You are on page 1of 13

Chapter 3

Hamiltonian dynamics
The main idea in hamiltonian dynamics is that instead of using only the coordinates
q
i
(t) and their derivatives to describe the system, we think of the coordinates q
i
and the
momenta p
i
as independent variables.
This may seem like an odd idea, since once we know the coordinates q
i
(t) as a function
of time, we also know their time derivatives q
i
(t), and through that the momenta p
i
(t).
So how can q and p be considered independent?
One way of making sense of this is to note that knowing the position of a body at a
particular time does not in itself tell us anything about its velocity (or momentum), or
vice-versa. It is only if we know the position at several dierent times that we will be
able to work out its velocity. And the full relation between q(t) and q(t), or between
q(t) and p(t), can only be known if we know the coordinate q(t) at all times t but
this amounts to having solved the problem of the motion of the body! So in this sense,
q and p (or, indeed, q and q) can be considered independent variables.
Secondly, the relation between coordinates and canonical momenta is not a simple one
like the relation between a coordinate and its derivative, but is related to the dynamics of
the system as encoded in the lagrangian (or, as we shall see, the hamiltonian). Therefore,
it makes sense to consider p as independent of q in a way that q cannot be.
Thirdly, as you will see in quantum mechanics, the two must be treated as independent
quantities there. From the quantum mechanical commutation relations between coordi-
nates and canonical momenta, [q
i
, p
k
] = i h
jk
, one can derive Heisenbergs uncertainty
relation, q
i
p
i
h, which holds for all (q, p) pairs. Therefore, in quantum mechanics,
it is impossible to know the coordinate and momentum of a particle at the same time.
Also, statistical mechanics, which forms the basis of the modern treatment of thermal
physics, is formulated in the phase space where coordinates and momenta are considered
as independent variables.
So our rst step in arriving at the hamiltonian formulation of mechanics will be to
use the relation between momenta p
i
and velocities q
i
to eliminate q, and write the
hamiltonian as a function of coordinates and momenta,
H = H(q
i
, p
i
, t) as opposed to L = L(q
i
, q
i
, t) . (3.1)
30
This will be our starting point in this chapter.
3.1 Hamiltons equations of motion
The EulerLagrange equations are equations of motion written in terms of the la-
grangian. We now want to nd similar equations in terms of the hamiltonian. To
achieve this, we rst note that the equations of motion are dierential equations, so we
can try to dierentiate the hamiltonian H. Starting from the denition of H in terms
of the lagrangian,
H =
N

i=1
p
i
q
i
L, (3.2)
we note that the left hand side of this equation is now a function of q, p and t, while the
right hand side is a function of q, q, t and p, since p enters into the rst term.
We can now take the derivative of both sides with respect to some parameter s, which
all quantities involved (including time) can depend on:
dH
ds
=
N

i=1
_
dp
i
ds
q
i
+ p
i
d q
i
ds
_

i=1
_
L
q
i
dq
i
ds
+
L
q
i
d q
i
ds
_

L
t
dt
ds
. (3.3)
Using the EulerLagrange equation and the denition of the canonical momentum,
L
q
i
= p
i
and
L
q
i
=
d
dt
L
q
i
=
dp
i
dt
= p
i
, (3.4)
we nd
dH
ds
=
N

i=1
_
q
i
dp
i
ds
+ p
i
d q
i
ds
p
i
dq
i
ds
p
i
d q
i
ds
_

L
t
dt
ds
. (3.5)
But H is a function of q
i
, p
i
, t, so using the chain rule we also have
dH
ds
=
N

i=1
_
H
q
i
dq
i
ds
+
H
p
i
dp
i
ds
_
+
H
t
dt
ds
. (3.6)
The two expressions (3.5) and (3.6) must be the same whatever the parameter s is, so
comparing the two expressions we see
q
i
=
H
p
i
; p
i
=
H
q
i
;
H
t
=
L
t
. (3.7)
The boxed equations are called Hamiltons equations of motion or the canonical
equations.
31
Using Hamiltons equations of motion, it is very straightforward to show that the hamil-
tonian is conserved if it does not depend explicitly on time, by taking the parameter s
in (3.6) to be the time t:
dH
dt
=
N

i=1
_
H
q
i
q
i
+
H
p
i
p
i
_
+
H
t
=
N

i=1
_
p
i
q
i
+ q
i
p
i
_
+
H
t
=
H
t
. (3.8)
Example 3.1
Consider a particle constraint to move on the cylindrical surface x
2
+ y
2
= R
2
,
subject to a central force

F = k

r .
Using cylinder coordinates (, z) with x = Rcos , y = Rsin we nd
V =
1
2
kr
2
=
1
2
k(R
2
+ z
2
) , (3.9)
T =
1
2
m( x
2
+ y
2
+ z
2
) =
1
2
m(R
2

2
+ z
2
) (3.10)
= L =
1
2
m(R
2

2
+ z
2
)
1
2
k(R
2
+ z
2
) . (3.11)
The canonical momenta are
p

=
L

= mR
2

, p
z
=
L
z
= m z . (3.12)
We can use this to nd

, z in terms of p

, p
z
:

=
p

mR
2
, z =
p
z
m
. (3.13)
The hamiltonian is
H = p
z
z p


L = p
z
p
z
m
+ p

mR
2

1
2
m
_
R
2
_
p

mR
2
_
2
+
_
p
z
m
_
2
_
+
1
2
k(R
2
+ z
2
)
=
p
2
z
2m
+
p
2

2mR
2
+
1
2
kz
2
+
1
2
kR
2
= H(z, p
z
, p

) . (3.14)
It is equal to the total energy since the potential energy does not depend on velocities
and the kinetic energy has the usual form. It is conserved since there is no explicit
time-dependence in L (or H).
Hamiltons equations of motion for this system are
p

=
H

= 0 ,

=
H
p

=
p

mR
2
(3.15)
p
z
=
H
z
= kz , z =
H
p
z
=
p
z
m
(3.16)
32
We can use (3.15), (3.16) to arrive at
p

= mR
2

= constant, z =
p
z
m
=
k
m
z . (3.17)
These are the EulerLagrange equations for the system, so we have shown that
Hamiltons equations are exactly equivalent to the EulerLagrange equations, as
they should be.
In the example above we see that does not appear in the expression for H: it is a
cyclic coordinate. This implies that the canonical momentum p

is conserved, but in the


hamiltonian framework it actually simplies the system even further: in the remaining
equations we can simply treat p

as any other constant, so the whole motion in the


-coordinate decouples from the remaining equations: instead of 3 variables z, z,

we
now just have 2: z, z.
This decoupling is a generic feature which can simplify the analysis of the system con-
siderably, as we shall see below.
3.2 Cyclic coordinates and eective potential
To see how this works out, let us look at a slightly more complex system, that of the
spherical pendulum. This is a pendulum that can swing freely in all directions, not just
in a plane. Using spherical coordinates (, ), where is the angle with the vertical axis,
we nd that the lagrangian of this system is
L =
1
2
m
2
(

2
+

2
sin
2
) + mg cos . (3.18)
The canonical momenta are
p

=
L

= m
2

(3.19)
p

=
L

= m
2
sin
2


(3.20)
From (3.19), (3.20) we nd

=
p

m
2
,

=
p

m
2
sin
2

. (3.21)
Since there is nothing funny going on in this system, the hamiltonian is equal to the
total energy,
H = T + V =
p
2

2m
2
+
p
2

2m
2
sin
2

mg cos . (3.22)
33
Hamiltons equations of motion are then

=
H
p

=
p

m
2
, p

=
H

=
p
2

cos
m
2
sin
3

mg sin , (3.23)

=
H
p

=
p

m
2
sin
2

, p

=
H

= 0 . (3.24)
Since p

is constant, the two last terms in (3.22) depend only on . They can be taken
to dene an eective potential,
V
e
() =
p
2

2m
2
sin
2

mg cos . (3.25)
Let us now look at a system with a single degree of freedom , with kinetic energy
p
2

/(2m
2
) and potential energy V
e
(). The hamiltonian of this system is exactly the
same as (3.22), and therefore Hamiltons equations of motion for , p

are exactly the


same as (3.23).
0 0.5 1 1.5 2 2.5 3

-1
0
1
2
3
V
e
f
f

/
m
g
l
k = 0
k = 0.02
k = 0.1
k = 0.5
k = 1.0
k = 1.5
Figure 3.1: The eective potential for the spherical pendulum, for dierent values of
k = p
2

/2m
2
g
3
.
We can use the eective potential to nd out what types of motion are possible in the
-direction. Figure 3.1 shows V
e
as a function of for several values of p

. We can see
that for all p

= 0 the eective potential goes to innity at = 0 and = . This means


that only bounded motion exists for p

= 0. The minimum of the potential corresponds


34
to circular motion at a xed angle , given by
mg cos (1 cos
2
) =
p
2

2m
2
. (3.26)
In contrast, for p

= 0 (the solid line), both bounded motion (through = 0) and


unbounded motion are possible, for mg < E < mg and E mg respectively. In
this case, the spherical pendulum reduces to the plane (simple) pendulum.
What we have seen in this example is quite typical of what happens if one or more of
the coordinates are cyclic. In general, if the hamiltonian can be written as
H(q
1
, q
2
, p
1
, p
2
) = f(q
1
)p
2
1
+ g(q
1
)p
2
2
+ V (q
1
) , (3.27)
where f and g can be any function of the coordinate q
1
, then we immediately see that
the second coordinate q
2
is cyclic, and the momentum p
2
is therefore conserved, ie it is
a constant. We can then dene an eective potential
V
e
(q
1
) = g(q
1
)p
2
2
+ V (q
1
) , (3.28)
and the hamiltonian will be equivalent to that of a 1-dimensional system with kinetic
energy T
1
given by
T
1
= f(q
1
)p
2
1
= H(q
1
, p
1
) = T
1
+ V
e
(q
1
) . (3.29)
It is clear that the equations of motion of q
1
and p
1
we obtain from (3.29) are the same
as what we obtain from (3.27) since the two hamiltonians are exactly the same: all we
have done is group the terms in a dierent way,
If are only interested in the motion in the q
1
coordinate we can stop as soon as we have
solved the equations of motion resulting from (3.29). If we also want to determine q
2
(t),
we use its equation of motion,
q
2
=
H
p
2
= 2p
2
g(q
1
) = q
2
(t) = 2p
2
_
t
q
2
(t
0
)
g(q
1
(t

))dt

. (3.30)
Once we have found q
1
(t) it is in principle straightforward to perform this integral to
obtain q
2
(t).
3.3 Hamiltons equations from a variational princi-
ple
We arrived at Hamiltons equations from the lagrangian and using the EulerLagrange
equations, but it is also possible to derive them directly from a variational principle, as
we shall now see. To this end, we can rewrite the action S =
_
Ldt as
S =
_
_
p q H(q, p, t)
_
dt . (3.31)
35
The equations of motion will now be derived by requiring S = 0, where now p and q can
be varied independently. We proceed analogously to how we derived the EulerLagrange
equations in Section 2.2, using the shorthand S instead of dS/d:
S =
_
_
p q H(q, p, t)
_
dt
=
_
t
2
t
1
_

i
_
p
i
q
i
+ p
i
q
i
_

i
_
H
p
i
p
i
+
H
q
i
q
i
__
dt
=

i
_
t
2
t
1
_
q
i
p
i
+
d
dt
(p
i
q
i
) p
i
q
i

H
p
i
p
i

H
q
i
q
i
_
dt
=

i
_
p
i
q
i
_

t
2
t
1
+

i
_
t
2
t
1
__
q
i

H
p
i
_
p
i

_
p
i
+
H
q
i
_
q
i
_
dt = 0 .
(3.32)
The rst term on the last line is zero because the endpoints q
i
(t
1
), q
i
(t
1
) are xed, so
q
i
(t
1
) = q
i
(t
2
) = 0. In the second term, the integral must be zero for any arbitrary
variations p
i
, q
i
, which can both be chosen indenpendently for all degrees of freedom
i. The only way this can be the case is if both terms inside the ordinary brackets are
zero at all times, and for all i. This gives us
q
i
=
H
p
i
, p
i
=
H
q
i
, (3.33)
which is Hamiltons equations of motion.
3.4 Phase space [Optional]
We can consider the values of {p, q} = (p
1
, . . . , p
N
, q
1
, . . . , q
N
) as coordinates of a point
in a 2N-dimensional space. This space is called phase space. The evolution of the system
in time can then be thought of as a trajectory in this space. The state of a system is
given by its location in phase space, ie by the values of all the generalised coordinates
and momenta in other words, every point in phase space describes a unique state.
There are two important aspects of this:
1. The trajectory of a system in phase space provides a complete description of the
motion of the system.
2. The complete trajectory is uniquely determined by the initial coordinates
{p
i
(t
0
), q
i
(t
0
)}. Mathematically, this is because Hamiltons equations of motion
constitute a set of rst order ODEs for the phase space coordinates as a function
of time, and the solution of a rst order ODE is uniquely determined by the initial
values. Physically, it is because the position and momentum together completely
determine the motion of any particle or system.
An important corrollary of this is that trajectories in phase space never cross. Identical
states will have identical futures (and identical pasts).
It is worth looking at how this changes in quantum mechanics.
36
3.5 Liouvilles theorem [Optional]
Liouvilles theorem can be expressed in words as
As the (representative) states of the system evolve in time, their density in phase
space remains constant.
Another way of expressing it is that the volume in phase space occupied by N trajectories
remains the same throughout the evolution of the system.
This result is a cornerstone of statistical mechanics, as well as of chaos theory.
3.6 Poisson brackets
Consider some function f(q
1
, . . . q
N
, p
1
, . . . , p
N
, t), ie an arbitrary function of coordi-
nates, momenta, and time. Let us now nd the total time derivative (ie rate of change)
of this function. Using the chain rule we have
df
dt
=
f
t
+
N

k=1
_
f
q
k
q
k
+
f
p
k
p
k
_
=
f
t
+
N

k=1
_
f
q
k
H
p
k

f
p
k
H
q
k
_
, (3.34)
where in the last step we have made use of Hamiltons equations of motion. We can
write this in shorthand form as
df
dt
=
f
t
+{H, f} . (3.35)
{H, f} is called the Poisson bracket of H and f.
For two general functions f and g of coordinates and momenta, we dene the Poisson
brackets as
{f, g}

k
_
f
p
k
g
q
k

f
q
k
g
p
k
_
(3.36)
3.6.1 Properties of Poisson brackets
From the denition (3.36) we can immediately see that the Poisson bracket is antisym-
metric, {f, g} = {g, f}.
37
We can also prove the following relations for arbitrary functions f, g, h and constants
a, b, c:
{f, f} = 0 , (3.37)
{f, c} = 0 , (3.38)
{af + bg, h} = a{f, h} + b{g, h} (3.39)
{fg, h} = f{g, h} + g{f, h} (3.40)
{f, p
k
} =
f
q
k
, (3.41)
{f, q
k
} =
f
p
k
. (3.42)
For example, the proof of (3.38) is
{f, c} =

k
_
f
p
k
c
q
k

f
q
k
c
p
k
_
=

k
_
f
p
k
0
f
q
k
0
_
= 0 , (3.43)
since c is a constant and therefore does not depend on either p or q.
To prove (3.42) we note that all coordinates and momenta are independent variables
and therefore depend only on themselves,
p
k
q
i
=
q
k
p
i
= 0 i, k ;
p
k
p
k
=
q
k
q
k
= 1 k ;
p
k
p
i
=
q
k
q
i
= 0 if i = k. (3.44)
We therefore have
{f, p
k
} =

i
_
f
p
i
p
k
q
i

f
q
i
p
k
p
i
_
=
f
q
k
1 =
f
q
k
. (3.45)
Proving the other relations will be left as an exercise.
From (3.42), (3.41), we immediately nd the fundamental Poisson brackets, which are
the Poisson brackets of coordinates and momenta. They are
{q
j
, q
k
} = {p
j
, p
k
} = 0 j, k (3.46)
{p
j
, q
k
} =
jk
=
_
1 , j = k
0 , j = k
(3.47)
You may note the similarity between these expressions and the commutation relations
between the position and momentum operators in quantum mechanics,
[ p
i
, x
j
] = i h
ij
. (3.48)
38
This analogy was noted early on, and is often used to formulate the quantum mechanical
version of a classical mechanics system. In general, one may nd the Poisson brackets
of any two quantities f and g, and this will give the commutation relation between the
quantum mechanical operators

f, g,
{f, g}
i
h
[

f, g] . (3.49)
In particular, applying this to (3.35) yields Heisenbergs equation of motion,
d

f
dt
=
i
h
[

H,

f] +


f
t
, (3.50)
which is valid for all quantum mechanical operators.
1
3.6.2 Poisson brackets and conservation laws
From (3.35) we can see that if f does not depend explicitly on time, ie f = f(p, q),
then df/dt = {H, f}. Specically, for any motion integral (conserved quantity which is
a function of coordinates and momenta) I, we have
{H, I} = 0 . (3.51)
We can use this to demonstrate that a quantity is conserved, even if it is not the canonical
momentum of a cyclic coordinate. The example below will illustrate this.
Example 3.2 Angular momentum
Consider a particle with mass m moving in a central force (spherically symmetric)
potential V (r). The hamiltonian for this system is, in spherical coordinates (r, , ),
H = T + V =
p
r
2m
+
p
2

2mr
2
+
p
2

2mr
2
sin
2

+ V (r) . (3.52)
Since the azimuthal angle does not appear in H (ie, it is a cyclic coordinate), we
can immediately see that the canonical momentum p

is conserved:
p

= {H, p

} =
H

= 0 . (3.53)
1
In the Heisenberg picture, where the operators depend on time. This is in contrast to the
Schrodinger picture, where the states depend on time and the operators are in general independent
of time.
39
In a spherically symmetric system all angles should be irrelevant, so one would
navely expect that p

would also be conserved. However, because of the way we


singled out the z-axis when dening our coordinates, is not cyclic, and we get
p

=
H

=
p
2

mr
2
cos
sin
3

= 0 . (3.54)
Let us now consider instead the quantity
J
2
p
2

+
p
2

sin
2

. (3.55)
We can use Poisson brackets to nd the rate of change of J
2
:
dJ
2
dt
= {H, J
2
}
=
H
p
r
J
2
r

H
r
J
2
p
r
+
H
p

J
2

J
2
p

+
H
p

J
2

J
2
p

= 0 0 +
p

mr
2

_

2p
2

cos
sin
3

p
2

cos
mr
2
sin
3

_
2p

+ 0 0
= 0 .
(3.56)
Therefore, J
2
is conserved.
What is the physical meaning of J
2
?
We can write the angular momentum

L in spherical coordinates as

L =

p = m

r (v
r
r + v

+ v

) = mr r ( r r + r

+ r sin

) . (3.57)
The vector products of r with the unit vectors are
r r = 0 ; r

= ; r

= . (3.58)
Therefore,

L = mr
2

mr
2
sin

= p

sin

(3.59)

L
2
= p
2

+
_
p

sin
_
2
= J
2
, (3.60)
since the two vectors

,

are orthogonal. We see that J
2
is just the square of the
total angular momentum, J
2
= L
2
x
+ L
2
y
+ L
2
z
.
We have already shown that p

is the z-component of the angular momentum. It


may also be shown, using Poisson brackets, that
J
x
= p

sin p

cot cos (3.61)


is conserved. This will be left as an exercise, as will be guessing the physical meaning
of J
x
.
40
3.6.3 The Jacobi identity and Poissons theorem
One important relation that Poisson brackets obey is theJacobi identity,
{f, {g, h}} +{g, {h, f}} +{h, {f, g}} = 0 , (3.62)
for any three arbitrary functions f, g, h. The proof of this is straightforward but ex-
tremely tedious, so we will not show it here. We will however note that the same identity
holds for other antisymmetric products, including commutators and vector products:
[A, [B, C]] + [B, [C, A]] + [C, [A, B]] = 0 , (3.63)

A (

C) +

B (

A) +

C (

B) = 0 . (3.64)
The proof for commutators is very straightforward and relatively short:
[A, [B, C]] + [B, [C, A]] + [C, [A, B]]
= [A, BC CB] + [B, CA AC] + [C, AB BA]
= ABC ACB (BCA CBA) + BCA BAC (CAB ACB)
+ CAB CBA (ABC BAC)
= 0 .
(3.65)
We can use this identity to prove that the Poisson brackets {F, G} of two conserved
quantities F and G is also conserved:
d
dt
{F, G} =

t
{F, G} +{H, {F, G}} (3.66)
= {
F
t
, G} +{F,
G
t
} {F, {G, H}} {G, {H, F}} (3.67)
= {
F
t
+{H, F}, G} +{F,
G
t
+{H, G}} = 0 . (3.68)
In (3.66) we have used (3.35); in (3.67) we have used the product rule for /t and used
the Jacobi identity to rewrite {H, {F, G}}. Finally, in (3.68) we have used the general
properties of Poisson brackets, {f, g} = {g, f} and {f, g + h} = {f, g} +{f, h}.
This gives us
Poissons theorem:
dF
dt
=
dG
dt
= 0 =
d
dt
{F, G} = 0 . (3.69)
We can use this to construct further conserved quantities from ones we already know,
although this is not quite as useful as it may at rst seem: in many cases the Poisson
brackets of two conserved quantities will simply vanish.
41
Example 3.3
Given the hamiltonian (3.52) and J
x
dened in (3.61), we can, with a minimum of
calculation, construct a further conserved quantity J
y
. We already know that p

is
conserved, and using the Poisson brackets we have found that J
x
is also conserved.
Therefore the Poisson brackets of the two,
J
y
= {p

, J
x
} =
J
x

= p

cos + p

cot sin , (3.70)


is also conserved. You can probably guess what the physical meaning of this quantity
is.
42

You might also like