Вы находитесь на странице: 1из 32

Notes on Classical Dynamics – Prof.

Jon E NGEL

Zack L. Hutchens

Fall 2018

Contents
1 Survey of Elementary Particles 3
1.1 Single Particle Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Multi-particle Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 D’Alembert’s Principle & Lagrange’s Equations . . . . . . . . . . . . . . . . . . . . . 4
1.5 Velocity-Dependent Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Variational Principles & Lagrange’s Equations 6


2.1 Hamilton’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Some Techniques on the Calculus of Variations . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Extending Hamilton’s Principle to Systems with Constraints . . . . . . . . . . . . . . 7
2.4 Symmetries and Conserved Quantities in Lagrangian Mechanics . . . . . . . . . . . 8

3 Central Potentials 11
3.1 Reduction to an Equivalent One-Body Problem . . . . . . . . . . . . . . . . . . . . . . 11
3.2 The Equations of Motion and First Integrals . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 The Virial Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.4 The Differential Equation for the Orbit . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.5 The Kepler Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.6 Scattering & Collision Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.7 Transformation of the Scattering Problem to Laboratory Coordinates . . . . . . . . . 14

4 Rigid-Body Kinematics 15
4.1 The Independent Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Euler Angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

5 Rigid-Body Dynamics 17
5.1 The Inertia Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Stability and Torque-Free Motion in the Body-Fixed Frame . . . . . . . . . . . . . . . 17
5.3 Symmetric Top with Gravitational Torque . . . . . . . . . . . . . . . . . . . . . . . . . 18

6 Hamiltonian Mechanics 20
6.1 The Legendre Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.2 A Variational Principle for the Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . 20
6.3 Conserved Quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1
6.4 Phase Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.5 Hamiltonian of a Free Particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.6 Hamiltonian of a Simple Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

7 Canonical Transformations 24
7.1 Generating Functions and the New Hamiltonian . . . . . . . . . . . . . . . . . . . . . 24
7.2 The Symplectic Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.3 Liouville Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
7.4 Poisson Brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

8 The Hamilton-Jacobi Equation 27


8.1 The Complete Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
8.2 Free Particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
8.3 Central Force Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8.4 Action-Angle Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8.5 Adiabatic Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

9 Problem-Solving Cookbook 31

2
1 Survey of Elementary Particles
1.1 Single Particle Systems
Consider a particle with position r from some origin with velocity v = ṙ. It then has momentum
p = mv = mṙ. Based on this, Newton’s laws state that

• N1L: There exist references frames called “Galilean” or “inertial” in which a system with no
force acting upon it has constant momentum p and in which Newton’s other laws are true.

• N2L: The equation F = ṗ is true and reduces to mr̈ when m is constant in time. In a
noninertial frame, it is impossible to discriminate between a true F and the acceleration of
the frame; as a result, this law holds only in inertial frames.

• N3L: Law of action and reaction: F ij = −F ji .

From these we can derive three important conservation laws of Newtonian mechanics.

1. Conservation of Linear Momentum: If there is no force on an object, i.e., F = 0, then linear


momentum p is conserved in time.

2. Conservation of Angular Momentum: The angular momentum L = r × p is constant in time if


there is zero torque N = L̇ = 0.

3. Conservation of Energy: If the forces acting upon a particle are conservative, i.e. F i = −∇Vi ,
then the quantity T + V is conserved where T = m 2 v · v and V are the kinetic and potential
energies, respectively.

1.2 Multi-particle Systems


The force on the ith particle in a system of many particles can be decomposed into its net internal
and external forces as
(e)
X
Fi = F ji + F i = ṗi . (1)
j

The sum of the external forces is


(e)
X
Fi = M R̈
i
P
where M = i mi and P
i mi ri
R= (2)
M
is the system center of mass (COM). The implication here is that only the total external force
changes the COM’s trajectory; after all, internal momentum changes should cancel out due to
N3L. In other words, the linear momentum of the system is conserved in the absence of a net
external force.
Using the decomposition of the total force we can determine the net torque on the system
(e)
X X X
N= r i × ṗi = ri × F i + r i × F ji . (3)
i i i,j,i6=j

3
When the forces between pairs of particles internal to the system lie along a line connecting the
two particles, then r ij × F ji = 0 and we arrive at the strong law of action and reaction:
dL
= N (e) (4)
dt
and therefore angular momentum is conserved in absence of a total external torque. The total
energy T + V of a multi-particle system is also conserved, see Goldstein p. 9-11 for the derivation.

1.3 Constraints
Constraints provide conditions on how the system can move within a coordinate system, which
parts of the coordinate system for which it has access, and how it moves relative to itself. A cheap
example is a rigid body of N solidified particles, which keeps the distances r ij constant in time.
A holonomic constraint is one that has the form f (r 1 , r 2 , ..., t) = 0. A“ holonomic system” 1 is a
system that has the seemingly-natural property of requiring n general coordinates to describe its
n degrees of freedom. For example, a simple pendulum is a holonomic system because you need
only one coordinate to describe its motion, φ (deviation angle from the vertical), and because this
system has only one degree of freedom (also φ). Holonomic constraints usually do not depend on
velocities.
Nonholonomic constraints are those which do not satisfy this description: in these cases, the
path taken affects the state of the system. For example, the condition of static friction (slip-less
rolling) v = rφ̇ is a nonholonomic constraint. For a disk rolled along a circular path, subject to this
constraint, changing the radius of its circular path can return it to the same (x, y, θ) position, but
it does not have the same rolling angle φ at that instant. Nonholonomic constraints are usually
restrictions on velocities, and are not derivable from constraints on the position coordinates.2
The use of generalized coordinates for holonomic constraints solves the problem of dependent
coordinates r i . For instance, in the simple pendulum example, the fact that φ is the lone degree of
freedom means that the Cartesian coordinates x = ` sin φ and y = −` cos φ are not independent,
and the use of φ as a generalized coordinate eliminates the need for the Cartesian coordinates as
part of the equations of motion.

1.4 D’Alembert’s Principle & Lagrange’s Equations


We can define a quantity called the virtual displacement δr, which according to Goldstein is “con-
sistent with the forces and constraints imposed on the system at the given instant t.” This dis-
placement is not equivalent to a true displacement dx = vdt that would occur over a time interval
dt. Rather, we can think of it as such a small displacement that it does not evenly slightly disturb
the state of the system. The utility of a virtual displacement is that it occurs in zero time, and thus
it is unaffected by time-dependent constraints. Virtual displacements are generally tangent to the
forces of constraint so that they do no virtual work. For more information on virtual work and
displacement see the denoted references.3 4
1
According to John Taylor in Classical Mechanics
2
A. M. Bloch, J. E. Marsden, D. V. Zenkov, Nonholonomic Dynamics, Not. of the AMS 52, 324–333 (2005).
3
https://physics.stackexchange.com/questions/129786/what-exactly-is-a-virtual-displacement-in-clas
129817#129817
4
S. Ray and J. Shamanna, “On Virtual Displacement and Virtual Work in Lagrangian Mechanics,” (2005):
arXiv:physics/0510204 .

4
Suppose a system is in equilibrium so that F i = 0 for all i. If the constraints of the system are
such that they do no virtual work, then f i · δr i = 0 and thus
(a)
Fi · δr i = 0,

since the applied forces F (a) are related to the constraints by F = f + F (a) . However, we wish to
apply this to dynamic systems (nonzero Fi ) and for generalized coordinates. To achieve this, we
can rewrite Newton’s second law as F i − ṗi = 0, for which it is then clear that

(F i − ṗi ) · δr i = 0.

If we again separate the total force as F = f + F (a) , and still assuming that constraints do no
virtual work, we arrive at D’Alembert’s principle:
 
(a)
Fi − ṗi · δr i = 0. (5)

Here we could drop the notation of applied forces: it is possible to generate the equations of mo-
tions without considering the forces of constraint. The translation of the coordinates r i to the
generalized coordinates qα is shown in Goldstein p. 18–21, which can be used to derive the Euler-
Lagrange equations.

1.5 Velocity-Dependent Potentials


As an example with the nonconservative Lorentz force F = q (E + v × B), it is possible to have
velocity dependent potentials such as

U = qφ + qA · v,

where φ is the electric scalar potential and A is the magnetic vector potential. In these cases we
can use Lagrange’s equations as  
∂L d ∂L
− = Qi ,
∂qi dt ∂ q˙i
where Qi , in general, is a nonconservative force. For dissipative forces that are proportional to
velocities, we can use the Rayleigh dissipation function

1X 2 2 2

F= kx vix + ky viy + kz viz .
2
i

Then we substitute into the Euler-Lagrange equations

∂F
Qi = − ,
∂vx,i

with the ability to recover the frictional force as

F f = −∇v F.

5
2 Variational Principles & Lagrange’s Equations
2.1 Hamilton’s Principle
Hamilton’s principle presents a formulation of classical mechanics different than that of Newton’s
laws. Newton’s laws are concerned with the analyzing of forces acting upon a particle through
its motion. Lagrangian mechanics presents a different viewpoint: we can determine a system’s
motion by minimizing its behavior in order to satisfy some endpoints. Hamilton’s Principle states
that the motion of the system from time t1 to time t2 is such that the action integral
Z t2
I= Ldt, (6)
t1

has a stationary value for the actual path of motion, where L = T − V.

2.2 Some Techniques on the Calculus of Variations


For some action integral J of a functional f between end points x1 and x2 , i.e.
Z x2
J= f (x, y, y 0 )dx, (7)
x1

we can find a path y(x) for which J has a stationary value relative to other paths which are in-
finitesimally variant of y(x). I now will define a new function

y∗ (x, a) = y(x) + az(x), (8)

where a is a constant scalar and z(x) is a function that ascribes a variation to the path y(x): that
is, y∗ is an infinitesimally different version of y(x). We now wish to minimize this action integral,
which we can do by differentiating under the integral sign:
∂f (x, y∗ , y∗0 )
Z x2
dJ
= dx
da x1 ∂a
∂f ∂y∗0
Z x2  
∂f ∂y∗
= + 0 dx
x1 ∂y∗ ∂a ∂y∗ ∂a
Z x2  
∂f ∂f 0
= z(x) + 0 z (x) dx
x1 ∂y∗ ∂y∗
Z x2   Z x2  
∂f ∂f 0
= z(x)dx + z (x) dx.
x1 ∂y∗ x1 ∂y∗0
Now we can do the second of these integrals by parts, choosing u as the partial so that du =
d ∂f 0
dx ∂y 0 dx, and dv = z (x)dx, so that v = z(x). Then the second integral is
Z x2    x2 Z x2   Z x2  
∂f 0 ∂f d ∂f d ∂f
z (x) dx = z(x) − z(x)dx = − z(x)dx,
x1 ∂y∗0 ∂y∗0 x1 x1 dx ∂y∗0 x1 dx ∂y∗0

since the evaluative term uv vanishes – z(x) is zero at the endpoints because any function y∗ (x)
must connect those endpoints. Now we can plug this back into our original integral:
Z x2   
dJ ∂f d ∂f
= − z(x)dx.
da x1 ∂y∗ dx ∂y∗0

6
We wish to minimize the action integral, and that condition is dJ da |a=0 = 0. We also know that
y∗ → y in the limit that a → 0. Evaluating at a = 0 results in
Z x2   
dJ ∂f d ∂f
|a=0 = − z(x)dx.
da x1 ∂y dx ∂y 0

This result must be true for any well-behaved z(x) (for an illustration, choose z(x) = δ(x − x0 )),
so we find that  
∂f d ∂f
− = 0. (9)
∂y dx ∂y 0
We have now derived the Euler-Lagrange equations. For a simple example, we can use these
equations to find the shortest
p path between
p two points (x0 , y0 ) and (x, y) in Euclidean space. For
a functional f = ds = dx + dy = 1 + y 0 (x)2 dx, Lagrange’s equations simplify to
2 2

y0
p = const.,
1 + y 02

which can be integrated to find a linear solution y − y0 = m(x − x0 ).

2.3 Extending Hamilton’s Principle to Systems with Constraints


Semi-holonomic constraints of the form

fα (q1 , ..., qn ; q̇1 , ..., q̇n ; t) = 0

can be solved using variational principles, where α is an index counting the m constraint equa-
tions. Constraints on the velocities of physical systems are typically written as
m
X
fα = aαk q̇k + a0 = 0.
k=1

As an example of this equation, a block sliding on an incline of angle θ has the position constraint
tan θ = xy on the Cartesian coordinates, and thus it has a velocity constraint ẏ − ẋ tan θ. However,
this is a silly example in the sense that the force of p
constraint is holonomic and thus could be
eliminated in the use of a generalized coordinate d = x2 + y 2 , per the usual method.
At any rate, we have the freedom to express the constraint in this way. From the notes corre-
sponding5 to section 2.4 that Lagrange’s equations become

d ∂L ∂L
− = Qα = λi aiα ,
dt ∂ q̇α ∂qα

with the λi being the undetermined Lagrange multipliers. Here α = 1, 2, ..., n counting the coordi-
nates, and i = 1, 2, ..., m counting the equations of constraint. We now have n + m equations and
n + m unknowns, so we can solve the problem. A nice example of a hoop rolling down an incline
is shown in Goldstein p. 49-51. There, the force of constraint is static friction, preventing the wheel
from slipping, so the equation of constraint is rθ̇ − ẋ = 0. Consequently, the coefficients aiα are
aθ = r and ax = −1, where the subscript i has been dropped since there is only one constraint.
5
Note: different notation in class notes than book.

7
2.4 Symmetries and Conserved Quantities in Lagrangian Mechanics
Consider a Lagrangian which contains a generalized velocity q̇j but does not include the gener-
alized coordinate qj . In these systems, qj is defined to be a cyclic coordinate. The significance of a
cyclic coordinate is this: Lagrange’s equations for the cyclic coordinate simplify to
 
∂L d ∂L d ∂L
− =0= ,
∂qj dt ∂ q˙j dt ∂ q̇j

so we see that the quantity


∂L
= const.
∂ q̇j
In other words, when qj is a cyclic coordinate, the generalized momentum is conserved in time.
Thus cyclic coordinates provide us a useful trick for solving problems in Lagrangian mechan-
ics. For example, if an angular position θ is a cyclic coordinate, then angular momentum of that
coordinate is conserved, and we can rewrite θ̇ as θ̇ = `/I, where I is some moment of inertia.
Thus θ̇ might disappear from the Lagrangian, either in the form of a constant or as a function of
other variables (e.g., θ̇ = mr` 2 ). Every conserved quantity reduces the number of coordinates to solve the
problem.
This leads naturally into the discussion of Noether’s theorem. Here we define a continuous
symmetry as some continuous set of transformations of the coordinates which does not change
the Lagrangian. We just saw that conservation of generalized momentum is precipitated by the
existence of a generalized coordinate as a cyclic coordinate. Consider a one-dimensional system of
particles with positions xk , which we can translate under the transformation xk → xk + . Here, as
we will soon show, if the Lagrangian is invariant under this transformation, the symmetry leads
directly to a conserved quantity.
We define qj = qj (x) and qj, = qj (xk + ). How does the Lagrangian change under this
transformation? We can expand qj, with a Taylor series to see that
X ∂qj
qj, = qj + 
∂xk
k

and !
d X ∂qj
q̇j, = q̇j + ,
dt ∂xk
k

neglecting higher-order terms in the assumption that  is small. The implication of a symmetry
between the coordinates xk and displaced coordinates xk +  is that

∂L
|=0 = 0,
∂
which is to say that the LagrangianP
is invariantP
under this transformation. We know this to be true
because the total potential energy V (xk ) = V (xk + ) is not changed under this translation,
and neither are the velocity-dependent kinetic energies (δL = δT + δV = 0). Expanding this
derivative with the chain rule shows that
∂L(qj , q̇j ) ∂L dqj ∂L dq̇j
|=0 = + .
∂ ∂qj d ∂ q̇j d

8
Remembering that qj, = qj +  this simplifies to

∂L X ∂qj ∂L d X ∂qj
+ =0
∂qj ∂xk ∂ q̇j dt ∂xk
k k

Several skips were skipped here, but primarily we rewrote the -partials, as, e.g.

dqj, X ∂qj, ∂xk, X ∂qj


|=0 = |=0 = ,
dxk, ∂xk, ∂ ∂xk
k k

dx
noting the fact that dk, = 1. We can apply the same process for the second term, but factoring
d
out the dt operator. Now we can make a substitution using the Euler-Lagrange equation:

d ∂L X ∂qj ∂L d X ∂qj
+ =0
dt ∂ q̇j ∂xk ∂ q̇j dt ∂xk
k k
" #
d X ∂L ∂qj
→ = 0,
dt ∂ q̇j ∂xk
k

for which the last equation is a simple application of the derivative product rule. This equation
then beautifully shows that
X ∂L ∂qj X ∂L X
= = pk = ptotal = const.
∂ q̇j ∂xk ∂ ẋk
k k k

We have now shown that when the Lagrangian is invariant under a spatial translation xk → xk +,
the total momentum (or of the center of mass) is then conserved in time.
This progression was a case-specific instance of invariance under position translations. In gen-
eral, continuous symmetries lead to conserved quantities through Noether’s theorem, proven in
1918 by Emmy Noether, a mathematical theorem which is not only fundamental to analytical me-
chanics but also quantum mechanics and quantum field theories.

Noether’s Theorem
Consider some transformation qj → qj, = qj + fi (q) + O(2 ). If the Lagrangian is invariant under
this transformation, i.e.
∂L
|=0 = 0,
∂
then
∂L
fi
∂ q̇j
is conserved.
Our next task to explore how the Lagrangian changes in time. Taking a simple derivative and
applying the multivariable chain rule, we see that

dL ∂L ∂L ∂L
= q̇j + q̈j +
dt ∂qj ∂ q̇j ∂t

d ∂L ∂L ∂L
= q̇j + q̈j + ,
dt ∂ q̇j ∂ q̇j ∂t

9
by Lagrange’s equations. The first two terms are an application of the product rule, so
 
dL d ∂L ∂L
= q̇j + .
dt dt ∂ q̇j ∂t
∂L
If we assume that L does not depend explicitly on time, so that ∂t = 0, we can define a quantity

∂L
h≡ q̇j − L = const.
∂ q̇j

which is constant in time. The quantity h is called the “Jacobi integral” or “energy function”, and
it is equivalent to the total mechanical energy E when kinetic energies are of the form

T = αik q̇i q̇k .

Thus it would not work for some  2 rotational velocities; for example, in our spinning bead + wire
example, we saw that T = m 2 . In that case, h 6= E. But either h or E could

2 α̇ + (d sin α · ω)
still be conserved. The energy function is numerically equivalent to the Hamiltonian, but they are
functions of different variables: h = h(q, q̇, t), but H = H(q, p, t).

10
3 Central Potentials
3.1 Reduction to an Equivalent One-Body Problem
A system of two masses m1 and m2 with positions r 1 and r 2 , interacting only due to some poten-
tial U , will have a Lagrangian of the form

L = T (Ṙ, ṙ) − U (r),

where r is the relative position vector r = r 2 − r 1 . Of course, we can write the total kinetic energy
as T = TCM + Trel , which is the sum of the center of mass’ kinetic energy and the kinetic energies
relative to the center of mass. Here Trel = 12 m1 ṙ12 + 12 m2 ṙ22 , which can be written with the reduced
mass
m1 m2
µ=
m1 + m2
as
1
Trel = µṙ2 .
2
Therefore, we can rewrite our Lagrangian as

1 1
L= (m1 + m2 ) Ṙ2 + µṙ2 − U (r),
2 2
since U depends only on the relative position vector rather than the CM position. Consequently,
R is a cyclic coordinate, and the momentum of the center of mass is constant in accordance with
Lagrange’s equations. This allows us to group the Lagrangian terms of the center of mass, as
   
1 2 1 2
L= (m1 + m2 )Ṙ + µṙ − U (r) = LCM + Lrel .
2 2

We can now address the Lagrangian in terms of its CM and relative components. As before, this
Lagrangian always returns the result (m1 + m2 )Ṙ = c for some constant c. The information of the
rotational dynamics lies within the relative component. With that, we have reduced a two-body
problem to an equivalent one-body problem. The simple choice of reference frame of the CM,
in which Ṙ = 0, means that L = Lrel .

3.2 The Equations of Motion and First Integrals


We now consider systems for which the central force is strictly conservative, corresponding to a
potential V (r) (rather than, e.g., U (r, ṙ)), and for which we have a one-body problem of reduced
mass m (we are in the CM frame). As the central force is always along r, the total torque N =
r × F = 0 always. Thus, the angular momentum L is conserved, and the motion is confined to a
plane defined by L. The effect, once again, is that the conserved quantity has reduced the degrees
of freedom of our problem. Our Lagrangian in this plane becomes

1
L = T − V = m(ṙ2 + r2 θ̇2 ) − V (r),
2

with ` = mr2 θ̇ the conserved quantity. Solving the Lagrange equation for r, we find the first
equation of motion
∂V
mr̈ − mr2 θ̇ + = 0,
∂r

11
∂V
or, substituting f (r) = ∂r and ` = mr2 θ̇,

l2
mr̈ − = f (r),
mr3
which is a second-order differential equation of r. We find a second equation of motion in the fact
that the total mechanical energy E = T + V is constant. For that constant of motion, we have

1 2 1 `2
mṙ + + V = E, (10)
2 2 mr2
which can be solved by separating variables, assuming the separable equation is analytically inte-
grable.

3.3 The Virial Theorem


Define the virial K as X
K= r i · pi . (11)
i

Since the particles are in a bound system, K < Kmax , and the time average of K̇ is
 
D E K(t) − K(−t)
K̇ = lim = 0.
t→∞ 2t
After all, the motion is periodic for the bound system so the symmetry K(t) = K(−t) should be
supplied. Then doing the simple derivative

K̇ = ṙ i · pi + r i · ṗi = 2T + r i · ṗi ,

we can see that D E


K̇ = h2T i + hr i · ṗi i = 0
or equivalently
h2T i = − hr i · ṗi i . (12)

3.4 The Differential Equation for the Orbit


Our new goal is to obtain a general equation for the physical orbit r(θ), rather than having the gen-
eralized coordinates written as functions of time. We can do this by expressing the first constant
of motion ` as
d ` d
=
dt mr2 dθ
and plugging it into the first equation of motion
∂V
mr̈ − mr2 θ̇ + =0
∂r
`2
 
1 d ` dr
→ 2 − = f (r).
r dθ mr2 dθ mr3
Now make the simple substitution u = 1r , and rewrite the force as the gradient of the potential:

d2 u
 
m d 1
2
+u=− 2 V . (13)
dθ ` du u

12
3.5 The Kepler Problem
For an inverse-square law central force, we have f (r) = kr−2 and V = −k/r. Now, if we substitute
this into the differential equation for the orbit, we arrive at the equation of the orbit
r !
1 mk 2E`2
u= = 2 1+ 1+ cos (θ − θ0 ) . (14)
r ` mk 2
q
2E`2
Defining the eccentricity  = 1+ mk2
, we have

1 mk
u= = 2 (1 +  cos (θ − θ0 )) . (15)
r `
We can classify the orbital equation for different values of . For  = 0, we have a circle (no
2
eccentricity) with E = −mk
2`2
. On the other extreme,  = 1 corresponds to a parabolic orbit of
E = 0. For  > 1 we have a hyperbola of E > 0, and likewise, for  < 0 we have a ellipse of E < 0.

3.6 Scattering & Collision Theory


We now want to examine a new type of central force problem. In the investigation of nuclear and
atomic physics, the scattering6 of particles against some target is important for understanding
fundamental physics principles.
We start our discussion with defining two quantities: the impact parameter b, which is defined
as the perpendicular distance from the target’s center to the projectile’s incoming trajectory, and
the scattering angle θ, which is the angle between the projectile’s outgoing and incoming velocity
vectors. Intuitively, there is an inherent relationship between the two b(θ), or θ(b): as the impact
factor gets very large, the projectile totally misses the target, and θ goes to zero. Contrarily, when
b → 0, the projectile strikes head-on, and we have a maximum scattering angle θ = π.
Now let us further specify that the projectiles enter through a ring of inner radius b (the impact
parameter) and width db. The collision7 cross section is clearly

dσ = 2πbdb.

We wish to derive an answer in terms of the number of particles scattered into a solid angle dΩ,
which in spherical coordinates is dΩ = 2π sin θdθ. Dividing, we find the simple equation for the
differential cross section
dσ b db
= . (16)
dΩ sin θ dθ
Denoting ψ as the angle between the center horizontal of the target and the radius of closest ap-
proach rm (“perapsis”), we can see that geometrically that θ = π − 2ψ. Applying the orbital differ-
ential equation (see Goldstein eq. 3.36) to this case, you can see that
Z ∞
dr
ψ= q ,
rm r 2 2mE 2mV 1
`2
− `2
− r2

where we have the initial conditions r0 → ∞ and θ0 = π, and secondary conditions at closest
approach r = rm and θ = π − ψ (see Goldstein Fig. 320, p. 108). An equivalent statement, given
6
see Goldstein 3.10, p. 106 & Taylor Ch. 14, p. 557: I draw from both books in this section.
7
Goldstein and Taylor adopt slightly different terminology here. Taylor defines the cross section as σ, Goldstein has
σ as the differential cross section.

13

the relationship θ = π −2ψ, and plugging in the constant angular momentum ` = mv0 b = b 2mE,
we have Z ∞
sdr
θ(b) = π − 2 r   , (17)
rm 2 V (r) 2
r r 1− E −s

which can be further rewritten in terms of u = 1/r (see Goldstein eq. 3.97, p. 108).

3.7 Transformation of the Scattering Problem to Laboratory Coordinates


To this point, we have only discussed systems of scattering particles in which the target remains
fixed. In reality, due to the third law and conservation of momentum, the target will also scatter
in the reciprocal force of the projectile. To solve, this we have to translate what we have in the
center-of-mass frame to the corresponding quantities in the laboratory reference frame. We define
Ψ as the scattering angle in the laboratory frame. The scattering angle in CM coordinates, θ, is
what we know. Thus our immediate goal is to find the relationship between these two angles.
First note that any velocity in the laboratory frame, vj , is related to the CM velocity by vj =
vj + vCM , where vj0 is a particle velocity in the CM frame. We can obtain two equations
0

v1f cos Ψ = v10 cos θ + vCM

and
v1f sin Ψ = v10 sin θ
by simply combining conservation of momentum and energy (as in introductory physics inelastic
scattering problems). Dividing these two, we find that

sin θ sin θ
tan Ψ = vCM = m1 .
cos θ + v0 cos θ + m 2
1

14
4 Rigid-Body Kinematics
4.1 The Independent Coordinates
In the first chapter, we saw that a rigid-body was defined as a group of particles subject to the
holonomic constraint that the distance between pairs of particles r ij is held constant. According
to Goldstein, the maximum number of degrees of freedom for a rigid body of N constituents is 3N .
However, these can be greatly reduced in the application of constraints. Notably, to determine the
location of a point in the rigid body, we need not know its relative location to all other particles, but
only to three noncollinear points on the body. Since these points are subject to the same constraint
rij = cij , for constant c, we can reduce the number of constraints to six.8
In principle, the orientation of a rigid body can be measured simply from the angular devia-
tion(s) of the coordinate axes and a set of axes “glued” to object. For example, consider an object
that is rotated in R2 . The Cartesian unit vectors transform under the rotation
 0   
x̂ cos θ sin θ x̂
= .
ŷ 0 − sin θ cos θ ŷ
Of course, for a physical system, we would need a 3×3 matrix to cover all of space. To do so, we
will index our Cartesian unit vectors as ei = {x̂, ŷ, ẑ}, with rotated vectors e0i . Then, any particle
will have the position
r = xi e i ,
with xi = {x, y, z}. This allows us to relate any rotated vector to its unrotated counterpart under
the relationship
ej = Aij e0i ,
such that from the former equation r = xj Aij e0i . Therefore, we find the elemental equation x0i =
Aij xj , which has the matrix equivalent
x0 = Ax.
Here A is simply some matrix that expresses the coordinates in a rotated basis. However, consider
the dot product
x0i yi0 = Aij xj Aik yk = xi yi ,
since dot products are invariant under rotation (lengths don’t change in the rotated frame). The
implication is that Aij Aik = δjk , or equivalently, that A is orthogonal matrix. Consequently, A is
an element of the Lie symmetry group SO(3).
As an extra note, consider that there are two interpretations of these rotations. One is the active
transformation, in which the vectors r i are physically rotated within in a fixed rotating frame. The
other is the passive view, in which vectors ”remain stationary” and the coordinate system itself
will rotate the corresponding amount. We’ll use this concept rigorously in the following sections
and chapter.

4.2 Euler Angles


Consider a rotation A1 for an object in 3D by an angle φ. We have a rotation matrix given by
 0   
x cos φ sin φ 0 x
y 0  = − sin φ cos φ 0 y  .
z0 0 0 1 z
8
More about constraints on rigid bodies: Goldstein section 4.1, p. 135.

15
Other rotations A2 and A3 out of the plane, with angles θ and ψ respectively, can be given by
   
1 0 0 cos ψ sin ψ 0
A2 = 0 cos θ sin θ  and A3 = − sin ψ cos ψ 0 .
0 − sin θ cos θ 0 0 1

Then, for a combination of these rotations, any rotated object will have the coordinates given by

x000 = A3 A2 A1 x,

where three primes represent that three rotations have taken place.
All this is to say that we have defined the orientation of the object using three angles: an az-
imuthal angle φ, a polar angle θ, and a spinning angle ψ: all contingent on the assumption that
the rigid body has one point fixed in space. More properly, Euler’s theorem guarantees that any
rotation of an object can be defined by a single direction and angle.

Euler’s rotation theorem


Any rotation of an object can be defined by a single direction and angle. That is, any matrix
A ∈ SO(3) has one and only one eigenvalue of 1, except for the identity element.
Define a direction n̂ and corresponding angle of rotation, α, which is a spin about n̂. Then any
rotated vector can be expressed as

r 0 = r cos α + n̂(r · n̂)(1 − cos α) + r × n̂ sin α.

By examining the rotations of these objects with these A matrices (see Engel notes and Goldstein
next) we can arrive at a force law for the noninertial frame

mr̈ = F 0 − 2mω × ṙ − mω × ω × r − mω̇ × r. (18)

Let’s examine these forces. The first of these, F 0 is the usual force that would be determined in the
corresponding inertial reference frame. It is the true net force in rotating coordinates; it, unlike the
latter, is not fictitious. The second term, −2mω × ṙ, is the Coriolis force. This term quantifies the
“curve” of moving objects seen by observers in a rotating frame. It has important applications to
meteorology because it is used in models of hurricanes’ spiraling winds in the Earth’s reference
frame.9
The third term −mω × ω × r is the centrifugal force which is ubiquitous to our everyday
experiences. It’s what causes to sway to the side in a turning car, and it’s used in salad spinners to
help dry out the leaves. As for the last term, −mω̇ × r: don’t worry ’bout it. It deals with reference
frames for which ω changes in time, and we are not really interested in solving those problems in
this class.

9
Contrary to belief, toilets in the Southern hemisphere do not flush backwards due to the Coriolis effect.

16
5 Rigid-Body Dynamics
5.1 The Inertia Tensor
We now want to give some equation like N = L̇ with an active interpretation (body-free frame/axes
fixed). To get L we have
X X
L= ma xa × v a = ma xa × (ω × xa ) (19)
a a

noting that v = ẋ = ω × x. Evaluating the ith component yields


X
ma |xa |2 δil − xai xal ωl

Li = (20)
a

or
Li = Iil ω l . (21)
The components Iil form the inertia tensor. Now we want to do the same in the body-fixed refer-
ence frame, as denoted by primed coordinates. Similarly we can show that
X  0 
Iil0 = ma |xa |2 δil − xai xal .
a

This is the same but with primed vectors, so we can conclude that I is a symmetric tensor. It is a
special case of a Hermitian operator, and it will have three real eigenvalues and three orthogonal
eigenvectors. In general, I is not diagonal, but we always have the freedom to choose a reference
frame in which it is. The physical directions specified by these three eigenvectors of L are called
the principal axes. For eigenvalues λi , we have L = λi ωi . In other words, the principal axes are
where L is parallel to ω, just like in introductory physics.

5.2 Stability and Torque-Free Motion in the Body-Fixed Frame


Suppose we throw some object, such as a chalkboard eraser, into the air. Which axes are stable?
What axes tumble? In this section we will attempt to address this question. We first need an
expression for the kinetic energy of the object:

1X a a2 1 0 1X
T = m |v | = ω · Iω 0 = λi ωi02 . (22)
2 a 2 2
i

With torque free motion we know L0 = Iω 0 and L0i = Iii ωi0 since I is diagonal. The total angular
momentum is given from
L02 = L02 02 02
1 + L2 + L 3 , (23)
and thus the total kinetic energy in terms of the angular momenta is
3
1 X L02
i
T = . (24)
2 λi
i=1

L doesn’t change in the lab frame because it is a torque-free system. But due to the noninertial
forces, it’s all over the place in the body-fixed frame.

17
Let’s choose the condition λ1 > λ2 > λ3 . Notice that equation (23) forms a sphere of allowed
momenta in angular momentum space, and likewise, equation (24) √is the equation for an ellipsoid
in angular momentum space which has semi-major axis length 2λi T . To test stability we can
“bump” it to give it more kinetic energy (or less angular momentum). The intersection points of
the sphere and the ellipsoid determine the stability. For any singular point of intersection, you
will have rotation about some corresponding axis only. Intersections that are loops or circles will
coincide with stable motion because it forces L to be confined to a particular region of angular
momentum space. Contrarily, wedges or strips (like colors on a beach ball, or like an orange
slice) do not confine the motion of the L vector, allowing it to swing all the way up and down,
so that motion will be unstable. Overall, we will have two stable and one unstable axis, for our
blackboard eraser.
What we would like to do now is write down Newton’s laws for torque-free motion in the
body-fixed frame. We have L̇ = 0, and from the derivation given in Taylor10 , we have
0
L̇ + ω 0 × L0 = 0. (25)

Rewriting this, we see that


λi ω̇i0 + ijk ωj0 λk ωk0 = 0. (26)
This gives the Euler equations of motion with no torque:

λ1 ω˙1 0 − ω20 ω30 (λ2 − λ3 ) = 0 (27)

λ2 ω˙2 0 − ω10 ω30 (λ3 − λ1 ) = 0 (28)


λ3 ω˙3 0 − ω10 ω20 (λ1 − λ2 ) = 0. (29)
Now let’s apply these to a body with one symmetry axis, i.e. λ1 = λ2 := λ. Equation (29) reduces
to ω30 = const. Solving equations (27) and (28) returns a coupled differential equation which can
be written as
ω̈10 = −Ωω˙2 0 = −Ω2 ω10 . (30)
0 cos(Ωt) and ω 0 = ω 0 sin(Ωt) for constant ω 0 . This means that
This can be solved to find ω10 = ω⊥ 2 ⊥ ⊥
0
ω1 precesses at frequency Ω around the symmetry axis. In the lab frame, L is constant, and the
object is spinning and precessing. We can define the frequencies as ωs = ψ̇ and ωp = φ̇, for which
L
ωp = . (31)
λ
The object precesses at the same speed regardless of angle. We just need to know the angular
momentum, although this itself could have an angle-dependence. We can also show that ωs = −Ω,
where Ω is the precession frequency of L.

5.3 Symmetric Top with Gravitational Torque


We now want to analyze the motion of a symmetric top subject to a uniform gravitational field.
Due to the symmetry, the gravitational force can be modeled as exerting a force M g at the center
of mass of the top. We can figure out the motion of the system using the Lagrangian formalism.
We know its kinetic energy will be
1  1
T = λ ω102 + ω202 + λ3 ω302 (32)
2 2
10
Taylor has a good derivation of rotated quantities in his book.

18
with potential energy M gl cos θ. Given velocity in the Euler angles (see Goldstein ch. 4), we can
write the Lagrangian as

λ 2  λ 
3
2
L= θ̇ + φ̇2 sin2 θ + ψ̇ + φ̇ cos θ − M gl cos θ. (33)
2 2
First of all, note that ψ and φ are cyclic coordinates, so their generalized momenta are conserved.
This gives us
pψ = λ3 ω30 = L03 = λa (34)
and
pφ = λ sin2 θφ̇ + λ3 (ψ̇ + φ̇ cos θ) = L3 = λb. (35)
Consequently, we will have to solve one – not three – differential equations. Algebra gives

b − a cos θ
φ̇ =
sin2 θ
and
(b − a cos θ) cos θ
ψ̇ = .
sin2 θ

19
6 Hamiltonian Mechanics
In this section we again wish to reformulate our approach to studying classical mechanics. We
will expand on our understanding of Lagrangian mechanics, demonstrating a new quantity called
the “Hamiltonian” which is related to the Lagrangian through a Legendre transformation. This
Hamiltonian will allow us to solve Hamilton’s equations and to understand the motion and inter-
actions of our system in a phase space defined by the conjugate momentum and conjugate position
coordinates. This phase space has important analytical implications for, e.g. Liouville’s theorem,
and for the following chapters.

6.1 The Legendre Transformation


We define our new quantity H as the the Legendre transformation of L:

H = q˙i pi − L, (36)

where the generalized momenta are given by


∂L
H = pi = . (37)
∂ q˙i
With these two equations we can compute the Hamiltonian of our system from its Lagrangian and
generalized momenta. Very often H is the sum of the kinetic and potential energies, T + V . Earlier
in our notes we discussed h, the energy function. To repeat an important note from that section,
H and h are numerically the same value. Both are transformations of the Lagrangian. However,
they are functions of different variables: h = h(q, q̇, t), whereas H = H(q, p, t). The Hamiltonian is
the total energy of the system when (i) the kinetic energy is quadratic, homogeneous function of
the generalized velocities and (ii) when the potential energy is strictly determined by generalized
coordinate and time only (no velocities).

6.2 A Variational Principle for the Hamiltonian


In Chapter 2 we defined the action of a system as
Z t2
I= Ldt. (38)
t1

Now, by replacing with our Legendre transform, we wish to obtain a variational principle for the
Hamiltonian: Z t2
I= (q˙i pi − H) dt. (39)
t1

In Hamiltonian mechanics we are interested in a phase space defined by the generalized coordi-
nate and its generalized momentum – these are taken to be independent variables. Thus both qi
and pi will satisfy Lagrange’s equations for our action integral. Doing the first, for qi , we see that
 
∂L d ∂L
− =0
∂qi dt ∂ q˙i

which for L = q̇i pi − H becomes


∂H
p˙i = − . (40)
∂qi

20
Likewise  
∂L d ∂L
− =0
∂pi dt ∂ p˙i
will yield
∂H
q˙i = . (41)
∂pi
These last two results are known as Hamilton’s equations. They are a pair of coupled differential
equations which, given some Hamiltonian, will generate equations for the motion of the system.
In principle these can be solved to find p(t) and q(t), as well as their phase space trajectories.

6.3 Conserved Quantities


There are two important notes to make in this section. The first is to note that like in Lagrangian
mechanics, when qi is cyclic, then corresponding momentum is conserved. We can easily see this
from Hamilton’s equations. If H is constant in q then

∂H
ṗ = − = 0,
∂qi

so p is conserved. The other item of interest is when the Hamiltonian itself is conserved. For this
to happen we require that dH
dt . Therefore, we need

dH ∂H ∂H ∂H
= q̇ + ṗ + = 0, (42)
dt ∂q ∂p ∂t

by the multivariable chain rule and since H = H(q, p, t). By inserting Hamilton’s equations into
the preceding line for q̇ and ṗ, we find that the first two terms will cancel and we are left with

dH ∂H
= = 0,
dt ∂t
which is to say that the Hamiltonian is a conserved quantity in time when it does not contain an
explicit time-dependence.

6.4 Phase Space


A principal distinction of Hamiltonian mechanics is that it provides solutions to mechanical prob-
lems in phase space. Previously, we have only considered such problems in a configuration space,
for we could define the location and orientation of any object with the generalized, independent
coordinates q1 , q2 , ..., qn ; t. These coordinates can be taken together as a vector space, which is the
aforementioned configuration space. In other words, configuration space is like the physical space
in which a system resides. For example, a point mass moving in three spatial dimensions might
have the independent coordinates t and r = hx, y, zi, and thus its configuration space is R3 .
Phase space is different in that it takes the parameters q and p = ∂L ∂ q̇ as independent coordi-
nates. Using Hamilton’s equations we can certainly find q(t) and p(t). Phase space is defined by
p and q and is parameterized by time. We will explore some examples of the Hamiltonian and
phase space in the following sections.

21
Figure 1: The phase space trajectories of a free particle. Each line corresponds to a unique initial
condition, and thus they will never intersect. The initial condition fully determines the motion of
the system. Every possible evolution of the motion of the system has a unique q0 and p0 , so the
phase space has infinitely-many trajectories, each of which will never intersect with the others.
Note: The trajectories are not continuous over all space since they start at a discrete point (q0 , p0 ),
which has not been illustrated here.

6.5 Hamiltonian of a Free Particle


A free particle of position q and mass m has a Lagrangian L = 12 mq̇ 2 . Therefore, the generalized
momentum is p = ∂L ∂ q̇ = mq̇. Since we want H(q, p, t), we can re-arrange to see that q̇ = p/m. The
Legendre transform yields
p2
H= (43)
2m
as we would expect. Simple application of Hamilton’s equations gives
∂H p
q˙i = =
∂pi m
and
∂H
p˙i = − = 0.
∂qi
p
Note that the first of the equations of motion can be separated to find that q = q0 + m t. Therefore
the phase space trajectory is a series of lines shown in Figure 1.

6.6 Hamiltonian of a Simple Pendulum


For another ordinary example, we will consider a simple pendulum. For a swinging angle φ and
origin at the pivot, we will have something like
1
L = mL2 φ̇2 + mgL cos φ. (44)
2

22
Figure 2: Phase space trajectories of a simple pendulum, taken from Ball & Holmes (2007). The
closed orbits about the origin indicate a pendular oscillation, for which the angular momentum
of the mass smoothly oscillates through its range of angle φ, approaching its maximum and min-
imum angular momentum at each axis. The other trajectories describe a motion for which the
pendulum completely orbits (in a single direction) the pivot in configuration space.

Then, the Hamiltonian will be

pφ 2
H = q̇i pi − L = − mgL cos φ, (45)
2mL2

since pφ = mL2 φ̇, where pφ is in this case an angular momentum about the pivot. Note here
that H = E = T + V which is conserved since there are no explicit appearances of time in the
Hamiltonian. Hamilton’s equation’s give us

φ̇ = (46)
mL2
and
ṗφ = −mgL sin φ. (47)
The astute reader might notice that this last equation is the torque on a point mass subject to a
uniform gravitational field, which can be derived using earlier methods or even first principles.
At any rate, Hamilton’s equations have given us a pair of first-order, coupled differential equations
which define the phase space of the system, as shown in Figure 2. 11

11
R. Ball & P. Holmes (2007). Dynamical Systems, Stability, and Chaos. arXiv:nlin/0702044

23
7 Canonical Transformations
7.1 Generating Functions and the New Hamiltonian
In the last chapter, we discussed a reformulation of classical mechanics dealing with phase space
and the Hamiltonian. We are now interested in determining transformations of coordinates which
are canonical, which means that a new Hamiltonian of the transformed coordinates will still obey
Hamilton’s equations. Consider a transformation of the coordinates from
qi , pi , H(q, p, t) → Qi , Pi , K(Q, P, t). (48)
Here K is the “new Hamiltonian,” which is also sometimes called the “Kamiltonian.” We want
Hamilton’s equations to obeyed in the new coordinates, which means we need physical paths that
minimize Z t2  
Pi Q̇i − K(P, Q, t) dt (49)
t1
and satisfy Hamilton’s equations. This will be true if
∂F
pi q̇i − H(q, p, t) = Pi Q̇i − K(Q, P, t) + . (50)
∂t
The function F is called a generating function. Let’s suppose we have some F1 (q, Q, t) – we just
want to see what happens.
dF ∂F ∂F ∂F
= q̇i + Q̇i + (51)
dt ∂qi ∂Qi ∂t
 
∂F
= pi − q̇i − H(q, p, t)
∂qi
 
∂F ∂F
= Pi + Q̇i − K(Q, P, t) + .
∂Q ∂t
This will be true if F1 obeys the following conditions:
∂F1
1. pi = ∂qi ,
∂F1
2. Pi = − ∂Q i
, and
∂F
3. H(q, p, t) = K(Q, P, t) − ∂t .
Now let’s do some examples.

Example F1 = qk Qk
Since this is a type-I (F1 ) generating function, we can use the previous equations. We find pi =
∂F ∂F
∂qi = Qi and Pi = ∂Qi = −qi . Together these yield, for all i,
Qi = pi , Pi = −qi . H = K, (52)
with Hamilton’s equations obeyed.

Example F = F2 (q, p, t) − Qk Pk
Given the last line of Eq. 51 we find that
∂F2 ∂F2 ∂F2
pi = , Qi = , H=K− . (53)
∂qi ∂Pi ∂t
Example F2 (q, P ) = qk Pk
∂F2 ∂F2
This will yield the identity: pi = ∂qi = Pi and Qi = ∂Pi , and H = K.

24
7.2 The Symplectic Condition
We have now done a few examples, but we have not determined how we can test if a transforma-
tion of the Hamiltonian is canonical in the first place. We will do just that in this section. First,
define η as a vector containing all the phase space coordinates, i.e.
   
q1 η1
 q2   η2 
   
 ...   η3 
   
η=  p1  =  η4  .
   (54)
 p2   ... 
   
 ...   ... 
pN η2N
We know the entries of η will obey Hamilton’s equations, so we can obtain a matrix equation
 ∂H 
   ∂η
∂H
1
0 I

 ∂η2
η̇ = , (55)

−I 0  ...


∂H
∂η2N
where the matrix is in block form – both I’s represent their own identity matrix. All this matrix
∂H
does is map the top of entries of η̇ (the q̇’s) with the bottom half of the ∂η i
’s (the ∂H
∂p ’s) and likewise
for the bottom entries. We call this matrix J so that
∂H
η̇ = J . (56)
∂η
Furthermore, if we define ζ as a vector containing the new coordinates (just like η but with Q’s
and P ’s), then this last result will yield
∂H
ζ̇ = M η̇ = J , (57)
∂ζ
since ζ will also obey Hamilton’s equations with the new Hamiltonian. In this equation, M is the
Jacobian matrix, where the elements are defined as
∂ζi
Mij = . (58)
∂ηj
We can now note two equations: ζ̇ = M η̇ = M J ∂H ∂H T ∂H
∂η and ∂η = M ∂ζ . In combining these we
obtain
∂H
ζ̇ = M JM T . (59)
∂ζ
Since this is supposed to follow Hamilton’s equations, we can conclude both that M IM T = I and
M JM T = J, (60)
which is to say that M is a symplectic matrix. These symplectic matrices form a Lie group. By
noting that J = J −1 we can find that
M T JM = J, (61)
which is called the symplectic condition. A transformation of the Hamiltonian is canonical when
its coordinates satisfy this condition. This will yield what are called the “direct conditions,” which
are given in Goldstein. The direct conditions are series of partial derivative tests based on the
symplectic condition to determine if the transformation is canonical. Note that an equivalent
verification is that det M = ±1.

25
7.3 Liouville Theory
We can show through a separate derivation that the flow in phase space is symplectic (see Prof.
Engel’s notes, or Goldstein). Given η(t) and η0 , the elements of the Jacobian will be

∂ηi
Mij = (62)
∂η0 j

Some volume in phase space dV0 will change in time with the phase space trajectories, due to the
actual motion of the system. Let’s say that the new volume is dV . Then,

dV = | det M |dV0 ,

which, since det M = ±1, means that the volume of the flow in phase space does not change.
Every point in phase space evolves according to Hamilton’s equations, flowing like a fluid, and
this fluid is incompressible – the volume remains constant everywhere in the flow. This result is
called Liouville’s theorem, which has significant implications not only in mechanics but is the root
of all classical statistical mechanics.

7.4 Poisson Brackets


In this section, we will look at yet another way of determining if a transformation is canonical. We
define the Poisson bracket of f (q, p) with g(q, p) as

∂f ∂g ∂f ∂g
[f (q, p), g(q, p)]η = − . (63)
∂qi ∂pi ∂pi ∂qi

This can be equivalently expressed as


   
∂f ∂g
= J .
∂η ∂η

Poisson brackets are canonical invariants, since


df (q, p, t) ∂f
= [f, H] + . (64)
dt ∂t
To show that the transformation is canonical, we need to show that {qi , qj } = 0, {pi , pj } = 0, and
{qi , pj } = δij . The first two are trivial, of course.

26
8 The Hamilton-Jacobi Equation
8.1 The Complete Integral
We have now discussed how the use of canonical transformations can allow us to transform our
problem into one that is easier to solve. Our new task to apply this knowledge to a special trans-
formation and apply what is called the Hamilton-Jacobi equation. Suppose we found a generating
function that makes the new Hamiltonian zero:

K = 0. (65)

Then, by virtue of Hamilton’s equations, we have

∂K ∂K
Q̇i = =0 and Ṗi = − = 0, (66)
∂Pi ∂Qi

which means that Qi and Pi are constant in time. The generating function that satisfies this trans-
formation is
F = −Qi Pi + F2 (q, P, t), (67)
which means that
∂F2
K =0=H+ .
∂t
We now have a partial differential equation of F2 . Let’s assume the solution is

F2 = S(q1 , q2 , ..., α1 , α2 , ..., αn+1 ; t) (68)

which is separable by addition of functions. The α’s are the momenta, so

∂F2
αi = pi =
∂qi

in accordance with a type-II transformation. The final result is

∂S
H+ = 0. (69)
∂t
which is the Hamilton-Jacobi equation. (Don’t worry – we’ll do some examples to help this make
sense.)

8.2 Free Particle


∂S
For a free particle, we have H = p2 /2m per usual. The momentum is ∂q , so the Hamilton-Jacobi
equation gives the PDE
 2
1 ∂S ∂S
=− .
2m ∂q ∂t

Let’s√assume a separable form as S(q, α, t) = W (q, α) − α1 t. Solving it yields W = 2mαq, so
S = 2mαq − αt. Then the momentum is
∂S √
p= = 2mα, (70)
∂q

27
which means that α is the total energy. Likewise, since S is a type-II generating function, we can
find Q: r
∂S ∂S m
Q=β= = = q − t. (71)
∂P ∂α 2α
To solve for the motion, just rearrange for q(t) in terms of Q, not forgetting that this method
guarantees that Q is constant:
p
q(t) = const. + t. (72)
m
Do not take confusion in the fact that α = E and α = P . In the transformed coordinate, the
“momentum” is actually in the energy, and the “coordinate” (Q) is actually the angle in phase
space!

8.3 Central Force Motion


Now let’s do a more complicated example: central force motion. Here the configuration space has
multiple degrees of freedom as r and φ for motion confined to a plane. The Hamiltonian will be
!
p 2
1 φ
H= p2r + + V (r),
2m 2m

where, of course, V (r) is a central potential. Separation of variables as S = Wr + αφ φ + α1 t (note


φ is cyclic) into the Hamilton-Jacobi equation reduces to
" #
∂W 2 αφ

1
+ 2 + V (r) = α1 , (73)
2m ∂r r

with the kind reminder that αφ = ∂S ∂φ – the generalized momentum in φ is conserved. We can
reduce this differential equation to quadrature to find that
Z r
αφ
W = 2m (α1 − V (r)) − 2 dr + αφ φ, (74)
r
where the second term is Wφ . Now we can find β:
Z
∂W ∂Wr αφ
βφ = − = +φ=− q dr. (75)
∂αφ ∂αφ r2 2m (α1 − V (r)) −
αφ
r2

The problem has been fully reduced to quadratures. The time-dependence of the motion is given
by
∂W
t + β1 = . (76)
∂α1

8.4 Action-Angle Variables


We are now going to introduce the idea of action-angle variables to solve problems. These are
radial coordinates in phase space that can help to describe the motion of physical systems, partic-
ularly for periodic motion. They are different ways of describing the same space. To do so we’ll
define the action12 as I
J = pdq (77)

12
This is not the same action we used in Chapter 2 for variational principles. We will use J for both, though.

28
Thus J is the area of a loop in phase space. We want a new Hamiltonian H̃ = K = α1 to be a
function of J, where α1 = α1 (J). To satisfy this demand we need a new generating function from
the original p and q to action-angle variables. We’ll have

W̃ (q, J) = W (q, α1 (J))

with
∂W (q, α) ∂ W̃ (q, J)
p= = .
∂q ∂q
Then “Q”= ∂∂J

= w, and consequently,

∂ H̃
ẇ = = ν(J) = const. (78)
∂J
This is simply another form of Hamilton’s equations, but with the action-angle variables. We see
that w is the canonical conjugate to J. At a closer look, let’s look at the change in w over a single
cycle: !
I   I I
∂w ∂ ∂ W̃ ∂ d
∆w = dq = dq = pdq = J = 1. (79)
∂t ∂q ∂J ∂J dJ
That means that w = θ/2π is the phase space relative angle, where θ is an arbitrary (and not
useful beyond instruction) variable. Now let’s do an example with a harmonic oscillator. The
Hamiltonian (which is the total energy) is

1
p2 + mω 2 q 2 = α1 = E.

H=
2m
Rearranging and integrating (using a trigonometric substitution) gives
I I p
2πE
J = pdq = 2mE − mω 2 q 2 dq = . (80)
ω
The frequency of rotation in phase space is then

∂ H̃ ω
ν(J) = = , (81)
∂J 2π
since the Hamiltonian as function of J is
ωJ
H̃ = α1 (J) = E = . (82)

This is nice, because means we didn’t have to find the generating function after all.

8.5 Adiabatic Invariance


For the last topic we are going to show that J is an adiabatic invariant. Adiabatic changes to the
system constitute those which occur very slowly (think adiabatic processes in thermodynamics).
Consider a harmonic oscillator in which the mass and spring stiffness are changing very slowly.
We have a Hamiltonian like
p2 1
H(t) = + k(t)q 2 = E.
2m(t) 2

29
The energy changes in time in accordance with

p2 p2
   
1 ṁ k̇ 1 2
Ė = ṁ + k̇q 2 = − + kq . (83)
2m2 2 m 2m k 2

Now don’t forget that m and k are changing very slowly, so the energy is also slowly changing. If
we return to the virial theorem (Chapter 3), we showed for homework that in a harmonic oscillator,
hT i = hV i = E2 . Therefore, !
E k̇ ṁ
Ė = − . (84)
2 k m
Now we can just separate and integrate to find that

E
∼ const. (85)
ω
By equation (82) this means that
J
∼ const., (86)

so we see that J is an adiabatic invariant.

30
9 Problem-Solving Cookbook
This “cookbook” outlines the solutions of different types of problems we considered during PHYS
701: Classical Dynamics. I used this during our doctoral written examination (qual-like) at UNC.

Coordinate systems.

1. Cylindrical Coordinates (r, φ, z): v 2 = ṙ2 + r2 φ̇2 + ż 2


2. Spherical Coordinates (r, φ, θ): v 2 = ṙ2 + r2 φ̇2 sin2 θ + r2 θ̇2 ..

Lagrangian Mechanics with only holonomic constraints.

1. Choose a coordinate system and write down the position r of the particle. Be sure to choose
the origin such that the frame is inertial.
2. Get v = ṙ, and compute the kinetic energy: T = 21 mv · v.
3. Write down the potential energy, and then the Lagrangian L = T − V.
4. Solve Lagrange’s equations to get the equations of motion (EOM), and solve those differen-
tial equations if needed.

Lagrangian Mechanics with constraint forces.

1. Identify the equation of constraint.


2. Get the kinetic and potential energies of the particle, as given in the above steps. Do not use
the constraint to reduce the number of degrees of freedom of the system.
3. If it isn’t already, massage the equations of constraint so that they have the form aαi q̇α + bi .
dL d ∂L
4. Construct the Euler-Lagrange equation: dq α
− dt ∂ q̇α = λi aiα , where λi is the corresponding
force of constraint.
5. Use Lagrange’s equations to get the equations of motion. De-couple them to solve for λ.
6. If the problem asks, e.g., where the constraint force is zero, let λ = 0 and solve accordingly.

Conserved quantities & symmetries of Lagrangian mechanics.


∂L ∂L
1. If qi is cyclic in L, then pq = ∂ q̇ is conserved. This is equivalent to saying ∂q = 0.
2. The energy function h is conserved when dh ∂L
dt = − ∂t = 0. Both conditions need to be satisfied.
3. If the Lagrangian is time-independent (i.e., t is cyclic), the energy is conserved. E is constant
when ∂L ∂t = 0.
4. h = E when (a) U = U (qi , t), i.e. the potential is not a function of the generalized velocities
q̇i and (b) q̇i ∂∂Tq̇i = 2T, which is to say that the kinetic energy is a homogenous, quadratic
function of the generalized velocities.

Central potentials: Keplerian orbits.

1. Use the first integral or the differential equation for the orbit to solve for the motion as
needed, see tagged pgs. in Goldstein.
2. To test the stability of the orbit, make the substitute r → r0 + (t). Then you can reduce the
equation of motion to ¨ = −ω 2  which will show that the orbit is stable with frequency ω.
df
3. Recall that β 2 = 3 + fr dr |r=r0 . If β(= p/q) is a rational number, then after q revolutions of the
radius vector, the orbit begins to retrace itself so that the orbit is closed.

31
Central potentials: scattering.
1. The goal is to get b(θ). For central potentials it is helpful to know the integral for θ(b), which
is eq. 3.96 on p. 108 in Goldstein. For other problems, e.g. billiard balls, you will just need a
geometrical analysis.
2. The differential cross section is σ(θ) = sinb θ | dθ
db
|. Compute it, and don’t forget that b is a
function of the θ.

3. The total cross section can be calculated as σT = 2π 0 σ(θ) sin θdθ.
Physics in noninertial reference frames.
1. Choose a coordinate system. Typical choice is z is up, y is North, and x is East.
2. Identify the direction of ω and v 0 . The direction of the coriolis force will be that of v 0 ×ω. The
direction of the centrifugal force will be outward the rotation, as given by the corresponding
cors
3. Write down the second law in rotating coordinates: mr̈ = F 0 − 2mω × ṙ − mω × ω × r.
Very often the centrifugal can be absorbed into F 0 with an effective gravitational field. Solve
accordingly.
Rigid-body dynamics; spinning tops; nutation and precession.
1. For a rigid body with one point fixed, Euler’s equations of motion are given on Goldstein p.
200.
Hamilton’s equations.
1. Find the Lagrangian as described above.
2. Generate the Hamiltonian through the Legendre transformation: H = q̇i pi − L.
3. Equations of motion can be generated from q̇i = ∂H ∂H
∂pi and ṗi = − ∂qi .

Canonical transformations.
1. See the table on canonical transformation generating functions in the book.
2. To test if a transformation is canonical you must verify the symplectic condition M JM T = J.
P  ∂f ∂g 
3. Alternatively, compute the Poisson bracket: {f, g} = N i=1 ∂qi ∂pi − ∂f ∂g
∂pi ∂qi .
4. It must be that {qi , qj } = 0, {pi , pj } = 0, and {qi , pj } = δij for the canonical transformation.
Hamilton-Jacobi formalism.
1. Get the Hamiltonian, if not given. Substitute H = − ∂S ∂t (the HJ equation), and for the mo-
∂S
menta, substitute, e.g. px = − ∂x .
2. If the Hamiltonian is time-independent, then you can separate variables using S = Wx (x) +
Wy (y) + ... − α1 t. If the Hamiltonian is time dependent you are going to have a bad day. You
have to be careful about the S you choose, and sometimes you will need to pick it to cancel
out the time-dependent part of the Hamiltonian.
3. Once you have all the individual W’s, recombine them to get an equation for S.
4. The new position coordinate is given as Q = β = ∂S ∂α . It will be a function of q and t. Then
you can rearrange it to get q(t).
5. The momentum will be p = ∂S ∂q . Alternatively just differentiate q and multiply by m.

32

Вам также может понравиться