Summary of the Above

d2 ~
r ~
• Newton’s second law: dt2
= −∇Φ(~
r)

− Complicated vector arithmetic & coordinate system dependence

 
∂L d ∂L
• Lagrangian Formalism: ∂q i
− dt ∂ q̇i
=0

− n second-order differential equations

∂H ∂H
• Hamiltonian Formalism: ∂pi
= q̇i ∂qi
= −ṗi

− 2n first-order differential equations

 
∂S
• Hamilton-Jacobi equation: H ∂q i
, qi = E

S(~
q, p
~) is generator of canonical transformation (~q, p~) → (Q,~ P
~ ) for
which H(~ ~) → H0 (P
q, p ~ ). If S(~
q, p
~) is separable then the Hamilton-Jacobi
equation breaks up in n ordinary differential equations which can be solved
by simple quadrature. The resulting equations of motion are:
 0

∂H
Pi (t) = Pi (0) Qi (t) = ∂Pi
t + ki
 Constants of Motion
Constants of Motion: any function C(~ q, p
~, t) of the generalized coordinates,
conjugate momenta and time that is constant along every orbit, i.e., if q
~(t)
and p
~(t) are a solution to the equations of motion, then
C[~
q (t1 ), p
~(t1 ), t1 ] = C[~
q (t2 ), p
~(t2 ), t2 ]
for any t1 and t2 . The value of the constant of motion depends on the orbit,
but different orbits may have the same numerical value of C
A dynamical system with n degrees of freedom always has 2n independent
constants of motion. Let qi = qi [~ q0 , p
~0 , t] and pi = pi [~
q0 , p
~0 , t] describe
the solutions to the equations of motion. In principle, these can be inverted
to 2n relations qi,0 = qi,0 [~
q (t), p
~(t), t] and pi,0 = pi,0 [~ q (t), p~(t), t].
By their very construction, these are 2n constants of motion.
If Φ(~
x, t) = Φ(~x), one of these 2n relations can be used to eliminate t.
This leaves 2n − 1 non-trivial constants of motion, which restricts the
system to a 2n − (2n − 1) = 1-dimensional surface in phase-space,
namely the phase-space trajectory Γ(t)
Note that the elimination of time reflects the fact that the physics are invariant
to time translations t → t + t0 , i.e., the time at which we pick our initial
conditions can not hold any information regarding our dynamical system.
 Integrals of Motion I
Integrals of Motion: any function I(~x, ~
v ) of the phase-space coordinates
(~
x, ~
v ) alone that is constant along every orbit, i.e.
I[~
x(t1 ), ~
v (t1 )] = I[~
x(t2 ), ~
v (t2 )]
for any t1 and t2 . The value of the integral of motion can be the same for
different orbits. Note that an integral of motion can not depend on time.
Thus, all integrals are constants, but not all constants are integrals.

Integrals of motion come in two kinds:

Isolating Integrals of Motion: these reduce the dimensionality of the

trajectory Γ(t) by one. Therefore, a trajectory in a dynamical system with n
degrees of freedom and with i isolating integrals of motion is restricted to a
2n − i dimensional manifold in the 2n-dimensional phase-space. Isolating
integrals of motion are of great practical and theoretical importance.

Non-Isolating Integrals of Motion: these are integrals of motion that do not

reduce the dimensionality of Γ(t). They are of essentially no practical value
for the dynamics of the system.
 Integrals of Motion II
q, p
A stationary, Hamiltonian system (i.e., H(~ ~, t) = H(~ q, p
~)) with n
degrees of freedom always has 2n − 1 independent integrals of motion,
which restrict the motion to the one-dimensional phase-space trajectory
Γ(t). The number of isolating integrals of motion can, depending on the
Hamiltonian, vary between 1 and 2n − 1.

DEFINITION: Two functions I1 and I2 of the canonical phase-space

coordinates (~ q, p~) are said to be in involution if their Poisson bracket
vanishes, i.e., if
∂I1 ∂I2 ∂I1 ∂I2
[I1 , I2 ] = ∂qi ∂pi
− ∂pi ∂qi
=0

A set of k integrals of motion that are in involution forms a set of k isolating

integrals of motion.

Liouville’s Theorem for Integrable Hamiltonians

A Hamiltonian system with n degrees of freedom which posseses n
integrals of motion in involution, (and thus n isolating integrals of motion) is
integrable by quadrature.
 Integrable Hamiltonians I
LEMMA: If a system with n degrees of freedom has n constants of motion
Pi (~q, p
~, t) [or integrals of motion Pi (~
q, p
~)] that are in involution, then there
will also be a set of n functions Qi (~q, p
~, t) [or Qi (~
q, p~)] which together
with the Pi constitute a set of canonical variables.

Thus, given n isolating integrals of motion Ii (~

q, p
~) we can make a
canonical transformation (~
q, p ~ P
~) → (Q, ~ ) with Pi = Ii (~
q, p
~) =constant
and with Qi (t) = Ωi t + ki

An integrable, Hamiltonian system with n degrees of freedom always has a

set of n isolating integrals of motion in involution. Consequently, the
trajectory Γ(t) is confined to a 2n − n = n-dimensional manifold
phase-space.
The surfaces specified by (I1 , I2 , .., In ) =constant are topologically
equivalent to n-dimensional tori. These are called invariant tori, because any
orbit originating on one of them remains there indefinitely.
In an integrable, Hamiltonian system phase-space is completely filled (one
says ‘foliated’) with invariant tori.
 Integrable Hamiltonians II
To summarize: if, for a system with n degrees of freedom, the
Hamilton-Jacobi equation is separable, the Hamiltonian is integrable and
there exist n isolating integrals of motion Ii in involution. In this case there
exist canonical transformations (~
q, p ~ P
~) → (Q, ~ ) such that equations of
motion reduce to:

Pi (t) = P
 i (0) 
0
∂H
Qi (t) = ∂Pi
t + ki

One might think at this point, that one has to take Pi = Ii . However, this
choice is not unique. Consider an integrable Hamiltonian with n = 2
degrees of freedom and let I1 and I2 be two isolating integrals of motion in
involution. Now define Ia = 1 2
(I 1 + I 2 ) and I b = 1
2
(I1 − I2 ), then it is
straightforward to proof that [Ia , Ib ] = 0, and thus that (Ia , Ib ) is also a
set of isolating integrals of motion in involution. In fact, one can construct
infinitelly many sets of isolating integral of motion in involution. Which one
should we choose, and in particular, which one yields the most meaningful
description of the invariant tori?
Answer: The Action-Angle variables
            

                       
                       

           
           

γ1
                       
                       

           

           

                       
                       

           

           

                       

                       

A1
           
           

                       

                       
Γ(t)

           
           

                       
                       

           
           

                       
                       

                       
                       

                       
                       
.

                       

                       

                       
                       

                       

                       

                       
                       

                       
                       

                       
                       

           

                       
                       

           

                       

                       
ω2

           

                       

                       

           

                       
                       



           

Ji =
                       
                       

           

                       
                       

The action variables are defined by:

           
ω1

                       

                       

1
2π
                       

                       

H
γi
p

q
~ · d~

.

A1 and A2 , which clearly are two integrals of motion.

with γi the closed loop that bounds cross section Ai .

A2

Action-Angle Variables I

γ2
ω2

Let’s be guided by the idea of our invariant tori. The figure illustrates a

specify a location on this torus by the two position angles ω1 and ω2 . The
torus itself is characterized by the areas of the two (hatched) cross sections
2D-torus (in 4D-phase space), with a trajectory Γ(t) on its surface. One can

labelled A1 and A2 . The action variables J1 and J2 are intimitally related to

 Action-Angle Variables II
The angle-variables ωi follow from the canonical transformation rule
ωi = ∂S
with S = S(~ ~ the generator of the canonical transformation
q , J)
∂Ji
(~
q, p
~) → (~ ~ . Since the actions Ji are isolating integrals of motion we
ω , J)
have that the corresponding conjugate angle coordinates wi obey
 0
ωi (t) = ∂H∂Ji
t + ω0

~ the Hamiltonian in action-angle variables (~

with H0 = H0 (J) ~ .
ω , J)
We now give a detailed description of motion on invariant tori:
Orbits in integrable, Hamiltonian systems with n degrees of freedom are
characterized by n constant frequencies

∂H0
Ωi ≡ ∂Ji

This implies that the motion along each of the n degrees of freedom, qi , is
periodic in time, and this can occur in two ways:
• Libration: motion between two states of vanishing kinetic energy
• Rotation: motion for which the kinetic energy never vanishes
 The Pendulum
To get insight into libration and rotation consider a pendulum, which is a
integrable Hamiltonian system with one degree of freedom, the angle q ..

The figures below shows the

corresponding phase-diagram.
l
q




p
F

r
s l=libration
l
r=rotation
−π +π q s=separatrix

• Libration: q(ω + 2π) = q(ω).

• Rotation: q(ω + 2π) = q(ω) + 2π
To go from libration to rotation, one needs to cross the separatrix
 Action-Angle Variables III
Why are action-angle variables the ideal set of isolating integrals of motion to
use?

• They are are the only conjugate momenta that enjoy the property of
adiabatic invariance (to be discussed later)
• The angle-variables are the natural coordinates to label points on
invariant tori.
• They are ideally suited for perturbation analysis, which is used to
investigate near-integrable systems (see below)
• They are ideally suited to study the (in)-stability of a Hamiltonian system
 Example: Central Force Field
As an example, to get familiar with action-angle variables, let’s consider once
again motion in a central force field.
As we have seen before, the Hamiltonian is
2
1 2 1 pθ
H= p
2 r
+ 2 r2
+ Φ(r)

where pr = ṙ and pθ = r 2 θ̇ = L.
In our planar description, we have two integrals of motion, namely energy
I1 = E = H and angular momentum I2 = L = pθ .
These are classical integrals of motion, as they are associated with
symmetries. Consequently, they are also isolating.
Let’s start by checking whether they are in involution
h i h i
∂I1 ∂I2 ∂I1 ∂I2 ∂I1 ∂I2 ∂I1 ∂I2
[I1 , I2 ] = ∂r ∂pr
− ∂pr ∂r
+ ∂θ ∂pθ
− ∂pθ ∂θ

∂I2 ∂I2 ∂I1 ∂I2

Since ∂p
r
= ∂r
= ∂θ
= ∂θ
= 0 one indeed finds that the two
integrals of motion are in involution.
 Example: Central Force Field
The actions are defined by
1 1
H H
Jr = 2π γr
pr dr Jθ = 2π γθ
pθ dθ

In the case of Jθ the θ -motion is one of rotation. Therefore the

closed-line-integral is over an angular interval [0, 2π].

2π
1
R
Jθ = 2π
I2 dθ = I2
0

In the case of Jr , we need to realize that the r -motion is a libration between

apocenter r+ and pericenter r− . Using that I1 = E = H we can write
p
pr = 2[I1 − Φ(r)] − I22 /r 2
The radial action then becomes
rR+ p
1
Jr = π
2[I1 − Φ(r)] − I22 /r 2 dr
r−

Once we make a choise for the potential Φ(r) then Jr can be solved as
function of I1 and Jθ . Since I1 = H, this in turn allows us to write the
Hamiltonian as function of the actions: H(Jr , Jθ ).
 Example: Central Force Field
As an example, let’s consider a potential of the form
β
Φ(r) = − α
r
− r2
with α and β two constants. Substituting this in the above, one finds:
 1/2
1
p
Jr = α 2|I1 |
− Jθ2 − 2β
Inverting this for I1 = H yields
 −2
α2
p
H(Jr , Jθ ) = 2
Jr + Jθ2 − 2β

Since the actions are isolating integrals of motion, and we have an

expression for the Hamiltonian in terms of these actions, the generalized
coordinates that correspond to these actions (the angles wr and wθ ) evolve
as wi (t) = Ωi t + wi,0
The radial and angular frequencies are
 −3
∂H 2
p
2
Ωr = ∂J r
= −α J r + J θ − 2β
 −3
∂H Jθ
2
p
2 √
Ωθ = ∂J = −α J r + J θ − 2β 2
θ Jθ −2β
 Example: Central Force Field
The ratio of these frequencies is
 1/2
Ωr 2β
Ωθ
= 1− Jθ2

Note that for β = 0, for which Φ(r) = − α

r
, and thus the potential is of the
Kepler form, we have that Ωr = Ωθ independent of the actions (i.e., for each
individual orbit).
In this case the orbit is closed, and there is an additional isolating integral of
motion (in addition to E and L). We may write this ‘third’ integral as
I3 = wr − wθ = Ωr t + wr,0 − Ωθ t − wθ,0 = wr,0 − wθ,0
Without loosing generality, we can pick the zero-point of time, such that
wr,0 = 0. This shows that we can think of the third integral in a Kepler
potential as the angular phase of the line connecting apo- and peri-center.
 Quasi-Periodic Motion
In general, in an integrable Hamiltonian system with the canonical
transformation (~q, p
~) → (Q,~ P ~ ) one has that qk = qk (ω1 , .., ωn ) with
k = (1, .., n). If one changes ωi by 2π , while keeping the other ωj (j 6= i)
fixed, then qi then performs a complete libration or rotation.
The Cartesian phase-space coordinates (~ x, ~
v ) must be periodic functions of
the angle variables ωi with period 2π . Any such function can be expressed
as a Fourier series
∞
~ = ~ exp [i(lω1 + mω2 + nω3 )]
P
~
x(~
ω , J) Xlmn (J)
l,m,n=−∞

Using that ωi (t) = Ωi t + ki we thus obtain that

∞
P
~
x(t) = X̃lmn exp [i(lΩ1 + mΩ2 + nΩ3 )t]
l,m,n=−∞

with X̃lmn = Xlmn exp [i(lk1 + mk2 + nk3 )]

Functions of the form of ~
x(t) are said to be quasi-periodic functions of time.
Hence, in an integrable systems, all orbits are quasi-periodic, and confined
to an invariant torus.
 Integrable Hamiltonians III
When one integrates a trajectory Γ(t) in an integrable system for sufficiently
long, it will come infinitessimally close to any point ω
~ on the surface of its
torus. In other words, the trajectory densely fills the entire torus.
Since no two trajectories Γ1 (t) and Γ2 (t) can intersect the same point in
phase-space, we thus immediately infer that two tori are not allowed to
intersect.

In an integrable, Hamiltonian system phase-space is

completely foliated with non-intersecting, invariant tori

 Integrable Hamiltonians IV
In an integrable, Hamiltonian system with n degrees of freedom, all orbits
are confined to, and densely fill the surface of n-dimensional invariant tori.
These orbits, which have (at least) as many isolating integrals as spatial
dimensions are called regular
Regular orbits have n frequencies Ωi which are functions of the
corresponding actions Ji . This means that one can always find suitable
values for Ji such that two of the n frequencies Ωi are commensurable, i.e.
for which
l Ωi = m Ω j
with i 6= j and l, m both integers.
A regular orbit with commensurable frequencies is called a resonant orbit
(also called closed or periodic orbit), and has a dimensionality that is one
lower than that of the non-resonant, regular orbits. This implies that there is
an extra isolating integral of motion, namely
In+1 = lωi − mωj
Note: since ωi (t) = Ωi t + ki , one can obtains that In+1 = lki − mkj ,
Note: and thus is constant along the orbit.
 Near-Integrable Systems I
Thus far we have focussed our attention on integrable, Hamiltonian systems.
Given a Hamiltonian H(~ q, p
~), how can one determine whether the system is
integrable, or whether the Hamilton-Jacobi equation is separable?
Unfortunately, there is no real answer to this question: In particular, there is
no systematic method for determining if a Hamiltonian is integrable or not!!!
However, if you can show that a system with n degrees of freedom has n
independent integrals of motion in involution then the system is integrable.
Unfortunately, the explicit expression of the integrals of motion in terms of
the phase-space coordinates is only possible in a very so called classical
integrals of motion, those associated with a symmetry of the potential and/or
with an invariance of the coordinate system.
In what follows, we only consider the case of orbits in ‘external’ potentials for
which n = 3. In addition, we only consider stationary potentials Φ(~ x), so
that the Hamiltonian does not explicitely on time and
H(~ q, p
~) = E =constant. Therefore

Energy is always an isolating integral of motion.

Note: this integral is related to the invariance of the Lagrangian L under time
translation, i.e., to the homogeneity of time.
 Near-Integrable Systems II
Integrable Hamiltonians are extremely rare. As a consequence, it is
extremely unlikely that the Hamiltonian associated with a typical galaxy
potential is integrable.
One can prove that even a slight perturbation away from an integrable
potential will almost always destroy any integral of motion other than E .
So why have we spent so must time discussing integrable Hamiltonians?
. Because most galaxy-like potentials turn out to be near-integrable.
Definition: A Hamiltonian system is near-integrable if a large fraction of
Definition: phase-space is still occupied by regular orbits
Definition: (i.e., by orbits on invariant tori).

The dynamics of near-integrable Hamiltonians is the subject of the

Kolmogorov-Arnold-Moser (KAM) Theorem which states:

If H0 is an integrable Hamiltonian whose phase-space is completely foliated

with regular orbits on invariant tori, then in a perturbed Hamiltonian
H = H0 + εH1 most orbits will still lie on such tori for sufficiently small ε.
The fraction of phase-space covered by these tori → 1 for ε → 0 and the
perturbed tori are deformed versions of the unperturbed ones.
 Near-Integrable Systems III
The stability of the original tori to a perturbation can be proven everywhere
except in small regions around the resonant tori of H0 . The width of these
regions depends on ε and on the order of the resonance.
According to the Poincaré-Birkhoff Theorem the tori around unstable
resonant tori break up and the corresponding regular orbits become irregular
and stochastic.
Definition: A resonant orbit is stable if an orbit starting close to it remains
Definition: close to it. They parent orbit families (see below)
Definition: An irregular orbit is an orbit that is not confined to a
Definition: n-dimensional torus. In general it can wander through
Definition: the entire phase-space permitted by conservation of energy.
Consequently, an irregular orbit is restricted to a higher-dimensional
manifold than a regular orbit. Irregular orbits are stochastic in that they are
extremely sensitive to initial conditions: two stochastic trajectories Γ1 (t)
and Γ2 (t) which at t = t0 are infinitesimally close together will diverge with
time.
Increasing ε, increases the widths of the stochastic zones, which may
eventually ‘eat up’ a large fraction of phase-space.
 Near-Integrable Systems IV
Note that unperturbed resonant tori form a dense set in phase-space, just
like the rational numbers are dense on the real axis.
Just like you can always find a rational number in between two real numbers,
in a near-integrable system there will always be a resonant orbit in between
any two tori. Since many of these will be unstable, they create many
stochastic regions
As long as the resonance is of higher order (i.e, 16 : 23 rather than 1 : 2)
the corresponding chaotic regions are very small, and tightly bound by their
surrounding tori.
Since two trajectories can not cross, an irregular orbit is bounded by its
neighbouring regular orbits. The irregular orbit is therefore still (almost)
confined to a n-dimensional manifold, and it behaves as if it has n isolating
integrals of motion.
. It may be very difficult to tell whether an orbit is regular or irregular.
However, iff n > 2 an irregular orbit may slip through a ‘crack’ between two
confining tori, a process know as Arnold diffusion.
Because of Arnold diffusion the stochasticity will be larger than ‘expected’.
However, the time-scale for Arnold diffusion to occus is long, and it is
unclear how important it is for Galactic Dynamics.
 Near-Integrable Systems V
A few words on nomenclature:
Recall that an (isolating) integral of motion is defined as a function of
phase-space coordinates that is constant along every orbit.
A near-integrable systems in principle has only one isolating integral of
motion, namely energy E .
Nevertheless, according to the KAM Theorem, many orbits in a
near-integrable system are confined to invariant tori.
Although in conflict with the definition, astronomers often say that the
regular orbits in near-integrable systems admit n isolating integrals of
motion.
Astronomers also often use KAM Theorem to the extreme, by assuming that
they can ignore the irregular orbits, and that the Hamiltonians that
correspond to their ‘galaxy-like’ potentials are integrable. Clearly the validity
of this approximation depends on the fraction of phase-space that admits
three isolating integrals of motion. For most potentials used in Galactic
Dynamics, it is still unclear how large this fraction really is, and thus, how
reliable the assumption of integrability is.