.15)
› fiÑ g · ›
fulfils ` ˘
Ïpgq˚ Ê › pX›˚ , Y›˚ q “ Ê› pX›˚ , Y›˚ q.

56.1.5 Example of adjoint and coadjoint orbit


Let us consider the following subgroup of SLp2, q
ˆ ˙
a n
“t where a ° 0 and n P u. (56.16)
0 a´1

The Lie algebra is ˆ ˙


x y
s“t where x, y P u. (56.17)
0 ´x
The following is a basis of s: ˆ ˙ ˆ ˙
1 0 0 1
H“ E“ (56.18)
0 ´1 0 0
2402 CHAPTER 56. CLASSICAL MECHANICS

and the commutation relation is rH, Es “ 2E. Since we are in a matrix group, the adjoint action
is given by AdpgqX “ gXg ´1 and we can compute
ˆ ˙
x ay 2 ´ 2nax
Adpa, nqpx, yq “ “ xH ` pa2 y ´ 2naxqE (56.19)
0 ´x
Thus the adjoint orbits are the straight lines.
For the coadjoint orbits, consider an element › “ ›H H ˚ ` ›E E ˚ P s˚ . We have
xAd˚ pg ´1 q›, X0 y “ x›, AdpgqXy
(56.20)
“ p›H ´ 2na›E qx ` a2 ›E y.
If we write › “ p›H , ›E q we have
ˆ ˙ˆ ˙
1 ´2na ›H
Ad pa, nq
˚ ´1
p›H , ›E q “ . (56.21)
0 a2 ›E
This is the orbit of ›. Notice that H ˚ is fix while E ˚ fiÑ ´2naH ˚ ` a2 E ˚ , so that E ˚ has a quite
large orbit. If ›2 ‰ 0, then dim Ad˚ p q› “ 2. If ›E “ 0, then the orbit is one single point. Thus
we have three kind orbits in the plane pH ˚ , E ˚ q:
(1) the upper half plane is one two dimensional orbit,
(2) each of the point on the ›E “ 0 is an orbit,
(3) the lower half plane is a second two dimensional orbit.
These orbits are very different from the adjoint orbits.

56.1.6 A second example


Let g “ SLp2, q and the basis tH, E, F u with the relations
rH, Es “ 2E
rH, F s “ ´2F (56.22)
rE, F s “ H.
The Killing form is given by the matrix
¨ ˛
8 0 0
— “ ˝0 0 4‚. (56.23)
0 4 0
It is nondegenerate and its signature is `, `, ´.

Lemme 56.9.
If the Killing form — is nondegenerate, the map
5 : g Ñ g˚
(56.24)
X 5 pyq “ —pX, Y q
is a linear Ad-equivariant isomorphism, that is
` ˘5
Ad˚ pgqpX 5 q “ AdpgqX . (56.25)
Proof. We have ` ˘
xAd˚ pgqX 5 , Y y “ — X, Adpg ´1 qY
` ˘
“ — AdpgqX, Y (56.26)
` ˘5
“ AdpgqX Y.

In that case, the 5 map intertwines the actions and the adjoint orbits are the same as the
coadjoint orbits.
56.1. SOME SYMPLECTIC AND POISSON GEOMETRY 2403

56.1.7 Kostant theorem


Let G act transitively on M and on G ˚ by the coadjoint action (56.13). We suppose this action
to be Hamiltonian. One can prove that J is equivariant:
Jpg · xq “ g · Jpxq (56.27)
where the first dot is the action on M and the second one, the action on G ˚ . The momentum map

of M and G ˚ are related by ⁄M X “ ⁄X ˝ J. Since G is transitive on M , for any x P M , we have
M “ G · x for any x in M , so
JpM q “ G · Jpxq.
We conclude that JpM q is only one orbit in G ˚ .

Proposition 56.10.
Let O be the coadjoint orbit of a connected Lie group G. Then O is canonically endowed by a
Ad˚ pGq-equivariant symplectic structure given by
Ê›O pX ˚ , Y ˚ q “ x›, rx, Y sy (56.28)
where X ˚ stands for the fundamental vector. Here the equivariance means
Ad˚ pgqÊ O “ Ê O (56.29)
for every g P G.

Proof. No proof.

The formula (56.28) is Kirillov-Kostant-Souriau. Notice that this proposition shows that the
coadjoint orbits are always even dimensional.
If › P G ˚ , we know a symplectic structure on its orbit ◊› :
Ê÷◊ pX÷˚ , Y÷˚ q “ ÷prX, Y sq (56.30)
for all ÷ P ◊› . We have

pJ ˚ Ê ◊ qx pX ˚M , Y ˚M q “ ÊJpxq

pX ˚ , Y ˚ q
“ JpxqrX, Y s
(56.31)
“ ⁄M
rX,Y s pxq
“ t⁄M M
X , ⁄Y upxq,

and if M is symplectic, it also fulfils ÊxM pX ˚M , Y ˚M q “ t⁄M


X , ⁄Y upxq, so that
M

J ˚Ê◊ “ ÊM . (56.32)
Théorème 56.11 (Kostant theorem).
Let pM, Êq be a symplectic manifold on which the connected Lie group G has a transitive Hamilto-
nian action. Then the momentum map J : M Ñ G ˚ takes his values in only one orbit ◊ and J is
a covering 1 of G in G ˚ .

Proof. Let us fix a x P M , we pose › “ Jpxq and define Gx , the stabilizer of x in G. Formula
Âpgq “ g · x defines a bijection  : G{Gx Ñ M . So we identify M “ G{Gx . In the same way,
◊› “ G{G› . So we can write Jpxq “ G› and
Jpg · Gx q “ g · Jpxq “ g · G› .
Since g · › “ g · Jpxq “ Jpg · xq “ Jpxq “ ›, we have Gx Ä G› . The nondegenerateness of Ê M
and Ê ◊ makes J nondegenerate because Ê M “ J ˚ Ê ◊ . Then dJ is injective and J is an immersion.
We conclude that Gx “ G› which proves that G› {Gx is discrete and finally that J is a covering.

1. revêtement, en français.
2404 CHAPTER 56. CLASSICAL MECHANICS

56.1.8 Central extension


Let G be a Lie algebra. A Chevalley coboundary is a 2-form which reads ”› for a certain
› P G ˚ with ” defined by
p”›qpA, Bq “ ´›prA, Bsq. (56.33)
Let be a 2-cocycle. If it is not a coboundary, we add an element C in G and we consider
G 1 “ G ‘ C with the Lie algebra structure

rA ` s, B ` tsG 1 “ rA, BsG ` pA, BqC. (56.34)

This is the central extension of G with respect to the 2-cocycle . The terminology comes from
the fact that the extension C belongs to the center of G 1 . The point is that is a coboundary
in G 1 because
` ˘
p”C ˚ qpA, Bq “ C ˚ rA, BsG 1 “ C ˚ rA, BsG ` pA, BqC “ pA, Bq, (56.35)

so that “ ”C ˚ .
Now we suppose that the group G acts on a manifold M . We define the action of the extended
group G1 “ G b e C by saying that the “new” part does not act: pg, sq · x “ g · x. Fundamental
fields remains unchanged:
pX, sq˚ “ X ˚ . (56.36)
If the action of G on M is weakly Hamiltonian, we have functions µx : M Ñ such that ipX ˚ qÊ “
dµX . These functions fulfil X ˚ “ tµX , .u. We define

⁄X,s “ µX ` s. (56.37)

Proposition 56.12.
The action of G1 is (strongly) Hamiltonian for these functions.

Proof. From equation (56.36), we have tµX , .u “ tµX,s , .u hence

t⁄pX,sq , ⁄pY,tq u “ tµX , µY u “ µrX,Y s ` CX,Y (56.38)


` ˘
for certain constants CXY which satisfy the property d tµX , µY u ´ µrX,Y s “ 0. Therefore

t⁄pX,sq , ⁄pY,tq u “ ⁄rX,Y s,CX,Y “ ⁄rpX,sq,pY,tqs . (56.39)

The sense of the whole construction is the following. When the action G is weakly Hamiltonian
on M , we have functions µX which define by

tµX , µY u “ µrX,Y s ` pX, Y q.

In this case, the corresponding group extension has a strongly Hamiltonian action with momentum
maps given by (56.37).

56.2 Constrained system


56.2.1 Holonomic constraints
References are [441, 500]. An holonomic constraint appears when one force a particle to
move for an example on a circle. More formally let Q be the configuration space; an holonomic
constraint is a submanifold N Ä Q, or more generally an integrable subbundle of T Q. From the
inclusion T N Ä T N , the Lagrangian restricts to LN : T N Ñ .
56.2. CONSTRAINED SYSTEM 2405

We assume that LN is a regular Lagrangian (see definition on page 2089) and that N is given
by an equation of the form N “ Ï´1 p0q for a certain section Ï : Q Ñ E ˚ of the dual of a vector
bundle over Q. The variational principle is given by
ª
” LN pq, qqdt
9 “0 (56.40)
` ˘
where the function q is constrained by Ï qptq “ 0 for every t. We enforce that condition by the
Lagrange multiplier method:
ª ´
` ˘ @ ` ˘ ` D¯
” 9
L qptq, qptq ´ ⁄ qptq, t , Ï qptqq dt (56.41)

where ⁄ is a function of pq, tq with values in E. The ” denotes a variation on the curves q in Q
and ⁄ in E. In local coordinates, the Euler-Lagrange equations read
ˆ ˙
d BL BL BÏa
“ i ´⁄ i , Ïa “ 0. (56.42)
dt B q9i Bq Bq
So the Euler-Lagrange equations for LN on N are the sames as the one of L on Q, with the
constraint Ï “ 0.

56.2.2 Primary constraints


The action is given by ª t2
Spqq “ 9
Lpq, qqdt,
t1
and we require the action to be stationary with respect ´
to variations
¯ ”q vanishing on t1 and t2 .
The Euler-Lagrange equations are given by ”q “ Bq ´ dt Bq9 “ 0 or, in coordinates,
”S BL d BL

BL BL BL
0“ ´ q9l l k ´ q:l l k .
Bq k Bq Bq B q9 B q9
when the Hessian matrix is invertible, i.e. when
ˆ ˙
BL
det ‰ 0,
B q9l B q9k
one can solve these equations in an algebraic way for the accelerations q: in functions of the positions
q and the velocities q.
9 When that determinant is vanishing, some of the Euler-Lagrange equations
are simple relations between components of q and components of q, 9 so that the solutions include
some arbitrary functions of time.

56.2.2.1 Example: Maxwell


The action is ª
1
SpAq “ ´ Fµ‹ F µ‹ d4 x
4
with Fµ‹ “ Bµ A‹ ´ B‹ Aµ . That actions has a gauge invariance
Aµ fiÑ A1µ “ Aµ ` Bµ f

for any function f . The equations of motion are B‹ F µ‹ “ 0, in particular the equation Ò ˆ B “
Bt E “ 0 provides the A:k in terms of the Al and A9 µ . Indeed the Hessian
¨ ˛
0
B2 L ` ˘ ˚ 1 ‹
“ ´4 g 0— g –0 ` g –— “ ˚ ‹,
B A9 – B A9 — ˝ 1 ‚
1
has vanishing determinant.
2406 CHAPTER 56. CLASSICAL MECHANICS

56.2.3 Passage to Hamiltonian formalism


The very principle is to define the momentums pk “ BL
B q9k
, so that

Bpk B2 L
“ ,
B q9l B q9l B q9k
so that the vanishing determinant condition makes that it is not possible to solve the velocities
q9k “ q9k pq, pq. That means that the momentums are not independent: there exists relations
„m pp, qq “ 0 (56.43)
for m “ 1, . . . , M . These relations are the primary constraints: they do not rely on the equations
of motion. In other terms, we have ` ˘
9 “0
„m q, ppq, qq
identically, and not only on the solutions.
In the Maxwell example, we pose the momentums fi – “ BL
B A9 –
“ F 0– . Notice that fi 0 ` F 00 “ 0
and that the remaining is fi k “ E k (the electric field).
If one suppose the rank of the Hessian to be constant, the constraints „m pq, pq “ 0 define a
submanifold of the phase space. This is the primary constraint surface.
Suppose that the primary constraints can be divided into one set of independent constraints
B„
„m , m “ 1, . . . , M 1 with the rank of BXm1 being M 1 (X “ pq, pq) and some dependent constraints
„m̄ such that vanishing of all the „m1 implies the vanishing of all the „m̄ .
In that case, we have two important results.

Théorème 56.13.
If a function G vanishes on the constraint surface „m “ 0, then there exists functions g m such that
G “ g m „m .

Théorème 56.14.
If ⁄k ”q k ` µk ”pk “ 0 for every variations ”q k and ”pk tangent to the constraint surface, then
B„m
⁄k “ um (56.44)
Bq k
B„m
µk “ um (56.45)
Bpk
for some functions um .

The tangent assumption means that


B„m k B„m
”„m “ ”q ` ”pk “ 0.
Bq k Bpk
Now the Hamilton equations with „m “ 0 are provided by the action
ª t2
` k ˘
Spq, p, uq “ pk q9 ´ H ´ um „m dt (56.46)
t1
where the functions um are Lagrange multipliers. The equations of motion for p and u provide
expressions pk “ Pk pq, qq
9 and um “ U m pq, qq
9 in a purely algebraic way, so that one has a reduced
action
SL pqq “ Spq, P, U q
which satisfies
”SL ”S
k
“ k,
”q ”q
so that the variational principle for q are the same with SL as with the original S.
One can show that the equations of motion of the action (56.46) read
F9 “ rF, Hs ` um rF, „m s (56.47)
for any function F “ F pq, pq.
56.3. LIE GROUPOIDS AND ALGEBROIDS 2407

56.2.4 Secondary constraints


For consistency, one can impose that the constraints are preserved in time: „9 m “ 0. Using the
evolution equation (56.47) on the function „m , one finds, for each m,

„9 m “ r„m , Hs ` un r„m , „n s

which are relations between q and p. In the case where that relation is independent of the „m , one
has a secondary constraint Xpp, qq “ 0 which has to be preserved with the time, so that

rX, Hs ` un rX, „n s “ 0.

Once again, this is a relation between p and q.

56.3 Lie groupoids and algebroids


The reference for this section are [501, 502].
Let M be a manifold and A, a vector bundle over M . Let r., .s be a Lie algebra structure on A.
Then A becomes a Lie algebroid when we endow it by a homomorphism fl : A Ñ T M such that
(1) it induces a Lie algebra homomorphism on the sections, i.e. a fl : pM, Aq Ñ pT M q such
that flrÂ, Ïs “ rflpÂq, flpÏqs,
(2) for all f P C 8 pM q and Â, Ï P pM, Aq,
` ˘
rf Â, Ïs “ f rÂ, Ïs “ f rÂ, Ïs ´ flpÏqf Â. (56.48)

The map fl is the anchor. In order to be more precise for item (1), the anchor function fl : A Ñ T M
induces the homomorphism fl1 : pAq Ñ pT M q by formula

fl1 pÂqpxq “ flpÂpxqq. (56.49)

We will omit the prime and simply write fl for both functions.
The algebroid A admit standard coordinates pq, ⁄q where q are coordinates on M and ⁄ some
coordinates on the fibres (whose are vector spaces). These coordinates depends on a choice of a
local base › of sections of A: the set of sections t›i pxqu is a basis of the fibre Ax for each x P M .
IN terms of these coordinates, the Lie algebra structure of A reads

r›i , ›j s “ cijk ›k

where the cijk P C 8 pM q are the structure constant. In the same way, the anchor is given in terms
of functions aij P C 8 pM q by
B
flp›i q “ aij .
Bqj

56.3.1 Example: tangent bundle T M


If we consider A “ T M , the anchor is fl “ Id. Then formula (56.49) is flpÂqpxq “ Âpxq, and
the condition (56.48), with more familiar notations, reduces to

rf X, Y s “ f rX, Y s ´ pY f qX.

56.3.2 Example: a Lie algebra


Any Lie algebra G can be seen as an algebroid on a manifold M “ tpu is a manifold containing
only one point. In this case, any path in M is constant and the tangent space T M reduces to only
the zero vector. The identically vanishing map fl : G Ñ T M is an anchor.
2408 CHAPTER 56. CLASSICAL MECHANICS

56.3.3 Example: gauge algebroid


We consider the following principal bundle:

G /P



M

Let’s begin to give a structure of vector bundle on the quotient T P {G. What is clear is that T P
is a vector bundle p : T P Ñ M . The action of G on T P is

X› · g “ d·g X›

where · : P Ñ P is the action. So the quotient T P {G is taken with respect to the equivalence
X› „ d·g X› . Notice that X› P T› P and d·g X› P T› · g P ; in fact in each class rX› s, there is one
and only one vector at each point of the fibre of ›.
The differential dfi of the projection passes to quotient: dfipX› q “ dfipX› · gq. It follows that
the next definition is correct:
dfi : T P {G Ñ T M
(56.50)
dfirX› s “ dfiX› P Tfip›q M.
Now let us study the sections  : M Ñ T P {G.

Lemme 56.15.
The sections of T P {G are the G-equivariant vector fields on P . (see definition 53.15.)

Proof. We consider the map L : pM, T P {Gq Ñ pP, T P q,

pLÂqp›q “ Âpfip›qq|›

where, when q P T P {G, the symbol q|› denotes the element of T› P which belongs to q. In
particular, rX› s|› “ X› . The section LÂ is G-equivariant, i.e.

d·g pLÂqp›q “ pLÂqp› · gq. (56.51)

Indeed, d·g X› P rX› s, thus “ ‰


d·g Âpfi›q |› P rÂpfi›qs X T› · g P,
“ ‰
and there is only one element which is Âpfi›q |› · g “ pLÂqp› · gq.

Now, taking dfi as anchor, this construction gives an algebroid structure to T P {G. In order to
prove that we have to prove that
` ˘
rf Â, Ïs “ f rÂ, Ïs ´ flpÏqf Â

for any choice of sections Â, Ï : M Ñ T P {G and of function f : M Ñ . Here we have

fl : pM, T P {Gq Ñ pM, T M q


` ˘ (56.52)
flpÂqpxq “ dfi Âpxq .

The first think to be remarked is that a good choice of local section a : M Ñ P and vector fields
X, Y P pP, T P q on P , one can express  and Ï under the form
“ ‰ “ ‰
Âpxq “ pX ˝ aqpxq , Ïpxq “ pY ˝ aqpxq (56.53)

with the same a. On T P {G, we put the Lie bracket inherited from the one of T P :
” ı
rÂ, Ïspxq “ rX, Y s ˝ apxq . (56.54)
56.3. LIE GROUPOIDS AND ALGEBROIDS 2409

If X corresponds to  by (56.53), we have


” ı
pf Âqpxq “ f pxqpX ˝ aqpxq
”` ˘` ˘ı
“ pf ˝ ÂqX apxq ,

thus pf ˝ fiqX corresponds to f  in the sense of


´ ¯
pf ˝ fiqX p›q “ f pfi›qXp›q.

Then we have
”“ ‰ ı
rf Â, Ïspxq “ pf ˝ fiqX, Y ˝ apxq
”` ˘ ı
“ pf ˝ fiqrX, Y s ´ Y pf ˝ fiqX ˝ apxq
“ ‰ “ ‰
“ f pxqrX, Y s ˝ apxq ´ Y pf ˝ fiqX ˝ apxq .

What lies in the bracket of the second term reads better under the form:
` ˘
Y pf ˝ fiqX ˝ apxq “ Yapxq pf ˝ fiqXapxq .

We want this thing to be equal to


´` ˘ ¯ ` ˘
flpÏqf  pxq “ flpÏqpxqf Âpxq
` ˘ “ ‰
“ dfi Ïpxq f pX ˝ aqpxq
“ ‰ “ ‰
“ dfi pY ˝ aqpxq f pX ˝ aqpxq .
“ ı
In the latter expression, pY ˝ aqpxq P Tapxq P while dfi pY ˝ aqpxq P Tx M . Since dfi : T P Ñ T M
fulfils dfipX› · gq “ dfiX› , the differential of fi passes to the classes (this is the reason for which
we chose it a anchor) and
dfirpY ˝ aqpxqs “ dfipY ˝ aq P T M.

It remains to be proved that


”` ˘ ı ` ˘ “ ‰
Y pf ˝ fiqX ˝ apxq “ dfipY ˝ aqpxq f pX ˝ aqpxq ,

which is true because

Yapxq pf ˝ fiq “ pY ˝ aqpxq
“ pdf ˝ dfiqpY ˝ aqpxq
“ dfipY ˝ aqpxqf.

56.3.4 Poisson structure


Let us describe a Poisson structure on the dual A˚ . For this, we put a Poisson bracket on
each A˚x , x P M . Since the bracket only depends on differential, we just have to define it on affine
functions on the fibres. Two remarks: firstly, the functions which are constant on fibres can be
seen as functions on M and second, linear functions can be seen as sections of A because a linear
function on a vector space is equivalent to the data of a single vector. So let f, g : M Ñ be two
functions and Â, Ï : M Ñ A be sections. The bracket on C 8 pA˚ q is defined as

tf, gu “ 0, tf, Âu “ flpÂqf, tÂ, Ïu “ rÂ, Ïs. (56.55)


2410 CHAPTER 56. CLASSICAL MECHANICS

56.4 Lagrangian formalism


Let A be a Lie algebroid on a manifold M and a function L : A Ñ which we will call
Lagrangian. The Legendre mapping is the fibre derivative F L : A Ñ A˚ given by

F Lpaq P A˚fipaq , F Lpaqpbq “ dLa pbq. (56.56)

More precisely, the differential of L at a P A is a map from Ta A. In order to define F Lpaq P A˚ ,


we begin to consider the restriction L|a of L to the fibre of a. The differential of L|a : Afipaq Ñ is

pdL|a qa : Ta Afipaq Ñ ,

but the fibre Afipaq is a vector space which can therefore be identified with its dual space. Then
we have
pdL|a qa P A˚fipaq Ä A˚ .
When one has a Lagrangian on A, one define the action as the function A : A Ñ ,

Apvq “ xF Lpvq, vy, (56.57)

i.e. the action of F Lpvq P A˚fipvq on v P Afipvq . The energy is the function

E “ A ´ L.

The Lagrangian L is a regular Lagrangian if F L is a local diffeomorphism. In this case, one can
bring the Poisson structure of A˚ on A. The resulting Poisson structure on A is the Lagrange-
Poisson structure. For this, we have to define t›, ÷u when ›, ÷ P C 8 pAq. The natural definition
is
t›, ÷uA :“ tF Lp›q, F Lp÷quA˚ (56.58)
with the following definition of F Lp›q P C 8 pA˚ q:

F Lp›qpÊq “ ›pF L´1 pÊqq. (56.59)

The latter definition says that when Ê P A˚fipaq the element F L´1 pÊq is the element a such that
pdL|a qa “ Ê (this is an equality in A˚fipaq ).
Since we have a Poisson structure, we can consider the Hamiltonian field (see definition (56.5))
corresponding to the energy function E : A Ñ . This is a vector field on A. This field is the
Lagrangian vector field
In standard coordinates (see page 2407), the Lagrangian is a function pq, ⁄q fiÑ Lpq, ⁄q. In
order to build F L, we first restrict F to a fibre; so L becomes a function ⁄ fiÑ Lpq0 , ⁄q and thus
the derivatives which appear in F L are
BL
µi “ .
B⁄i
Now we try to express the Poisson structure on A in the standard coordinates. Fist, qi is a function
on A which associates to one point the component i of its projection. If Ê and ÷ both belong to
A˚x , the elements F L´1 pÊq and F L´1 p÷q both belongs to the same fibre Ax . Thus F Lpqi q is a
function that is constant on the fibres. So

tqi , qj u “ 0.

56.5 Groupoids
A set is a groupoid when we consider some maps
— –, — : Ñ 0 Ä ,
— m : 2 Ñ where 2 “ tpx, yq P ˆ tel que —pyq “ –pxqqu,
— i: Ñ
56.5. GROUPOIDS 2411

such that
(1) mpx, mpy, zqq is defined if and only if mpmpx, yq, zq is defined, and in this case, they are
equal,
(2) mp—pxq, xq “ mpx, –pxqq “ x,
(3) mpx, ipxqq and mpipxq, xq are defined to be respectively equal to —pxq and –pxq.
Notice that the fact for mpx, yq to be defined means that –pxq “ —pxq. As notations and terminol-
ogy, we adopt the following conventions. The map i is the inversion, m is the multiplication and
is denoted by a dot: mpx, yq “ x · y. We also write ipxq “ i´1 .
The first hypothesis is called associativity.

Lemme 56.16.
@ x P , we have

–p—pxqq “ —pxq (56.60a)


—p–pxqq “ –pxq. (56.60b)

Proof. From definition of a groupoid, —pxq · x “ x, so –p—pxqq “ —pxq. The other statement
comes in the same way.
Notice that the conclusion does not come from the equality —pxq · x “ x, but only from the
existence of the product —pxq · x.

Lemme 56.17.
If px, yq P 2 , we have

—px · yq “ —pxq (56.61a)


–px · yq “ –pyq (56.61b)

Proof. Since px, yq P 2, one can write x · y. Using x “ —pxq · x, and associativity we find:
` ˘
x · y “ —pxq · x · y “ —pxq · px · yq.

Existence of the last product implies –p—pxqq “ —px · xq. Using the equality –p—pxqq “ —pxq, we
find the second relation. The first one is proven by the same:
` ˘
x · y “ x · y · –pyq “ px · yq · –pxq.

Lemme 56.18.
For all x P , we have

–p–pxqq “ –pxq (56.62a)


—p—pxqq “ —pxq. (56.62b)

Proof. Using formula x · –pxq “ x in itself, and using associativity,


` ˘
px · –pxqq · –pxq “ x · –pxq · –pxq “ x.

The existence of the product –pxq · –pxq and the fact that — ˝ – “ – give the result.

Lemme 56.19.
Let us mention the following other properties:

–px´1 q “ —pxq —px´1 q “ –pxq (56.63)


–pxq · –pxq “ –pxq —pxq · —pxq “ —pxq (56.64)
2412 CHAPTER 56. CLASSICAL MECHANICS

the simplification rule:

x · y 1 “ x · y2 ñ y 1 “ y 2 (56.65)
x1 · y “ x2 · y ñ x1 ´ x2 , (56.66)

and the corollary


px´1 q´1 “ x. (56.67)

Proof. No proof.

Proposition 56.20.
The set 0 is the set of fixed points of – and —.

Proof. We proof that 0 “ tx P tel que –pxq “ xu; the same is true for —. The definition
of 0 is to be the image of –. Let x P 0 : there exists a y P such that x “ –pyq. So
–pxq “ –p–pyqq “ –pyq “ x.
For the reciprocal, let x “ –pxq. Then x P 0 because x belongs to the image of – — for
instance, the image of itself.

56.5.1 Example: when 0 “ teu


Let us prove that in the case when 0 reduces to only one point teu, the groupoid is a group
whose unit is e.
First, remark that for all x P , we have –pxq “ —pxq “ e, so that 2 “ and the multiplication
is everywhere defined. Associativity is not a problem. The map x ބ ipxq is the inverse because
x · ipxq “ —pxq “ e and ipxq · x “ –pxq “ e. We also have x · e “ e · x “ x because e “ –pxq
and x · –pxq “ x.

56.5.2 Example: the null groupoid


A null groupoid is a groupoid in which 0 “ . In this case, since 0 is the set of fix points
of – and —, we have
– “ — “ Id .

In order for x · y to exist, we need —pyq “ –pxq, which in the case of the null groupoid gives x “ y.
So the only products that are defined are

x · x “ x.

56.5.3 The case – “ —


The case – “ — regroup the two preceding cases. We will prove that for each u P 0 , the set
–´1 puq is a group with u as unit. Indeed let x, y P –´1 puq; it is clear that x · y exists because
—pyq “ –pyq “ x “ –pxq. So the product is defined everywhere. Proof of the fact that u is the
unit is easy:

x · u “ x “ x · –pxq “ x
u · x “ —pxq · x “ x,

and

x · ipxq “ —pxq “ u
ipxq · x “ –pxq “ u.
56.6. LIE GROUPOID 2413

56.5.4 An example on a vector bundle


We consider a vector bundle fi : E Ñ M and

0 “ tox tel que x P M u,

the set of the zero of each fibre. As groupoid law, we choose the addition in the fibres: when
fipvq “ fipwq, we define v · w “ v ` w, and as map – “ —, we naturally choose

–pvq “ —pvq “ ox

if v P Ex . In this case, –´1 poq is the fibre of o which is a group for the addition.

56.5.5 Orbits
Let be a groupoid and u P 0. The isotropy group of u is

u “ –´1 puq X — ´1 puq

In order to prove that it is a group, remark that if x, y P u,

–pxq “ –pyq “ —pxq “ —pyq “ u,

in particular, x · y exists and u has a law. It is easy to prove that u is th unit.


Notice that —p–´1 pxqq “ –p— ´1 pxqq. Indeed if y P —p–´1 pxqq, there exists a z such that
y “ —pzq and –pzq “ x. We have to find a z 1 such that y “ –pz 1 q and —pz 1 q “ x. The element
z 1 “ —pzq works.
The set —p–´1 pxq “ –p— ´1 pxqq is the orbit of trough x.

Proposition 56.21.
The set of orbits of is a partition of 0.

Proof. The proof is as easy as I want not to give you.

56.5.6 Morphism
If and 1 are two groupoids, the map f : Ñ 1 is a morphism if for each existing product
x · y in , we have
f pxq · f pyq “ f px · yq.
We see that automatically

— 1 pf pxqq · f pxq “ f pxq “ f p—pxq · xq “ f p—pxqq · f pxq;

so that the simplification rule gives

—1 ˝ f “ f ˝ — (56.68a)
1
– ˝ f “ f ˝ –. (56.68b)

56.6 Lie groupoid


A Lie groupoid is a (maybe non Hausdorff) manifold endowed with a groupoid structure such
that
(1) the set 0 is a Hausdorff manifold,
(2) the maps – and — are differentiable submersions,
(3) the multiplication m : 2 Ñ is differentiable,
(4) the inversion x Ñ x´1 is a diffeomorphism.

You might also like