Chapter 1

Classical Mechanics
1.1 Lagrangian Mechanics
1.1.1 Newtons mechanic in R
n
Recall that according to the Newtons law
F = ma
where F is the sum of forces acting on a point particle of mass m
and a is its acceleration. If is the trajectory of such particle, the
acceleration is a =
1
. The force F is called a potential force if
F =
U
q
, where U(q) is the potential, or potential energy.
Thus, Newtons law, describing the motion of a point-wise par-
ticle in the force eld with the potential U(q) is equivalent to the
second order dierential equation for the trajectory of the particle:
m (t) =
U
q
_
(t)
_
Thinking of a solid as a collection of constrained points, one can
describe the motion of solids as constrained solutions to Newtons
equations.
1
Here we will use notations q =
dq
dt
, and q =
d
2
q
dt
2
, etc., which are standard
in classical mechanics
1.1.2 Constrained motion and Newtonian me-
chanics on manifolds
The motion of a mechanics particle can be constrained to a sub-
manifold. Physically it can be achieved by a potential force which
does not allow the particle leave a thing neighborhood of this sub-
manifold. Newton equations for a particle in such potential eld
give the Newton equations on the submanifold in the limit when
the width of the neighborhood goes to zero.
The resulting equation sis best to describe in terms of the metric
induced on the submanifold by its embedding in R
n
. Let N R
n
be the submanifold, denote by x
a
local coordinates on S and by
g
ab
(x) the metric tensor on N. In local coordinates

x
a
is the
basis in tangent spaces and g
ab
(x) = (

x
a
,

x
b
)
x
where (, )
x
is
the scalar product on the tangent spaces to S, i.e. the metric on
N.
The Newton equations for a particle constrained to S are
d
dt
(g
ab
(x(t)) x
a
) =
U
x
a
(x) (1.1.1)
where U(x) is the potential (on N).
The derivation of these equations from Newtons equations in R
n
1
is an exercise.
The equations (1.1.1) do not require an isometric embedding of
N into R
n
. They are written in terms of intrinsic characteristic of
N. So they can be considered as the extension of the denition of
Newtons equations from R
n
to an arbitrary Riemannian manifold.
The manifold N is called the congurations space. The tangent
vector (t) T
(t)
N to a trajectory (t) is a tangent vector to N
at the point (t) N.
1.1.3 Lagrangian formulation of Newtons me-
chanics on TN
Lets reformulate these dierential equations as a variational prob-
lem. Assume the particle is moving on a manifold N, which well
refer to as conguration space.
A Lagrangian function of a mechanical system is a smooth func-
tion /(, q) on the tangent bundle TN, q N and T
q
N.
For example, if N is equipped with a Riemannian metric, we can
choose
/(, q) =
(, )
2
U(q) (1.1.2)
where (, ) is the Riemannian metric, and U is the potential func-
tion.
For a parameterized path = ()
t
=0
in N, dene the action
functional
/[] =
_
t
0
/
_
(), ()
_
d.
Let us show that Newtons equations are Euler-Lagrange equa-
tions for the action with the Lagrangian(1.1.2), i.e. they describe
extrema of the action functional.
Heuristically the (rst) variation is the innitesimal change in
action when is changed innitesimally.
/[] = /[ +] /[]
More precisely, if f is a function on paths, and
s
is a family
of parametrized paths such that
0
= the variation of f in the
direction of
s
at s = 0 is
f() =
df(
s
)
ds

s=0
Dene =
ds
ds
[
s=0
. This is a vector eld along . Now for the
variation of the action we can write
/[] =
d/[
s
]
ds

s=0
=

i
_
t
0
_
/

i
( , )
i
() +
/
q
i
( , )
i
()
_
d
=

i
_
t
0
_

d
dt
/

i
( , ) +
/
q
i
( , )
_

i
() d
. .
bulk term
+

i
_
/

i
_
(t), (t)
_

i
(t)
/

i
_
(0), (0)
_

i
(0)
_
. .
boundary terms
These expressions written in local coordinates. Coordinate free
formulation in given in the Appendix ??. We call the rst term the
bulk term and the last terms boundary terms.
If is a solution to the Euler-Lagrange equations

d
dt
/

i
( , ) +
/
q
i
( , ) = 0 (Euler-Lagrange equations)
then the bulk term vanishes. The variation of the action /[]
is given now only by boundary terms.
It is easy to see that the Euler-Lagrange equations for (1.1.2)
are exactly the Newtons equations. The Euler-Lagrange equations
have complete local solution if the matrix of second derivatives

2
L

j
(, q) is non-degenerate for all (, q).
1.1.4 Boundary conditions
Usually, a mechanical problem requires a solution of a boundary
problem, i.e. Euler-Lagrange equations should be supplemented by
a boundary condition.
Typically, a boundary condition is determined by a submanifold
B TN TN. A solution to such boundary problem is a clas-
sical trajectory
cl
(i.e. a solution to the Euler-Lagrange equa-
tions) with initial and target points of the trajectory being in B:
_
(t), (t), (0), (0)
_
B.
We will consider only boundary problems of the form B = B
out

B
in
where B
in
are initial conditions at = 0 and B
out
are target
conditions when = t.
Example 1. Boundary value problem when the position is xed
at time 0 and time t: (0) = Q and (t) = Q

. In this case,
B = T
Q
N T
Q
N. If we vary paths within the family satisfying
the constraint
_

s
(t),
s
(t),
s
(0),
s
(0)
_
B, then we see that (t)
and (0) are zero. So the boundary terms in /[] vanish. Thus,
the extrema of /[] (subject to the boundary conditions) are pre-
cisely solutions to this boundary problem. This is an example of
what we call variational boundary conditions.
Note that in general if we look for extrema of the action functional
on paths satisfying certain boundary conditions we may not nd
any. The bulk term with vanish on solutions to Euler-Lagrange
equations, but there could be no solutions which would satisfy both,
the boundary conditions, and the vanishing of boundary terms in
the variation of the action. To remedy this, one can try to add to
the action functional some boundary terms. It works in some cases.
Example 2. Boundary problem when the position is xed at
= 0, (0) = Q, and the velocity is xed at = t: (t) = V .
In this case, B = V (N) T
Q
N, where V (N) is the image of the
vector eld V : N TN in TN. Now, because we xed (0), we
have (0) = 0 , so the remaining boundary terms are
d/
_
(t)
_
=

i
/

i
_
(t), (t)
_

i
(t)
So extrema of our action functional are not solutions to the bound-
ary problem. But now consider the modied action functional

/[] = /[]

i
P
i

i
(t).
The boundary condition P
i
=
L

i
_
(t), (t)
_
now is variational for
the modied action functional.
However, this does not work for all types of boundary conditions.
Here is an example.
Example 3. The initial value problems: when the initial and
coordinate and the initial velocity are xed (0) = Q and (0) = V .
In this case, B = TN(V, Q). This problem does not have a vari-
ational reformulation. The action functional can not be modied by
adding extra terms depending on the boundary characteristic of the
path such that the extrema of modied action solve the boundary
problem.
We will see later that there is a simple description of variational
boundary problems in terms of Lagrangian submanifolds in sym-
plectic manifolds.
1.2 Legendre transform, Hamiltonian
formulation
1.2.1 The Legendre transform
Denition 1.2.1. A smooth function / C

(TN) is non-
degenerate at q N if det
_

2
L
ij
_
,= 0 for all T
q
N.
Denition 1.2.2. / is strongly non-degenerate at q N if
_

2
L
ij
_
is positive denite for all T
q
N.
For a Lagrangian function / C

(TN), dene its Hamiltonian

function 1 C

(T

N) as the Legenedre transform

1(p, q) = max

TqN
_
p(

) /(

, q)
_
.
If the function / is strongly positive, it is convex and there is
unique maximum which is also unique extremum if / is smooth.
Thus, the maximum is achieved at

which solves the equation
p(

) =
/

, q). (1.2.3)
Denition 1.2.4. The Legendre transfrom is the map T
q
N
T

q
N given by (, q)
_
d

/(, q), q
_
.
If / is strongly non-degenerate (resp. non-degenerate), then the
Legendre transform is an isomorphisms (resp. local isomorphism).
In particular, equation (1.2.3) has a unique solution

for a given
(p, q).
Remark 1.2.5. Clearly, the denition of the Legendre transform
of / also works when it is continuous but not necessary smooth
convex function.
The Legenedre transform of the Hamiltonian

/(, q) = max
pT

q
N
_
p() 1( p, q)
_
.
in general does not give the function / but the convex hull of / (see,
e.g. [?]). If / is berwise convex (for all , each q), then

/ = /.
If we assume / is smooth, convexity is equivalent being strongly
non-degenerate.

In our applications, / will either be strongly non-degenerate,

which insures that it is berwise convex, or the Hessian
_

2
L
ij
_
will be identitcally zero, that is / is degenerate.
1.2.2 Hamiltonian formulation of Lagrangian
mechanics
Theorem 1.2.6. The image of classical trajectories on TN in
T

N (with respect to the Legendre transform) are solutions to the

following rst order system.
p
i
=
1
q
i
q
i
=
1
p
i
Where 1 is the Hamiltonian of the system (given by the Legendre
transform of the Lagrangian).
Proof. The proof is a straightforward exercise.
Well see a coordinate-free formulation of this theorem later
(Corollary 1.2.7).
The symplectic form on T

N.
Corollary 1.2.7 (to Theorem 1.2.6). The image x() in T

N
of a classical trajectory
_
(), ()
_
on TN (with respect to the
Legendre transform) is a ow line of the Hamiltonian vector eld
v
H
=
1
(d1). In other words, x() =
1
(d1)(x()) .
The main moral of this transformation is that under the Legendre
transform, the trajectories become ow lines of the Hamiltonian
vector eld.
1.3 Hamiltonian dynamics on symplec-
tic manifolds
1.3.1 Hamiltonian dynamics on symplectic man-
ifolds
The Hamiltonian mechanics exists not only on M = T

N but any
symplectic manifold. Some basic notions about symplectic mani-
folds are reminded in Appendix ??.
Let (M, ) be a symplectic manifold. Fix the Hamiltonian
function H C

(M). It denes the Hamiltonian vector eld

v
H
=
1
(dH).
The Hamiltonian dynamical system generated by H is a dynam-
ical system with trajectories being ow lines of the Hamiltonian
vector eld v
H
. Flaw lines are solution to the system of dierential
equations
x(t) = v
H
(x(t)). ()
Assume that for suciently small t and for each x M there
exists unique solution to these equations. Then we can write x(t) =
F
t
(x) for the ow line passing through x at t = 0. Here F
t
is a
dieomorphism of M. For f C

(M), dene f
t
= F

t
(f), i.e.
f
t
(x) = f
_
x(t)
_
. This gives a local action of R on C

(M), i.e an
evolution in the space of functions.
Theorem 1.3.1.
1.
dft
dt
= H, f
t
for f
0
= f. Now even though the space is innite-
dimensional, the equations of motion are linear.
2. F

t
(f, g) = F

t
(f), F

t
(g) (it is clear that F

t
(fg) =
F

t
(f)F

t
(g), so F

t
is a Poisson algebra homomorphism).
[3.] An innitesimal version:
H
f
def
= H, f is a derivation of the
Poisson algebra (C

(M), , ).
Proof. (1)
df
t
dt
=
d
dt
f(x(t)) =
_
x(t), df(x(t))
_
=
_
v
H
(x(t)), df(x(t))
_
=
_

1
(dH), df(x(t))
_
=
1
(dH df
t
) = H, f
t

(3), then we can exponentiate it to get (2).

H
(f, g) =
H
f, g +f,
H
g
H, f, g = H, f, g +f, H, g
which is just the Jacobi identity for , .
Proposition 1.3.2. v
H
is tangent to the level surfaces of H.
In other words, the dynamics is parallel to the level surfaces of
H; if a trajectory originates on a particular level surface, it will
remain there. This is good news because instead of solving dier-
ential equations on a 2n-dimensional manifold, we can solve them
on a (2n 1)-dimensional manifold. The level surfaces of H are
physically surfaces of constant energy. Notice that this proposition
is true only for conservative systems, when the Lagrangian (and
thus the Hamiltonian) do not depend on time. This is also known
as conservation of energy.
This property of the Hamiltonian function inspires the following
denition.
Denition 1.3.3. A function G C

(M
2n
) is an integral of the
Hamiltonian dynamics generated by H if
F

t
(G) = G
equivalently (because of the Theorem 1.3.1), H, G = 0. That
is, G is i n involution with H.
When we have such a function, it is clear that the level surfaces
G
(c)
= x M[G(x) = c are also preserved by evolution (i.e. v
H
is tangent to G
(c)
).
If the system is conservative i.e., if the Hamiltonian does not
depend explicitly on time, we have at least one integral (namely H).
If another integral G exists, then we know that if a trajectory starts
at the intersection H
(E)
G
(c)
of level surfaces, it will remain there.
So one can reduce the dynamics to a (2n 2)-dimensional space.
In general, if the system has lots of integrals, one can signicantly
reduce the dimension of our space.
1.3.2 Integrable Hamiltonian ows
Let (/
2n
, ) be a symplectic manifold and H C

(/) be a
Hamiltonian generating vector eld v
H
=
1
(dH).
Denition 1.3.4. The system is called integrable if there exist n
functionally independent
1
F
1
, . . . , F
n
C

(/) such that

F
i
, F
j
= 0 = F
i
, H.

Recall that F, G (dF dG,

1
), where
_
2
T

/ is the
symplectic form and
1

_
2
T/ is the corresponding Poisson
tensor.
Algebraically this means that functions F
1
, F
2
, . . . , F
n
generate
a maximal commutative Poisson subalgebra in the algebra of func-
tions on /, on which the Hamiltonian evolution generated by H
acts trivially.
Functions F
1
, . . . , F
n
are called rst integrals of the system.
Geometrically, n independent functions in involution (Poisson
commutative) determine the Lagrangian bration F : /
R
n
, F(x) = (F
1
(x), . . . , F
n
(x)) where ( the image of F) is an n-
dimensional open subvariety in R
n
. The Hamiltonian H generate
integrable dynamics if the ow lines of v
H
are parallel to the bers
1
This means that dF1 . . . dFn = 0 on an open dense subset of M.
(equivalent to commutativity of H with F
i
). Clearly ber of F are
level surfaces of functions F
1
, . . . , F
n
:
M
c
= F
i
(x) = c
i
[x M.
We will also denote level surfaces by F
1
(c).
Theorem 1.3.5 (Arnold-Jost-Liouville). For an integrable system
on a symplectic manifold the following holds:
i) Level surfaces are invariant with respect to the ow of H.
ii) Each generic level surface has a canonical ane structure:
i.e. an atlas on M
c
such that transition functions are ane
transformations of R
n
(compositions of translations and linear
transformations), such that ow lines of the Hamiltonian are
linear in this coordinates.
iii) Connected components of level surfaces are tori for compact
manifolds.
The ane structure in the second statement is determined by the
ow lines of the Hamiltonians F
1
, . . . , F
n
. The ane coordinates on
level surfaces are called usually the angle variables. The coordinates
parameterizing level surfaces are called action variables.
If (
1
, . . . ,
n
) is an such ane coordinate system then aw lines
of H are linear:

i
(t) =
i
(H, c
1
, . . . , c
n
)t +
i
.
The frequencies
i
are important characteristics of the system.
Finally, it is worth to emphasize that integrable Hamiltonian sys-
tems have more functorial ingredients then Hamiltonian systems.
A functorial, geometrical, ingredient of a Hamiltonian system is a
symplectic structure on a manifold (on a phase space ) of a system.
In the integrable case we have not only this but also a Lagrangian
bration on this manifold.
In this sense from geometrical point of view integrable systems
are Lagrangian bration on symplectic manifolds.
Specic integrable Hamiltonian ows are, usually, interesting
only because of their applications to some other problems lying
outside of the realm of symplectic geometry. We will see some ex-
amples of such systems in the next lecture.
1.4 Variational principle and Hamilto-
nian mechanics
Assume / is strongly non-degenerate, so it gives an isomor-
phism between tangent and cotangent bundles. Dene p() :=
L

_
(), ()
_
. Recall that the Hamiltonian is a function on
T

N given by 1
_
p(), ()
_
= p(

) /
_

, ()
_
, where

satis-
es p =
L

, q) (that is, = ()). So we have that

/
_
(), ()
_
= p
_
()
_
1
_
p(), ()
_
.
We can write the action as
/[

] := /[] =
_
t
0
_
p()( ()) 1
_
p(), q()
__
d
=
_

_
t
0
1
_
p(), q()
_
d
where

= p(), ()
t
=0
is the image of = (t), (t) in
T

N under the Legendre transformation. The rst term only de-

pends on

and the parameterization of , and the second term

really describes the dynamics.
The variation of the action on a classical trajectory can now be
written now as
/[

cl
] =

i
/

i
_
q(t), q(t)
_
q
i
(t)(t 0) =
_
di(q(t))
_

_
di(q(0))
_
where (q) is the value of the form on the tangent vector di(q),
i : N T

N is the zero section, and di : TN T(T

N) is its
dierential.
The goal now is to show that solutions to the Euler-Lagrange
equation dene the Lagrangian submanifold in TN TN endowed
with the symplectic form =

2
where
1
and
2
are
projections to the rst, respectively second, copy of T

M and is
the symplectic form on T

M.
Let

be a classical trajectory in the phase space originating at

q N at time t
1
and ending at time t
2
at Q N. We can evaluate
the action on such a trajectory, giving a function on NNRR
/[

] = /

(q, Q, t
1
, t
2
).
Theorem 1.4.1 (Hamilton-Jacobi). For each as above we have
/

t
2
+H(d
Q
/

, Q) = 0
/

t
1
H(d
q
/

, q) = 0.
Exercise 1 Finish the proof of the theorem ?? and complete the
proof of the theorem 1.2.6
1.4.1 Variational boundary problems
If we want to modify the action function such that solutions to the
boundary problem B TN TN will be exact extrema of the
modied action. In other words we want the whole variation to
vanish, including boundary terms, not only the bulk terms, which
give the Euler-Lagrange equations. Consider the modied action
functional
/
F,B
[] = /[] +F
for some function F on B. Its variation is
/
F,B
[
cl
] = boundary terms[
B
+F = 0
where
cl
is a solution to the Euler-Lagrange equations and the
family
s
is contstrained by the boundary conditions.
We want this to be identically zero.
If we use coordinates (V, v) on T
Q
NT
q
N, and local coordinates
x
a
on B this is equivalent to F satisfying the dierential equations
F
x
a
=

i
/

i
_
V (x), Q(x)
_Q
i
(x)
x
a
+

i
/

i
_
v(x), q(x)
_q
i
(0)
x
a
.
The identity
__

x
b

i
/

i
_
V (x), Q(x)
_Q
i
(t)
x
a
_

_
(V, Q) (v, q)
_
_

_
a b

= 0
is a necessary condition is that such F exists. This identity ensures
that F exists locally, i.e. that the form

a
(

i
/

i
_
V (x), Q(x)
_Q
i
(x)
x
a
+

i
/

i
_
v(x), q(x)
_q
i
(0)
x
a
)dx
a
is closed. Once this condition is satised, the obstruction to nding
such function F is an element of H
1
dR
(B)
The physical systems weve described so far are called conserva-
tive, meaning that / is independent of time. More generally, we
could take / to be a smooth function on TN R. In this case, the
action functional is
/[] =
_
t2
t1
/
_
(), (),
_
d.
Most of the previous analysis also works for such systems. We will
discuss such systems later.
Exercise Prove that solutions to the Euler-Lagrange equation
for the Lagrangian /(, q) =
1
2
(, ) on a Riemannian manifold are
geodesics.
Remark If N = R
n
with the at metric, solutions to the Euler-
Lagrange equations are straight lines (t) = A + Bt. In this case
there is a unique solution connecting any two points on the plane
On S
2
with constant positive curvature, two points which are
not opposite to each other are connected by two geodesics. Two
opposite points are connected by innitely many geodesics. This
example demonstrates that boundary problems may have many so-
lutions.
1.5 Variational principle and La-
grangian submanifolds
Recall from lecture 2 that given a boundary condition B TN
TN, we described the function F on B. Such that the modied
action functional /
F
= / + F is extremal (including bulk and
boundary terms) on trajectories originating and ending on B.
Such function F should satisfy the condition
dF(V, Q, v, q) =
/

(v, q)dq
/

(V, Q)dQ = p dq P dQ.

Here (V, Q, v, q) B, p =
L

(v, q), P =
L

(V, Q), and

(P, Q, p, q) B

.
From here we conclude that the necessary condition for such F
to exist is the closedness of the form on the right, i.e. on B

one
should have dP dQdp dq = 0. If x
a
are local coordinates on
B

this translated to

i
_
P
i
x
a
Q
i
x
b

q
i
x
a
q
i
x
b
_
[a b] = 0
Thus, a necessary condition for existence of such an F is that B

should be an isotropic submanifold in T

N T

N, i.e.:
[
B
= 0
where =

2
is the symplectic form on T

N T

N.
1.5.1 Variational principle contd
Now lets look at the E-L solutions. From the theory of ODEs, we
know that generically one need to have 2n conditions to get nitely
many solutions. Thus, if we want generically nitely many classical
trajectories connecting two points from B, the boundary manifold
B

should be Lagrangian.
Thus, a boundary condition B is variational and the number of
solutions is generically nite only if B

is a Lagrangian submanifold
in T

N T

N with the symplectic form .

For each extremum of / with xed end points (Q, q) dene the
function /

= /[] on a neighborhood of (Q, q) N N. Assume

that each of these functions extends to a function on N N.
Remark 1.5.1. In a free motion on a Riemannian manifold N
these will be the geodesics connecting two generic points.
Now dene the submanifold
L
()
t
:= (P, Q, p, q) T

N T

N[P = d
Q
/

, p = d
q
/

I other words, (P, Q, p, q) L

t
means that if one starts at (p, q) at
time 0 and ow for time t along the Hamiltonian vector eld v
H
,
one will ends up at the point (P, Q).
Proposition 1.5.2. L
t
is a Lagrangian submanifold.
Proof. This follows from the fact that owing for time t,
F
t
: T

N T

N is a symplectomorphism, and L
t
is the graph
of this symplectomorphism. In general, the graph of a symplec-
tomorphism M M is a Lagrangian in M M
op
where M
op
is
the symplectic manifold M with the reversed sign of the symplectic
form.
If there are more then one trajectories one can dene
L
t
=

t
as the set of solutions of Euler-Lagrange equations. In general it is
not a submanifold. But every subset of it which is a submanifold
is a Lagrangian submanifold.
Now it is clear that solutions to the E-L equations with boundary
conditions B can be identied with points in L
t
B

.
In these lectures we will be interested mostly in the case when
B

= B
1
B
2
, where B
i
T

N are Lagrangian submanifolds.

1.5.2 Hamilton-Jacobi equation
Now we return to the situation where our symplectic manifold is
T

N for some smooth manifold N, Hamiltonian H C

(T

N)
with corresponding Lagrangian / (assume strongest possible non-
degeneracy). Suppose (q, Q) is a solution to the Euler-Lagrange
equations with xed end points so that
0
= q and
t
= Q. Lets
assume that this solution is unique for every pair (q, Q). Then we
have
/(q, Q)
t1,t2
=
_
t2
t1
/
_
(), ()
_
d =: /
t1,t2
(q, Q)
So we have a family of functions /
t1,t2
on NN. These functions
are quite remarkable. For each interval [t
1
, t
2
] such function denes
a Lagrangian submanifold
L
[t1,t2]
= (P, Q), (p, q)[P = dQ/
t1,t2
, p = dq/
t1,t2

in T

N T

N. and this Lagrangian submanifold is The second

wonderful property is that if we want to compute it we dont have
to use the Euler-Lagrange equations.
Theorem 1.5.3 (Hamilton-Jacobi). The function /(q, Q)
t1,t2
sat-
ises the following system of partial dierential equations
/
t1,t2
t
2
+H(d
Q
/
t1,t2
, Q) = 0
/
t1,t2
t
1
H(d
q
(/
t1,t2
), q) = 0
Proof. For the derivative of the action function we have
A
t1,t2
t
2
= /
_
(t
2
), (
2
)
_
+
_
t2
t1
/
_
(), ()
_

()
t
2
d
+
_
t2
t1
/
_
(), ()
_
q
()
t
2
d = /
_
(t
2
), (
2
)
_
+
_
t2
t1
_

d
d
/
_
(), ()
_

+
/
_
(), ()
_
q
)
()
t
2
d
+
d
d
/
_
(t
2
), (t
2
)
_

()
t
2
[
=t2

d
d
/
_
(t
1
), (t
1
)
_

()
t
2
[
=t1
(1.5.4)
The second term is zero because of the Euler-Lagrange equations.
In order deal with the third and forth terms notice (t
2
) = Q and
(t
1
) = q. Dierentiating these conditions with respect to t
2
we
have:
()

[
=t2
+
()
t
2
[
=t2
= 0,
()
t
2
[
=t21
= 0
From here we have:
A
t1,t2
t
2
= /
_
(t
2
), (
2
)
_
p(t
2
) (t
2
) = H(p(t
2
), q(t
2
))
This proves the rst equation. The second equation can be proven
similarly.
1.6 First order Lagrangian systems
The non-degeneracy condition of

2
L

j
(, q) is violated in an im-
portant class of rst order Lagrangians.
Let be a 1-form on N and b is a function on N. Dene the
action
S
[t2,t1]
[] =
_
t2
t1
(< ((t)), (t) > +b((t)))dt
where is a parameterized path.
Euler-Lagrange equations for this action are:
( (t)) +db((t)) = 0
where = d.
Naturally, the rst order Lagrangian system is called non-
degenerate if the form is non-degenerate.
It is clear that non-degenerate rst order Lagrangian system de-
nes a symplectic structure on the manifold N. The Euler-Lagrange
equations for such system are Hamiltonian on the symplectic man-
ifold (N, ) with the Hamiltonian H = b. It is also clear that the
action of the non-degenerate rst order system is the Hamilton-
Jacobi action for this Hamiltonian system.
Assuming that satises the Euler-Lagrange equations the varia-
tion of the action is not yet vanishes but it is given by the boundary
terms (??). For rst order Lagrangians this becomes:
S
[t2,t1]
[] = ((t))(t)[
t2
t1
If (t
1
) and (t
2
) are constrained to Lagrangian submanifolds in N
these terms vanish.
Thus, constraining boundary points of to Lagrangian submani-
folds is a natural boundary condition for non-degenerate rst order
Lagrangian systems. As we will see, it is a part of the more gen-
eral concept where Lagrangian submanifolds are natural boundary
conditions for Hamiltian systems.
1.7 Degenerate Lagrangian systems
Assume a Lie group G acts on N and the Lagrangian function of
TN is invariant with respect to the induced action of G. This is
an example of a degenerate Lagrangian system: in the case matrix

2
L

j
is degenerate.
1.8 States in classical mechanics
Denition 1.8.1. A state on manifold M (doesnt have to be sym-
plectic) is a probability measure on M (i.e. a measure so that
(M) = 1).
Such probability measure can be interpreted as follow: (U M)
is the probability that in the state , during an observation, the
system can be observed in a subset U of the whole phase space M.
There is a natural way to assign a measure
f
on R to a function
f C

(M). If E R, then
f
(E) is

f
(E) =
_
f
1
(E)
_
or

f
(E) =
_
Mf (E)

where M
f
(E) = f
1
(E) = x M[f(x) E.
The probabilistic interpretation of the measure
f
is: the value

f
(E) is the probability that if we will be measuring the function
f in the state , it will have a value in the subset E R.
Remark 1.8.2. If
1
and
2
are such measures, then any convex
combination
1
+ (1 )
2
is also a probability measure.
Denition 1.8.3. Pure states are states that are supported at
points. That is, they are measures
x
(U) =
_
1 x U
0 x , U
.
The Poisson algebra C

(M) is called the algebra of classical

observables. The phase space M can be identied with the set of
pure states.
An important characteristic of an observable is its expectation
value
E

(f) =
_
M
f =: f

in a state .
Another important characteristic is the dispersion around E

(f)

(f)
2
=

(f f

)
_

= f
2

f
2

0.
The Hamiltonian dynamics, i.e. the local action of R on M gen-
erated by the function H induces the dual dynamics on states is

t
, where E
(t)
(f) := E

(f
t
).
The evolution of states is called Liouville evolution. The phase
space M is symplectic a symplectic manifold with the volume form

n
. If (x) is the density of a state relative the symplectic volume
measure, its evolutions is given by the dierential equation
d
t
dt
= H,
t