S7 and BT VII: Classical mechanics

John Magorrian, HT 2012

magog@thphys.ox.ac.uk

These notes accompany the lectures for the Physics Short Option S7 / Physics & Philosophy BT VII course on classical mechanics. You can nd the most up-to-date version at http://www-thphys.physics.ox.ac.uk/user/JohnMagorrian/cm.pdf . Last updated: 13 Feb 2012.

Syllabus

(Sections marked

are not covered in the HT2012 course)

Calculus of variations: EulerLagrange equation, (variation subject to constraints) . Lagrangian mechanics: principle of least action; generalized co-ordinates; conguration space. Application to motion in strange co-ordinate systems, particle in an electromagnetic eld, normal modes , rigid bodies . Noethers theorem and conservation laws. Hamiltonian mechanics: Legendre transform; Hamiltons equations; examples; principle of least action again ; Liouvilles theorem ; Poisson brackets; symmetries and conservation laws; canonical transformations. HamiltonJacobi equation .

Recommended reading T. W. B. Kibble & F. H. Berkshire, Classical mechanics, 5th ed. About 19. The single most suitable book for this course. L. D. Landau & E. M. Lifshitz, Mechanics. About 30. First volume of the celebrated Course of Theoretical Physics. Succinct. H. Goldstein, C. Poole & J. Safko, Classical mechanics, 3rd ed. About 50. Covers more advanced topics too. Verbose. Supplementary reading The following books are more dicult, but some might nd them inspiring for a second pass at the subject. V. I. Arnold, Mathematical methods of classical mechanics Adopts a more elegant, more mathematically sophisticated approach than the other books listed here, but develops the maths along with the mechanics. G. J. Sussman & J. Wisdom, Structure and interpretation of classical mechanics. About 45, but also freely available online. Uses a modern, explicit functional notation and breaks everything down into baby steps suitable for a computer.

S7 & BT VII: Classical mechanics

0 Some maths
0.0 Notation
Vectors:
x = xi + y j + z k

(0.1) (0.2)

(in 3d) or
x = x1 x 1 + x2 x 2 + + xn x n

for the general case. Gradients of function f (x, x ): f f x 1 + x x1 f f x 1 + x x 1 So,

p

f x 2 + , x2 f x 2 + . x 2

(0.3)

f f f = p1 + p2 + . x x1 x2

(0.4)

0.1 An introduction to the calculus of variations

Recall that a function is simply a rule for mapping elements of one set (the functions domain) to elements of another set (its range). A functional is a mapping from the set of all functions that satisfy some specied conditions (e.g., the set of all smooth maps from the real line to three-dimensional space) to the real numbers. Examples include: the nth moment
x1

In [y ] =

xn y (x) dx

(0.5)

x0

of a one-dimensional function y (x) dened for x0 < x < x1 and having y (x0 ) = y (x1 ) = 0; the length of a curve x(t) joining two xed points x(t0 ) and x(t1 ) in n-dimensional space,
t1

L[x] =

t0

|x | dt;

(0.6)

the gravitational potential energy of a mass distribution (x),

1 V [] = 2 G

(x)(x ) d3 xd3 x . |x x |

(0.7)

For this course we need only consider functionals of the form

t1

F [ x] =
t0

L(x, x , t) dt

(0.8)

that eat smooth one-dimensional curves x(t) with xed endpoints x(t0 ) = x0 , x(t1 ) = x1 in an n-dimensional space. The rst two examples above are of this form. The third is not. Internally, the functional runs over the curve, feeding the local values of (x, x , t) to a function L and accumulating the results. Note that this L treats x, x and t as independent variables; it does not know that x = dx/dt!

S7 & BT VII: Classical mechanics Now lets look at how the output of the functional changes when we distort the curve slightly from x(t) to x(t) + h(t). The variation of the curve h(t) must be smooth and vanish at the endpoints in order that x + h be admissible to F , but is otherwise arbitrary. The variation or dierential of the functional F [x; h] lim
0

F [ x + h] F [ x ]

(0.9)

An extremal is a curve x(t) for which F [x; h] = 0 for all admissible h(t). Finding these extremals (if they exist) is the business of the calculus of variations. Fundamental lemma of the calculus of variations t0 < t < t1 , satises
t1 t0

If a smooth curve f (t), dened on the range (0.10)

f (t) h(t) dt = 0

for all continuous h(t) having h(t0 ) = h(t1 ) = 0, then f (t) 0. Proof by contradiction: we show that if f = 0 then equation (0.10) would not be true for all h(t). Suppose that there were some tblip between t0 and t1 for which f (tblip ) = 0. Then, because f has no discontinuous jumps we can always nd a small interval (tleft , tright ) around this tblip where f = 0. Now consider the function
h(t) = f (t)

(tright t)(t tleft ), for tleft < t < tright , 0, otherwise.

(0.11)

This clearly satises the conditions of the lemma, but

t1 t0 tright

f (t) h(t) dt =

tleft

f 2 (t)(tright t)(t tleft ) dt > 0,

(0.12)

since the integrand is positive between tleft and tright . So, weve shown that if f = 0 anywhere then we can always nd some h(t) that makes f h dt = 0. Turning this around, if there is no h for which the integral is non-zero, then we must have f = 0 between t0 and t1 . Now we come to the key result of this section. Let F [x] be a functional of the form
t1

L(x, x , t) dt,
t0

(0.13)

dened on the set of smooth functions x(t) satisfying boundary conditions x(t0 ) = x0 and x(t1 ) = x1 . Then a curve x(t) is an extremal of F if and only if it satises the EulerLagrange equation, d dt L x L = 0. x (0.14)

Proof: If x is an extremal of F then for any variation h we have 0 = F = lim =

t0 t1

t1 t0

0 t1

, t) L(x, x L(x + h, x + h , t) dt (0.15) L h x

t1

L L h+ h dt x x L d x dt L x h dt + ,
t0

=
t0

where the last line follows from the previous one using integration by parts. The nal term on the last line vanishes because the boundary conditions mean that h(t0 ) = h(t1 ) = 0. Thus (0.15) becomes
t1

0=
t0

L d x dt

L x

h dt

(0.16)

S7 & BT VII: Classical mechanics for any smooth h. Applying the fundamental lemma, our extremal curve x(t) must satisfy the Euler Lagrange equation (0.14). Conversely, if a curve x(t) satises (0.14) then it is clear from (0.15) that it is an extremal of the functional F . Example: the shortest path between two points Consider the set of smooth curves in the (t, x) plane that pass between two xed points x (t0 ) = x0 and x(t1 ) = x1 . The path length of any such curve x(t) 2 . The EL equation for this L is is given by the functional (0.13) with L = 1 + x d dt x 1+x 2 = 0, (0.17)

since L/x = 0 and L/ x = x/ 1+x 2 . Therefore extremals satisfy x = A, a constant. Integrating, x = At + B , with the constants A and B completely determined by the condition that the curve pass through the two xed points. Important: When writing down the EL equation, remember that x and x are independent arguments of L. Use the fact that along extremals x(t) satisies x = dx/dt only when solving (i.e., integrating) for x(t). Easy rst integral when L does not depend explicitly on t (Beltrami identity) Solving the EL equation often leads to lots of messy algebra. But if L = L(x, x ), the EL equation can be reduced to the rst-order dierential equation x (L/ x ) L = constant. To see this, note that on solution paths f dx f dx f f df = + = x + x dt x dt x dt x x for any function f (x, x ). So, d x dt L x L = x L d L L L +x x x x dt x x x d L L =x = 0. dt x x (0.18)

(0.19)

Look out for a more physical way of deriving this result later in the course! Example: minimal surface of revolution Among all the curves that pass through the points (t0 , x0 ) and (t1 , x1 ), nd the one that generates the surface of minimum area when rotated about the t-axis. A physical example is a soap bubble drawn between two coaxial circular hoops. The surface area generated by a curve x(t) is
t1

2
t0

x 1+x 2 dt.

(0.20)

Comparing to equation (0.13), we see that L(x, x ) = 2x 1 + x 2 , independent of t. Using the result above for general L(x, x ), the extremals of (0.20) satisfy xx x 1+x 2 1+x 2 = A. (0.21)

This is easily rearranged, via x = A 1 + x 2 , to give Ax =

x 2 A2 . t+B A

(0.22)

Integrating, the curve that extremizes the surface area of revolution is x(t) = A cosh , (0.23)

where the constants A and B are chosen to satisfy the boundary conditions x(t0 ) = x0 and x(t1 ) = x1 . Depending on the choice of x0 and x1 , there can be zero, one or two solutions for (A, B ). Of course, in the zero-solution case an extremal does exist, but it is not smooth and therefore lies beyond the remit of the machinery developed above.

S7 & BT VII: Classical mechanics

0.2 Variation subject to constraints

Sometimes it is necessary to nd extremals of a functional F [x] (equation (0.13)) subject to a constraint of the form g (x, t) = 0 (0.24) among the co-ordinates. From (0.15), the condition for a curve x(t) to be an extremal is then
t1

F [x, h] =
t0

L d x dt

L x

h dt = 0

(0.25)

for any smooth h(t) that satises g g + + hn = 0, (0.26) x1 xn since we must have g (x + h, t) = 0. This last condition means that we cannot use the fundamental lemma directly. Instead let us multiply (0.26) by an arbitrary function (t) and insert it into the integrand of (0.25). This combines the two conditions (0.25) and (0.26) into one:
h g = h1
t1

0=
t0

L d x dt

L x

g h dt. x

(0.27)

The function (t) is a Lagrange multiplier. Now suppose that, say, g/x1 = 0. Then we may choose (t) to make L d L g + =0 (0.28) xi dt x i xi for i = 1, so that the x 1 term in the integrand of (0.27) vanishes. We are then free to vary (h2 (t), . . . , hn (t)) independently as long as we choose h1 (t) to ensure that the constraint condition (0.26) holds. Using the fundamental lemma on (0.27) we see that the relation (0.28) must apply for i = 2, . . . , n as well as for i = 1. Therefore extremals of F [x] subject to the constraint g = 0 satisfy d dt L x L g = . x x (0.29)

This results in n + 1 equations (n components of EL equation plus the constraint g = 0) for n + 1 unknowns (x1 , . . . , xn and ). In practice one usually takes linear combinations of dierent components of the EL equation to eliminate (t). For this reason is sometimes known as Lagranges undetermined multiplier. Alternatively Introduce a new co-ordinate and a new functional
t1

G[x, ]

(t)g (x, t) dt
t0

(0.30)

that acts on curves {x(t), (t)} in this (n + 1)-dimensional space. Obviously, variations of {x(t), (t)} that satisfy g = 0 will have G = 0 too. Combining the two conditions F = 0 and G = 0 into one,
t1

(F + G) =
t0

L dt = 0

(0.31) (0.32)

with

Writing down the x and components of the EL equation for L , we nd that d L L g = , dt x x x g = 0.

}, t) L(x, x L ({x, }, {x , , t) + g (x, t).

(0.33)

It is clear that a path x(t) found by solving (0.33) satises the constraint g = 0 and therefore G 0 and so G = 0 too. Since (F + G) = 0 by construction, the path is an extremal of F too, F = 0. Exercise: At this point one might object that the EL equation for L applies only to paths {x(t), (t)} with xed endpoints. The conditions on the functional (0.13) mean that we are given x(t0 ) = x0 and x(t1 ) = x1 . What do we know about (t0 ) and (t1 )?

S7 & BT VII: Classical mechanics

0.3 Legendre transforms

Given a function f (x), its Legendre transform g (p) is another function that encodes the same information as f (x) but in terms of p = df /dx instead of x. A necessary condition for the Legendre transform to exist is that the rst derivative f (x) be strictly monotonic, so that either f > 0 everywhere or that f < 0 everywhere. Consider the set of (non-vertical) lines in the (x, y ) plane, y = ax b. Introduce another plane and to each line in the original plane assign a single point (a, b) in the new plane. The (a, b) plane is known as the (projective) dual of the original (x, y ) plane. Since the relation b + y = ax still holds if we exchange (a, b) with (x, y ), it follows that the dual of the (a, b) plane is the original (x, y ) plane; each plane is the dual of the other. Now take a smooth curve y = f (x) in the original (x, y ) plane. This curve traces out another curve in the dual (a, b) plane, the point (x, f (x)) being mapped to a = f (x), b = xf (x) f (x). For example, the plots below show the curve y = f (x) = x sin x (left) and its image (right) in the dual (a, b) space. The second derivative f (x) changes sign at the point B .
B

A B A

a x

If f (x) is convex (f > 0) then a increases monotonically with x and we can dene the Legendre transform of f (x) as g (a) xf (x) f (x) (0.34) = xa f (x), where x(a) is the point on the original curve where f (x) = a. That is, b = g (a) is the dual to the curve y = f (x) and vice versa. Exercise: For the higher-dimensional case in which hyperplanes y = a x b map to points (a, b) in the dual space, show that the Legendre transform of a function f (x) is g (a) = x a f (x), where x(a) is the point for which f = a.

In particular, for later use note that the Legendre transform of a function L(q ) is given by H (p) = q p L(q ), where q in the RHS is expressed in terms of p = L/ q .

Example from thermodynamics: the Helmholtz free energy A(T, V, N ) = U T S is the Legendre transform of the internal energy U (S, V, N ) with respect to T = U/S .

y=

b ax

S7 & BT VII: Classical mechanics

1 Lagrangian mechanics

1.1 Hamiltons principle of least action

Consider a particle of mass m whose potential energy V (x; t) is independent of its velocity. Its equation of motion is d V mx = . (1.1) dt x This is equivalent to the EL equation (0.14), d L L = 0, dt x x (1.2)

1 if we choose L/ x = mx and L/ x = V / x, or L = 2 mx 2 V . Now suppose we were given the instantaneous positions of the particle at times t0 and t1 . The results above imply that the path that the particle takes between these two xed points is an extremal of the action integral, t1

S [ x]

L(x, x , t) dt,
t0

(1.3)

Now let us consider a system of N particles, having masses mi , positions xi and for which the potential energy is V (x1 , . . . , xN ; t). The latter includes the eects of inter-particle interactions, such as gravity or electrostatic repulsion, as well as any externally applied forces, but we assume that it does not depend on the particles velocities. If we again take L({xi }, {x i }; t) = T V to be the dierence between the kinetic and potential energies of the whole system of particles, then the EL equations L d L =0 dt x i xi reduce to the standard Newtonian equations of motion: d V mi x i = dt xi (i = 1, . . . , N ). (1.5) (i = 1, . . . , N ), (1.4)

where the Lagrangian L = T V is the dierence between the particles kinetic and potential energies.

We can think of the system of particles as moving in 3N -dimensional conguration space. Given snapshots of the 3N co-ordinates of the full system at times t0 and t1 , we see that the path the system traces out in conguration space at intermediate times is an extremal of the action
t1

S ({xi (t)}) =

t0

L({xi }, {x i }, t) dt.

(1.6)

Notice that the condition for a curve to be an extremal of the action (1.3) is independent of the particular co-ordinate system we use to describe the curve. This means we can use any sensible co-ordinate system to parametrize the curves we feed in to the action integral and the EL equation will return the extremal curve (i.e., the equation of motion) in that co-ordinate system. We describe our mechanical system using a set of generalized co-ordinates, q (t) (q1 (t), . . . , qn (t)), that pin down the instantaneous position of the system in n-dimensional conguration space. We assume that there is no redundancy among the qi , so that the system has n degrees of freedom. The system moves through conguration space with a generalized velocity q (q 1 , . . . , q n ). Now suppose we know that q (t0 ) = q0 and q (t1 ) = q1 . Then the general form of Hamiltons principle of least action states that the path in conguration space the system takes between these two times is an extremal of the action integral
t1

S [q ]

L(q , q , t) dt,
t0

(1.7)

S7 & BT VII: Classical mechanics where the Lagrangian L is a function only of the generalized co-ordinates, the generalized velocities and time. Therefore, the equation of motion of the system is d L L = 0. dt q q (1.8)

The quantity p L/ q is known as the generalized momentum of the system, F = L/ q is the generalized force. Some comments: (i) In this formulation we assume only that that L is some scalar function of (q , q , t), which we are free to choose in order to make the EL equations (1.8) match the true equations of motion of the system. For the common case in which the particles move in a velocity-independent potential V (q , t) we know from the examples above that a suitable choice is L = T V . (ii) L is not unique. For example, for any function (q , t) we can add d/dt to L and still obtain the same equations of motion. (Prove it!) (iii) Dierent elements of q can have dierent units. Therefore dierent elements of the generalized momentum p and generalized forces L/ q can have dierent units too. (iv) If one has external (generalized) forces that are not accounted for in L, they can be added to the RHS of (1.8).

1.2 Why bother?

The most obvious advantage of the Lagrangian approach to mechanics over the elementary Newtonian approach is that it allows us to derive the equations of motion of many mechanical systems without the tedious task of resolving forces. As L is a scalar quantity we are free to use whatever co-ordinate system we like to label points in conguration space: we may express L in terms of those co-ordinates, turn the handle and obtain the equations of motion. It does not directly tell us how to solve the equations of motion though. A related benet of the Lagrangian approach is that it makes a deep connection between symmetries and conservation laws. Problems involving mechanical systems are often invariant under some continuous transformations (e.g., rotation about a particular axis or translation in a certain direction), which means that there is a corresponding constant of motion (see 1.6 below). This often suggests the most natural coordinate system to use for the problem, which can then help us to solve the equations of motion explicitly, or at least teach us something qualitative about the behaviour of solutions. Lagrangian mechanics really comes into its own when modelling the motion of rigid bodies (A.2 below). We probably wont have time to cover that topic during lectures though. The methods were applying to mechanical systems in this course can also be applied to other problems (see, e.g., the rst two chapters of Goldstein). Much of modern, non-classical physics is derived from some form of action principle.

1.3 Equations of motion for some simple systems

Simple pendulum A bob of mass m is attached to one end of a rigid massless rod of length l. The other end of the rod is attached to a xed point, about which the rod can rotate in a xed vertical plane. The most natural parameter to use to describe the instantaneous conguration of this one-dimensional pendulum is the angle the rod makes with the vertical. Since the potential energy V () = mgl cos is independent , we have that of generalized velocity
1 2 + mgl cos , ml2 L=T V = 2

(1.9)

= ml2 . The equation of motion (1.8) for the system is then + (g/l) sin = 0. so that p L/

Springy pendulum Replace the rigid rod in the simple pendulum above with a massless spring of natural length l and spring constant 2 , so that when the string is extended or compressed to a length r its

S7 & BT VII: Classical mechanics

2 2 potential energy Vspring (r) = 1 2 (r l) . The natural generalized co-ordinates to use for this system are (r, ). A Lagrangian in these co-ordinates is 1 2 + mgr cos 1 2 (r l)2 , 2 + 2 mr2 L = T (Vgrav + Vspring ) = 1 2 mr 2

(1.10)

which yields the equations of motion d 2 + mg cos 2 (r l), mr = mr dt d = mgr sin . mr2 dt

(1.11)

Spherical pendulum Now lets return to our simple pendulum constructed from a rigid rod, but relax the constraint that the rod can rotate only in a xed plane. The Cartesian co-ordinates of the location of the bob with respect to the pivot can be written as x = l sin cos y = l sin sin z = l cos , where (, ) are the usual polar co-ordinates of a point on the surface of a sphere. We orient our co-ordinate system with the Oz axis pointing downwards, so that is the angle the bob makes with the downwards vertical. Dierentiating (1.12) with respect to time to nd x (, ), we have that the Lagrangian
1 L= 1 2 V = 2 m[x 2 + y 2 + z 2 ] + mgl cos 2 mx 2 + 2 sin2 ] + mgl cos . = 1 ml2 [ 2

(1.12)

(1.13)

The generalized momenta (p , p ) conjugate to the generalized co-ordinates (, ) are given by p L = ml2 , p L = ml2 sin2 . (1.14)

Notice that these are both angular momenta. In particular, p is the angular momentum about the z axis and the EL equation for , p = L/ = 0, tells us that p is conserved. The EL equation for is = ml2 2 sin cos mgl sin ml2 = p2 cos
3

ml2 sin

mgl sin ,

(1.15)

to eliminate . where in the second line we have used our expression for the constant p = ml2 sin2 Exercise: It is dicult to integrate equation (1.15) to obtain an explicit expression for as a function = (d/ d), explain how (1.15) can be used to obtain an expression for as of t. Using the fact that a function of . Show that the motion reduces to motion in a one-dimensional eective potential Ve () = p2 2ml2 sin2 mgl cos , (1.16)

and explain how to nd the minimum and maximum values of taken by the pendulum for a given set of initial conditions. Particle in a central eld The location of particle of mass m moving in three dimensions in a spherically symmetric gravitational potential (r) is most naturally expressed using spherical polar coordinates, q = (r, , ), in terms of which (x, y, z ) = (r sin cos , r sin sin , r cos ). (1.17)

S7 & BT VII: Classical mechanics ), the Lagrangian Since V = m does not depend on q = (r, , 2 + r2 sin2 2 V (r), L=T V = 1 2 + r2 2m r (1.18)

10

where the velocity x 2 in the square brackets comes from dierentiating the co-ordinate transform (1.17) with respect to t. The EL equation (1.8) gives the equations of motion 2 + mr sin2 2 dV , p r = mr dr where the components of the generalized momentum pr mr, p mr2 , and p mr2 sin2 . (1.20) 2, p = mr2 sin cos p = 0, (1.19)

p is a constant because p = 0. As the potential is spherically symmetric, we can orient our co-ordinate = 0 from (1.19), showing that the motion remains conned to the system so that = 2 initially. Then p plane = 2. and Exercise: Write down the Euler equation for r. In the equation you get, use (1.20) to express in terms of the constants p and p . Show that this motion is identical to that obtained from the one-dimensional eective Lagrangian, 2 Ve (r), L(r, r ) = 1 2 mr with Ve (r) = V (r) +

2 p2 + p . 2mr2

(1.21)

and from L(r, , r, ) and then A common temptation is to try to save work by rst eliminating , to obtain the EL equations from the resulting Lagrangian L(r, , r, p , p ). This is wrong! Why? If a co-ordinate qi does not appear explicitly in L, then L/qi = 0 and the EL equation tells us that the corresponding momentum pi L/ q i is conserved. Such qi are known as cyclic or ignorable coordinates.

1.4 Particle in a magnetic eld

So far we have considered problems in which the Lagrangian can be written as L = T V , where T is kinetic energy and V (q , t) is the (velocity-independent) potential energy of the system. It turns out that these same methods can be used to describe more general systems. Consider a particle of charge Q and mass m moving in an electromagnetic eld. Its equation of motion is d mx = QE + Qx B. dt (1.22)

Since the Lorentz force F = Qx B does no work on the particle, it makes no contribution to either T or V and so we cannot derive (1.22) from a Lagrangian of the form L = T V (x). Nevertheless, one can still concoct a Lagrangian that produces this motion. Recall that we can express
E =

A t

and B = A

(1.23)

in terms of an electrostatic potential (x, t) and a magnetic vector potential A(x, t). Here we show that the Lagrangian L= 1 2 + Q(x A ). (1.24) 2 mx produces the equation of motion (1.22).

11

S7 & BT VII: Classical mechanics d (mx + QA) = Q( x A) = 0. (1.25) dt (Notice how the presence of the velocity-dependent force from the magnetic eld means that the generalized momentum p = mx + QA = mx .) The derivative with respect to time in the LHS of (1.25) is to be carried out along the curve x(t). Therefore, using the chain rule, Ai dAi = + dt t dxj A Ai = + (x )A dt xj t .
i

The EL equation for this L is

(1.26)

Substituting this into (1.25) and rearranging, we have that d A mx = Q + Q ( x )A (x A) = 0. dt t (1.27)

The expression inside the rst square bracket is simply E . The expression inside the second is x B . To see this, use the vector identity ( A) +A ( x ), ( x A) = ( x )A + ( A )x +x
0
B

(1.28)

in which the second and fourth terms vanish because x and x are independent variables: x i /xj = 0. Therefore the EL equations (1.25) for the Lagrangian (1.24) reduce to the familiar (1.22). Here I have simply pulled this Lagrangian out of a hat, but when one looks at the problem in a proper, relativistically covariant way the action S [x] =
1 2 2 mx

+ Q(x A ) dt

(1.29)

pops out naturally. As ever, adding a total derivative d = +x dt t (1.30)

to the integrand of S (and thus to the Lagrangian (1.24)) has no eect on the extremal x(t) obtained by solving S = 0. This is equivalent to the gauge transformation , t
A A + .

(1.31)

1.5 Motion in non-inertial co-ordinate systems

Ant on a turntable An ant nds itself on a turntable that rotates with constant angular velocity . The ant sets up cartesian (X, Y, Z ) co-ordinates co-rotating with the turntable, so that the lab co-ordinates (x, y, z ) of a point (X, Y, Z ) on the turntable are given by x cos t sin t 0 X y = sin t cos t 0 Y . (1.32) z 0 0 1 Z The Lagrangian of a particle in the ants co-ordinates 2 + y 2 + z 2) V L=T V = 1 2 m(x 2 2 2 1 2 2 2 =1 2 m(X + Y + Z ) + m(X Y XY ) + 2 m (X + Y ) V,

(1.33)

S7 & BT VII: Classical mechanics the second line following from the rst on dierentiating the transformation matrix (1.32). The (X, Y ) equations of motion for a free particle (V = 0) on the turntable in the ants co-ordinate system are therefore d = 2mY + m2 X, mX dt d + m2 Y. mY = 2mX dt (1.34)

12

Notice that these are simply x =y = 0 in the co-rotating frame. Turning to the ant itself, if friction keeps =Y = 0) with respect to the turntable, then it feels an outward force of magnitude mR2 , it at rest (X where R2 = X 2 + Y 2 (second term on RHS of each of (1.34)). This is reduced if the ant tries to walk against the rotation of the turntable (rst terms on RHS), and vanishes completely if the ant runs around the circle R = constant with speed R. Writing r = (X, Y, Z ) for the co-ordinates of a particle in the rotating frame (r for rotating, x for xed) and (0, 0, ), the Lagrangian (1.33) can be expressed as
1 1 mr 2 + mr ( r ) + 2 m( r)2 V (r, t) L(r, r , t) = 2 2

+ r) V (r, t). =1 2 m (r

(1.35)

Instead of wrestling with the matrix (1.32), a simpler way of deriving (1.35) is to note that a particle moving with respect to the rotating r frame with velocity r has in the x frame a velocity whose magnitude |x | = |r + r |, (1.36)

1 which, together with L = 2 mx 2 V , gives (1.35) directly. Notice that equation (1.36) is a statement only about the magnitude of the vector x , not its direction; we show below that x and (r + r) are related by a rotation, as one might expect.

The equations of motion for the particle in the rotating frame are easy to obtain from (1.35). Making use of the relation a (b c) = c (a b), the partial derivatives of L are found to be
p

L = mr + m r, r V L = mr + m( r) . r r

(1.37)

Therefore the equation of motion d V r 2m r mr = m m ( r ) , dt r (1.38)

showing that in this non-inertial, rotating frame the particle moves as if it were subject to three additional r, the Coriolis force 2m r pseudo-forces: the inertial force of rotation m and the centrifugal force m ( r).

Exercise: Show that in the northern hemisphere the Coriolis force deects every body moving across the earths surface to the right and every falling body towards the East.

Exercise: In cosmology it is often useful to express the equations of motion of dust (stars, gas) in terms of co-moving co-ordinates, r, which are related to physical co-ordinates, x, through x = a(t)r where a(t) is the scale factor of the universe. Show that in these co-ordinates the motion of a dust particle satisies a 1 a + r= 2 . (1.39) r + 2 r a a a r

13

S7 & BT VII: Classical mechanics More general moving co-ordinates There is a more general way of dealing with moving frames. Consider a co-ordinate transformation of the form
x = R + B r,

(1.40)

in which the co-ordinates of a particle P in the xed x system are given in terms of those in the r system by a rotation B (t) followed by a translation R(t). Dierentiating (1.40), the velocity of the particle in the x frame is given by r + Br +B x =R . (1.41)

For example, a rock on the earths equator has co-ordinates r = (R , 0, 0). Equations (1.40) and (1.41) give its co-ordinates and velocities in a frame centred on the sun and oriented with respect to the xed stars if we choose R(t) to be the location of the centre of the earth in this xed frame and use B (t) to describe the rotation of the earth about its axis. = 0 in which the (three-dimensional) x and r coPure rotation Let us rst consider the case R = R ordinate axes are related by a pure rotation, so that x = B r. Since B is a rotation, BB T = I , so B 1 = B T and r = B 1 x = B T x. Substituting this into (1.41) gives T x + Br x = BB . T , dierentiate the relation BB T = I to obtain To understand the eect of BB T + BB T = 0 BB T + (BB T )T = 0. BB (1.43) (1.42)

T is a skew-symmetric matrix. Now write out the expression x = (2 x3 3 x2 , 3 x1 Thus BB 1 x3 , 1 x2 2 x1 ) in matrix form. The result is 0 3 2 x1 3 0 1 x2 . (1.44) 2 1 0 x3 So, by choosing appropriately, any skew-symmetric matrix can be represented by the operation x. In particular, the relation (1.42) can be written as
x = x + Br

(1.45)

T with eigenvalue 0. This is the for some (possibly time-dependent) , which is an eigenvector of BB instantaneous angular velocity of the r frame with respect to the x frame. Using x = B r and introducing B 1 , the instantaneous angular velocity in the r frame, we have that
x = B Br + Br

= B ( r + r ),

(1.46)

1 mx 2 V = 1 T x ) V gives the because B b B c = B (b c). Substituting this x into L = 2 2 m(x Lagrangian (1.35). So, the Lagrangian we derived earlier for the special case of a steady rotation t about the x 3 = r 3 axis holds even when the rotation axis and rotation rate change with time, provided we take to be the instantaneous angular velocity in the rotating frame. T for each of the following matrices and nd by comparing your results with Exercise: Calculate BB equation (1.44). What is in each case? cos t sin t 0 cos t sin t 0 B1 = sin t cos t 0 , B2 = 0 0 1 . (1.47) sin t cos t 0 0 0 1

S7 & BT VII: Classical mechanics

14

Pure translation, no rotation B = I and equation (1.41) becomes

If the r and x co-ordinates are related by a pure translation, then

2

+r x 2 = R

2 + 2R r =R +r 2 2 + 2 d (R r) 2R r+r =R 2. dt

(1.48)

A suitable Lagrangian is r V (r, t), L= 1 2 mR 2 mr dropping the rst two terms from (1.48) because they contribute nothing to the equations of motion. General case translation plus rotation For the general case, we introduce an intermediate co-ordinate system x related to x by a translation and to r by a rotation:
x=R+x, x = B r.

(1.49)

(1.50)

x V . Taking x 2 mR = B (r + r) from (1.46), we Using (1.49), the Lagrangian L(x , x , t) = 1 2x have nally that 2 (B r) V. L(r, r , t) = 1 + r) mR (1.51) 2 m (r = V / x. Show Exercise: Let B be a constant (time-independent) rotation matrix and choose mR d that in this freely falling frame the equations of motion become dt mr = 0. Exercise: Show for the case r = 0 that T (x R) = BB x R = (x R). Explain why this means that and are independent of the choice of R. (1.52)

1.6 Noethers theorem

A constant of motion is any function C (q , q , t) for which the total time derivative dC C C C = +q +q dt t q q (1.53)

vanishes along a trajectory q (t) that satises the equations of motion. For example, if L/t = 0 then we already know from the Beltrami identity (0.19) of 0.1 that H (q , q ) = q L L q (1.54)

is a constant of motion. Similarly, if L contains a cyclic co-ordinate qi (one for which L/qi = 0), then the generalized momentum pi = L/ q i is a constant of the motion. In general, a system with n degrees of freedom has 2n 1 independent constants of motion. To see this, suppose that a system has (q , q ) at some time t. Then one can in principle integrate the system forwards/backwards to some reference time, t0 . The values of q and q at t0 are some complicated functions qi (t0 ) = fi (q , q , t), q i (t0 ) = gi (q , q , t), of their values at time t. Eliminating t from these 2n equations leaves 2n 1 constants of motion. There are few mechanical systems for which one can write down expressions for all 2n 1 constants of motion, but we have already seen (e.g., motion of particle in central eld) that nding

15

S7 & BT VII: Classical mechanics just n constants of motion is enough to understand the behaviour of a mechanical system with n degrees of freedom, at least qualitatively. Noethers theorem states that for every continuous symmetry of the Lagrangian there is a corresponding conserved quantity. Suppose we apply a transformation to our mechanical system, which results in a small change in co-ordinates q q + K (q ), (1.55) where K (q ) is a vector-valued function of q . For example, we might move our favourite pendulum slightly to the left, or turn it anticlockwise a little. If the Lagrangian L(q , q , t) is invariant under this transformation then there is a constant of motion L K. (1.56) C (q , q ) = q Proof: Since the transformation (1.55) leaves L unchanged then, at 0= dL L q L q = + d q q L L = K+ K. q q = 0,

(1.57)

Using the EL equation to replace the L/ q factor, 0= Example: homogeneity of space L= d dt L q K+ L d K = q dt L K . q (1.58)

The Lagrangian for a closed system of N particles,

1 2 i

mi x 2 i

ij

V (|xi xj |),

(1.59)

is invariant if we apply the translation xi xi + n to all the particles co-ordinates, for any choice of direction n . By Noethers theorem, this symmetry means that L x i n = mi x i
i

n ,
i

(1.60) mi x i is an invariant. Thus,

is a constant of the motion. Since the relation holds for any n , we have that translation invariance of L implies conservation of total linear momentum.

Example: isotropy of space Similarly, the Lagrangian (1.59) is invariant if we pick any direction n and carry out an innitesmal rotation of the system about this axis: xi xi + n xi . Noether tells us that there is a conserved quantity L x i ( n xi ) =
xi mi x i

n .

(1.61)

In other words, rotational invariance of L leads to conservation of angular momentum. A particle moves in a uniform magnetic eld + Q(x A ). Since E = 0, we are free to choose = 0. To nd the constants of motion it proves easiest to consider two dierent choices for the vector potential A, each of which lead to the same B and therefore to the same equations of motion.
B = (0, 0, B ) = B k. From (1.24), the Lagrangian L =
1 2 2 mx

Example: particle in a uniform magnetic eld

Our rst choice is A = (By, 0, 0). Then we get L = 1 2 QxBy . This is invariant under translations in 2 mx either the i or k directions: x x + i, x x + k. Therefore, two constants of the motion are L i = px = mx QBy x and L k = pz = mz. x (1.62)

S7 & BT VII: Classical mechanics Our second choice is A = (0, Bx, 0). Then L = 1 2 QyBx , which is invariant under translations in either 2 mx j or k directions, leading to the additional constant of motion L j = py = my + QBx. x The physical meaning of pz is obvious. To understand px and py , consider + iQB, P px + ipy = m(x + iy ) + QB (ix y ) = m (1.64) (1.63)

16

where x + iy . This is a rst-order ODE for . Multiplying by the integrating factor eit , where the Larmor frequency QB/m, the solution is (t) = P + Keit , im (1.65)

where K is a constant of integration. We now see that px and py (through P ) encode the x and y co-ordinates of the guiding centre around which the particle gyrates. The radius of gyration is given by the integration constant |K |, which sets the particles energy.

17

S7 & BT VII: Classical mechanics

2 Hamiltonian mechanics

2.1 Hamiltons equations

The EulerLagrange equation (1.8), d dt L q L = 0, q (2.1)

when written out in component form becomes a set of n coupled second-order ODEs. Like any set of n coupled second-order ODEs, we can turn it into a set of 2n rst-order ODEs by introducing n additional variables. In this case, introduce p L/ q to obtain
p =

L , q

p=

L , q

(2.2)

the second of which is an awkward implicit equation for q . Wed like to have a new function that somehow encodes the same information as L(q , q , t), but with q replaced by p L/ q . Provided L(q , q , t) is a convex function of the velocities q , then we can do just this by taking the Legendre transform (0.3) of L(q , q , t) with respect to q . This gives a new function, the Hamiltonian, H (q , p, t) p q L(q , q , t), (2.3) which is a function of the generalized co-ordinates q , the conjugate momenta p and time; we have to use the relation p = L/ q to express all q on the RHS in terms of p and q (and possibly t). To obtain the equations of motion in terms of this new function, take the total dierential of each side of (2.3). The RHS gives dH = q dp + p dq L L L dq + dq + dt q q t L L dq dt, =q dp q t

(2.4)

using p = L/ q to cancel two of the terms on the rst line. This must equal the total dierential of the LHS, H H H dH = dq + dp + dt. (2.5) q p t Since (2.4) and (2.5) have to be equal for any choice of (dq , dp, dt), it follows that
q =

H ; q

L H = ; q q

L H = . t t

(2.6)

Using the relation p = L/ q , these become Hamiltons equations:

q =

H , p

p =

H . q

(2.7)

S7 & BT VII: Classical mechanics

18

2.2 Why bother?

In Hamiltonian mechanics we can think of our mechanical system as a point (q , p) moving in 2n-dimensional phase space with velocity given by (2.7). Contrast this to Lagrangian mechanics in which there is no such simple geometrical interpretation of the correpsonding paths through conguration space, even when the equations are written out in the coupled form (2.2). In practice it usually turns out that Hamiltons equations are no easier to solve than the corresponding EL equations. The power of Hamiltonian mechanics comes from the ease with which one can use the explicit ODEs (2.7) to discover general properties of trajectories in phase space. This helps uncover much of the hidden structure that underpins classical mechanics. It turns out that very similar structures underlie quantum mechanics.

2.3 Examples
2 + mgl cos , so that p L/ = ml2 . ) = 1 ml2 Simple pendulum The Lagrangian (1.9) L(, 2 Using the Legendre transform (2.3) to turn this L into a function of (, p ) gives L H (, p ) = p = (2.8)

p2 mgl cos , 2ml2

in terms of p . Hamiltons in which we have used the expression for p to express all occurrences of equations (2.7) become H = H = p , p = = mgl sin . (2.9) 2 p ml Notice that there is essentially no dierence between these and the corresponding EL equation. The Hamiltonian approach does, however, encourage us to think of the motion of the pendulum as taking place in a two-dimensional phase plane (, p ), in which the velocity eld is given p by (, ) = (H/p , H/) (red arrows on plot). The phase-space co-ordinates of the bob follow the integral curves of this velocity eld (blue curves), so called because they are obtained by integrating (i.e., solving) Hamiltons equations to nd the trajectory.
2

1.0

0.5

0.0

0.5

1.0

Particle in a potential well If the particles potential energy V = V (x, t) then the Lagrangian 1 L= 2 mx 2 V , giving p = L/ x = mx , the usual momentum familiar from Netownian mechanics. The Legendre transform (2.3) of L is H (x, p, t) = p x L(x, x , t) . (2.10) p2 = + V (x, t) 2m Hamiltons equations (2.7) reduce to the very familiar
x =

H p = , p m

p =

H V = . x x

(2.11)

This example serves as a useful reminder of which of Hamiltons equations has the minus sign. 2 + r2 sin2 2 ) V (r), which gives Particle in a central eld The Lagrangian (1.18) L = 1 2 + r2 2 m(r 2 and p = mr2 sin . Applying the Legendre transform (2.3) to turn this momenta pr = mr , p = mr2 L(r, r, , ) into something that depends explicitly on (r, pr , p , p ) gives the Hamiltonian + p L(r, r, ) H (r, pr , p , p ) = pr r + p , =

p2 p2 p2 r + + + V (r). 2m 2mr2 2mr2 sin2

(2.12)

19

S7 & BT VII: Classical mechanics Hamiltons equations (2.7) become pr , m p2 dV p2 + 3 2 p r = 3 , r dr r sin r = = p , mr2 p2 cos p = , mr2 sin3 p , mr2 sin2 p = 0. =

(2.13)

Just as in the Lagrangian case (1.3), if we orient our co-ordinate system so that the particle starts with = 2 and = 0 then Hamiltons equations tell us that p and remain zero throughout the motion and p is a constant of motion. This is equivalent to motion in the simpler eective Hamiltonian He (r, pr ) = p2 r /2m + Ve (r ), where the one-dimensional eective potential Ve (r) = p2 + V (r ). 2mr2 (2.14)

Notice that it is much easier to exploit the conservation of the momenta p and p in Hamiltonian mechanics than in Lagrangian mechanics. Motion of a particle referred to a rotating co-ordinate system The Lagrangian (1.35) L = 1 2 m [ r + r ] V ( r , t ) from which p L/ r = m ( r + r ). Using (2.3) to construct the corresponding 2 Hamiltonian gives H (r, p) = p r L
1 =1 2 2 m( r)2 + V 2 mr 1 = m[r + r] r 2 m[r + r ]2 + V

(2.15)

p2

2m

p ( r) + V.

Exercise: Write out Hamiltons equations for this system and show that they are equivalent to equation (1.38). Particle in an electromagnetic eld Similarly, for the Lagrangian (1.24), L = 1 2 + Q(x A ) , 2 mx the momenta p = L/ x = mx + QA. The Hamiltonian H =px L = mx 2 + QA x =
1 2 2 mx 1 2 2 mx

+ Q

+ Q(x A ) (2.16)

(p QA)2 = + Q. 2m Exercise: By writing the second of Hamiltons equations for this case in the form p i = 1 2m xi (pj QAj )(pj QAj ) Q xi

(2.17)

show that p = Q(x A) Q (remember x and p are independent co-ordinates in phase space). Hence show that Hamiltons equations reduce to the expected mx = Qx ( A) Q.

2.4 General remarks

1. If a co-ordinate qi doesnt appear in the Lagrangian, then, by construction, it doesnt appear in the Hamiltonian either. The corresponding momentum pi is conserved because, from Hamiltons equations, H p i = = 0. (2.18) qi

S7 & BT VII: Classical mechanics 2. The rate of change of H along a trajectory (q (t), p(t)), dH H H H = +q +p dt t q p H H H H H H + = , = t p q q p t

20

(2.19)

3. If L = T V , where T is a homogeneous quadratic form in the velocities q , T =

1 2 ij

the second line following from the rst on using Hamiltons equations (2.7). Thus H is conserved if it does not depend explicitly on time. (We have already seen this from the Beltrami identity in 0.1.) aij (q , t)q i q j , (2.20)

with aij = aji and V = V (q , t), then the Hamiltonian H = T + V . To see this, notice that pk L = q k
1 2 ij

(aij ik q j + aij q i jk ) =
i

aki q i .

(2.21)

Constructing the Hamiltonian in the usual way, we have that H =pq L=

k i

aki q i q k 1 2

ij

aij q i q j V = T + V.

(2.22)

2.5 Liouvilles theorem

The instantaneous state of a mechanical system is described by a point in phase space with co-ordinates (q , p). This point moves through phase space with a velocity (q , p ) given by Hamiltons equations (2.7). If we release an ensemble of systems with the same Hamiltonian but slightly dierent initial conditions, the phase points ow through phase space like a uid. Before tackling this particular ow, we rst need a standard result: any ow in which the divergence of the velocity eld is identically zero preserves volume (i.e., is incompressible). To show this, let us take a general velocity eld x = f (x, t) (2.23) in an n-dimensional space and examine the relative motion of two nearby points x0 (t) and x1 (t). Let (t) x1 (t) x0 (t). From a Taylor expansion of (2.23) we have that after time t, dropping terms O (t2 ), j j + fj k t = xk jk + fj t k = Jjk k , xk (2.24)

where Jjk = jk + (fj /xk )t. The change in volume eected by this transformation is given by det Jjk . Exercise: Show by direct expansion that det(I + At) = 1 + (tr A)t + O(t2 ) for any square matrix A. Using the result of the exercise, the ow preserves volume if zero divergence.
i (fi /xi )

= 0, i.e., if the velocity eld has

Now we return to phase space. Let us introduce the shorthand w (q , p) = (q1 , . . . , qn , p1 , . . . , pn ) for the co-ordinates of a point in 2n-dimensional phase space. By Hamiltons equations, the velocity eld in phase space is w = (H/ p, H/ q ). Its divergence div (w (w, t)) 2H 2H (w (w, t)) = = 0. w q p p q (2.25)

21

S7 & BT VII: Classical mechanics So, the phase-space volume enclosed by the points representing our ensemble of systems is conserved. This is Liouvilles theorem. Similarly we can introduce the phase-space mass density f (x, v , t), which has to respect the (phase-space) continuity equation 0= f f + div (f w ) = + (f q ) + (f p ) t t q p f f q p f + q + p +f + = t q p q p div w =0 f f H f H = + . t q p p q This is known as Liouvilles equation.

(2.26)

2.6 Poisson brackets

The Poisson bracket [A, B ] of any two smooth phase-space functions A(q , p, t), B (q , p, t) is dened as [A, B ] A B A B . q p p q (2.27)

It is straightforward to show that the Poisson bracket has the following properties (verify them!): (i) [A, B ] = [B, A] (antisymmetry); (ii) [A + B, C ] = [A, C ] + [B, C ] for any real numbers , (linearity); (iii) [AB, C ] = [A, C ]B + A[B, C ] (chain rule); (iv) [[A, B ], C ] + [[B, C ], A] + [[C, A], B ] = 0 (Jacobi identity). Furthermore, phase-space co-ordinates satisfy the canonical commutation relations or fundamental Poisson bracket relations [pi , pj ] = 0, [qi , qj ] = 0 and [qi , pj ] = ij . (2.28)

This follow directly from the denition (2.27) and, remembering that (q , p) are independent coordinates in phase space, the relations pi /qj = qi /pj = 0 and pi /pj = qi /qj = ij . In terms of Poisson brackets, Hamiltons equations become q i = [qi , H ], p i = [pi , H ]. (2.29)

The rate of change of any function f (q , p, t) along a trajectory (q (t), p(t)) is f f f df = + q + p dt t q p f f H f H = + t q p p q f = + [f, H ]. t

(2.30)

Alternative notation It is sometimes convenient to combine q and p into the single vector w (q , p) = (q1 , . . . , qn , p1 , . . . , pn ). Then Hamiltons equations (2.7) can be written as
w =J

H , w

(2.31)

S7 & BT VII: Classical mechanics where the symplectic matrix J 0n In In 0n , (2.32)

22

and 0n and In are the n n zero and identity matrices, respectively. The expression (2.27) for the Poisson bracket becomes 2n T A B B A [A, B ] = J = J . (2.33) w w w w
, =1

Because wi /w = i the canonical commutation relations (2.28) are simply [wi , wj ] = Jij . (2.34)

2.7 Symmetries and conservation laws

Constants of motion If a function F (q , p) is a constant of the motion then, using (2.30), = dF = F + [F, H ] 0=F dt t [F, H ] = 0. (2.35)

The functions F and H are then said to (Poisson) commute. Conversely, if we can nd a function F (q , p) for which [F, H ] = 0, then F is a constant of motion. Given two constants of motion, F (q , p) and G(q , p), we have from the Jacobi identity that [[F, G], H ] + [[G, H ], F ] + [[H, F ], G] = 0.
0 0

(2.36)

So, [F, G] is also a constant of motion. In some cases the new function [F, G] will turn out to be trivial (e.g., it might be zero or a straightforward function of the known invariants F and G), but sometimes it will be a new, independent constant of motion. Example: Let r = (r1 , r2 , r3 ) be Cartesian co-ordinates and p the corresponding conjugate momentum. The angular momentum J = r p has components J1 = r2 p3 r3 p2 , The Poisson bracket of the rst two, [J1 , J2 ] = [r2 p3 r3 p2 , r3 p1 r1 p3 ]
0

J2 = r3 p1 r1 p3 ,

J3 = r1 p1 r2 p1 .

(2.37)

= [r2 p3 , r3 p1 ] [r2 p3 , r1 p3 ] [r3 p2 , r3 p1 ] +[r3 p2 , r1 p3 ]

0

(2.38)

= [r2 p3 , r3 p1 ] + [r3 p2 , r1 p3 ] = r2 p1 + p2 r1 = J3 . If J1 and J2 are constants of motion, then so too is J3 . Or, more generally, the vector J is conserved if any two of its components are. Exercise: Show that [J 2 , Ji ] = 0. Symmetries and conservation laws In section 1.6 we saw that if a Lagrangian L is invariant under a small change in co-ordinates, q q + K (q ), then C = K (L/ q ) is a constant of motion. There is a corresponding relationship between symmetry and conservation laws in Hamiltonian mechanics.

23

S7 & BT VII: Classical mechanics We have seen how the Hamiltonian function denes a phase ow (q , p ) = ([q , H ], [p, H ]) = (H/ p, H/ q ). Similarly, any function G(q , p) denes a ow through phase space with dq () = [q (), G], d dp() = [p(), G], d

(2.39)

in which the parameter takes the place of time. If we integrate the coupled ODEs (2.39) for a xed interval in , we obtain a one-parameter mapping G : (q (0), p(0)) (q (), p()) of phase space onto itself. The function G is the generator of this mapping. The solutions to (2.39) are the integral curves of G. Exercise: Suppose that (x, p) are Cartesian co-ordinates in phase space. What mapping is generated by G = x1 ? By G = p1 ? By G = x1 p2 x2 p1 ? F dq F dp dF = + = [F, G]. d q d p d For small , G is the innitesmal map
q q+

Exercise: Show that the derivative of a function F (q , p) along the ow generated by G is

(2.40)

G , p

pp

G , q

(2.41)

or q = G/ p, p = G/ q . If H is invariant under this map, then G is called a symmetry of the Hamiltonian. The condition for this is that 0 = H = H H q + p q p H G H G = q p p q = [H, G].

(2.42)

= [G, H ], so if G is a symmetry then G is conserved. Conversely, for any constant of motion G(q , p) But G then there is a map (2.41) that leaves H invariant. (In constrast, in Lagrangian mechanics Noethers theorem says only that symmetryconstant; it does not show that constantsymmetry too.)

Exercise: What mapping does H generate? What condition do we need to impose on H for this to work? Exercise: Show that the innitesmal map (2.41) can be written as
w w + J
G w G T J w G T J w

=w = I where w =
q p

(2.43)
w

w,

and J is the symplectic matrix (2.32). By chaining many such mappings together,

obtain a formal expression for the mapping G for general using the denition exp(X ) = limn (I + X/n)n for the exponential of a linear operator.

S7 & BT VII: Classical mechanics

24

2.8 Poincar e integral invariants

The following two sections provide another way of looking at the structure of phase space in which time plays a more explicit role: you may have noticed that many of the results concerning Poisson brackets required us to restrict our attention to functions F (q , p) that do not depend explicitly on time. Some hydrodynamics: Stokes lemma Suppose we have a uid in which points at position x move with velocity x = u, for some u = u(x). At time t0 we set up a closed loop 0 of dye-releasing particles. At any later instant the particles will lie along another closed loop (t). Between times t0 and t the particles will have traced out a surface S , the ends of which are the curves 0 and (t). The surface S is known as a vortex tube: individual particles within the loop move along integral curves of u, which are vortex lines of u. Because the particles always move parallel to u we have that S ( u) n dS = 0, where n is the outward-pointing unit normal to the vortex tube. Stokes theorem tells us that
u dx u dx =

( u) n dS = 0,

(2.44)

where is any other loop that encircles the vortex tube. Extended phase space Given a mechanical system with n degrees of freedom the corresponding phase space has 2n dimensions. If we treat time t as an additional co-ordinate we obtain a (2n + 1)-dimensional space, known as extended phase space.
1 2 (p + q 2 ). Show that Exercise: A one-dimensional simple harmonic oscillator has Hamiltonian H = 2 solutions to Hamiltons equations yield helical curves in extended phase space. 2 2 2 Exercise: Now suppose that H = 1 2 (p + q ), where = (t). How does the time dependence of the spring constant change the trajectories in extended phase space?

Vortex tubes in 3d extended phase space Consider a system with one degree of freedom so that extended phase space has three dimensions. We recycle our use of w from the previous section to label points , where (p ) in (extended) phase space and look at one-dimensional loops w( ) = p( )p + q ( ) q + t( )t , q , t are the natural basis vectors to use for extended phase space. We assume that (p , q , t) (in that order) form a right-handed set. . The vortex lines of u are given by the solutions w( ) of We introduce a special vector u = pq Ht dp dq dt , , d d d dw = =u= d
p
p

q
q

t H
t

H H , ,1 , q p

(2.45)

using p/t = 0 because (p, q, t) are independent co-ordinates of extended phase space. The vortex lines of the special vector u are the solutions to Hamiltons equations of motion! Any one-dimensional loop 0 we draw in extended phase space will form a vortex tube, the vortex lines of which are the integral curves of Hamiltons equations. The circulation about 0 is
u dw =

p
0

dq dt H d d

d =
0

(p dq H dt) .

(2.46)

Stokes lemma tells us that

1

(p dq H dt) =

(p dq H dt)

(2.47)

for any other loop 1 that encircles the same vortex tube. Comments: is not unique: for any S (q, p, t) the vortex lines of u + S are the 1. The special vector u = pq Ht same as those of u; we can add a total derivative dS to either integrand in (2.47).

25

S7 & BT VII: Classical mechanics 2. If we choose 0 and 1 to lie on constant-t slices, t = t0 and t = t1 respectively, of extended phase space, then dt = 0 and (2.47) becomes p dq =
t0 t1

p dq

dp dq =
S (t0 ) S (t1 )

dp dq.

(2.48)

That is, the area S (t) of the phase plane enclosed by a loop does not change as the loop evolves: we have rederived Liouvilles theorem for the simplest case, n = 1, in a much more complicated way... The gure on the right shows why p dq = S dp dq , where S is the surface enclosed by the loop : both terms on the LHS of this graphical equation contribute to p dq , but dq is negative in the rst term and positive in the second.
p

+
q

General case It turns out that this result can be generalized to systems with more than one degree of n , the vortex lines of which again are precisely i H t freedom. The special vector becomes u = i=1 pi q the integral curves of Hamiltons equations: every vortex line of u is a solution to Hamiltons equations and every solution to Hamiltons equations is a vortex line of u. A generalization of Stokes lemma to (2n + 1)-dimensional space tells us that (p dq H dt) = (p dq H dt) (2.49)

for any one-dimensional loops 0 and 1 that encircle the same two-dimensional vortex tube. The quantity (2.49) is known as a Poincar e (Cartan) integral invariant. Comments 1. For later use we note that, for any function S (p, q , t), S S S dp + dq + dt p q t = 0. (2.50)

This means we can add a total derivative of S to either, or both, sides of (2.49). 2. If we choose 0 and 1 to be constant-t slices, t = t0 and t = t1 , of extended phase space then (2.49) becomes p dq = p dq
0 1

pi dqi =
i 0 i 1

pi dqi dpi dqi ,

i Si (1 )

(2.51)

dpi dqi =
i Si (0 )

where Si ( ) is the projection of the loop onto the (pi , qi ) plane. If we set up a loop of particles at some time t0 , the sum of the projected areas of the loop onto the (pi , qi ) planes is preserved as the loop evolves. Volume preservation (Liouvilles theorem) can be viewed as a consequence of these more fundamental integral invariants.

S7 & BT VII: Classical mechanics

26

2.9 Canonical maps

We have seen how easy it is to change variables q Q in the Lagrangian formulation of mechanics. The price we pay for the more interesting and powerful structure of phase space in Hamiltonian mechanics is that co-ordinate transformations are not so straightforward. In this section we investigate how to change to new phase-space co-ordinates (Q, P ), Qi = Qi (q , p, t), Pi = Pi (q , p, t), (i = 1, . . . , n), (2.52)

that preserve the Poincar e invariants of the previous section. First, some denitions: If for any loop in extended phase space we have that (P dQ K dt) = (p dq H dt) (2.53)

in which the function K (Q, P , t) is independent of the choice of , then the transformation (2.52) is called a canonical map (or a canonical transformation) and the new co-ordinates (P , Q) are called canonical co-ordinates. Evolution under time is an example of a canonical map. To see this, suppose that we dene (Q, P ) to be the values that (q , p) will have one second in the future. Then (2.53) is clearly satised if we take K (Q, P , t) = H (Q, P , t + 1 sec). Hamiltons equations in the new co-ords The RHS of (2.53) is u dw, where w = (p( ), q ( ), t( )) and u = i pi q i H t. We have already seen that the vortex lines for this u are given by
q i =

H , pi

p i =

H . qi
i

(2.54) i K t . The vortex Pi Q (2.55)

Similarly, the LHS of (2.53) is lines of U are clearly given by

U dW , where W = (P ( ), Q( ), t( )) and U =

= Q

K , P

= P

K . Q

Both (2.54) and (2.55) describe the same vortex lines in the same extended phase space, but expressed in dierent co-ordinates. Therefore Hamiltons equations in the new co-ordinates are given by (2.55) with Hamiltonian K (Q, P , t). Generating functions Equation (2.53) can hold for all loops only if the integrands dier by a total derivative dS of any well-behaved function S (P , Q, t):
P dQ K dt + dS = p dq H dt.

(2.56)

A powerful way of constructing canonical maps is by playing with the function S . Let us assume that we can express P = P (q , Q, t) so that we can eliminate P from S to obtain S = F1 (q , Q, t), a function of both the old and new co-ordinates and time, but not the momenta. Substituting this S = F1 into (2.56) and using the chain rule gives
P dQ K dt +

F1 F1 F1 dq + dQ + dt = p dq H dt. q Q t

(2.57)

As (dq , dQ, dt) can be varied independently (the equality above has to hold for any loop ) we must have
p=

F1 , q

P =

F1 , Q

K=H+

F1 . t

(2.58)

NB: Some books dene a canonical map as one that preserves the form of Hamiltons equations. Our condition (2.53) is more stringent.

27

S7 & BT VII: Classical mechanics where H in the last equation is to be interpreted as the original H (q , p, t) substituting for q = q (Q, P , t), p = p(Q, P , t) to make it a function of (Q, P , t). Thus the function F1 (q , Q, t) generates an implicit transformation from (q , p) (Q, P ). By construction, it satises the condition (2.53) and therefore is canonical.

2 2 2 Exercise: What mapping is generated by F1 = q Q? Show that the Hamiltonian H (q , p) = 1 2 (p + q ) 1 2 2 2 is transformed to K (Q, P ) = 2 (Q + P ).

Unfortunately, generating functions of the form F1 (q , Q, t) are not suitable for constructing mappings close to the identity. So, instead of writing S = F1 (q , Q, t), let us take S = P Q + F2 (q , P , t), (2.59)

in which we treat Q as a function Q(q , P , t). Substituting this into (2.56) and using the chain rule to expand dS gives:
P dQ K dt P dQ Q dP +

F2 F2 F2 dq + dP + dt = p dq H dt. q P t

(2.60)

One way of explaining the P Q term that appears in (2.59) that it comes from taking the Legendre transform of F1 . An alternative, simpler approach is to note that (a) we are free to choose (almost) whatever we like for S and (b) including P Q in S nicely cancels out the P dQ on the LHS of (2.56). As (dP , dq , dt) vary independently, we must have that
p=

F2 , q

Q=

F2 , P

K=H+

F2 . t

(2.61)

This is another implicit canonical mapping between (q , p, t) and (Q, P , t). Exercise: Show that F2 (q , P ) = q P generates the identity map. What mapping does F2 (q , P ) = qP + n q produce? What about F2 = q P + n P? Exercise: Show that for small the generating function F2 (q , P ) = q P + G(q , P ) produces the innitesmal map (2.41). (Use the fact that P p as 0.)

Is a given mapping canonical? A simple way of testing whether a mapping is canonical is by examining P dQ p dq . If that can be expressed as a total derivative dS (p, q , t) or dS (P , Q, t) then the mapping is canonical. Exercise: Show that the mapping Q = log p, P = pq is canonical. Find a function F1 (q, Q) that generates this mapping. Find another generating function of the form F2 (q, P ). Another test is to return to the denition (2.53) of a canonical map and to check whether dPi dQi =
i Si ( ) i si ( )

dpi dqi

(2.62)

for all loops , where si ( ) and Si ( ) are the projections of onto the (pi , qi ) and (Pi , Qi ) planes. Let us look at the projections of the (pk , qk ) planes onto all of the (Qi , Qj ), (Pi , Pj ) and (Pi , Qj ) planes. We have that Qi Qj Qj Qi dpk dqk , dQi dQj = qk pk qk pk
k

dPi dQj =
k

Pi Qj Qj Pi qk pk qk pk Pi Pj Pj Pi qk pk qk pk

dpk dqk , dpk dqk ,

(2.63)

dPi dPj =
k

S7 & BT VII: Classical mechanics where the quantities in parentheses are the Jacobians of the transformation from (qk , pk ) to the new (Qi , Qj ) etc coordinates. The only way of making (2.62) hold for all choices of the loop is by requiring that Qi Qj Qj Qi qk pk qk pk Qj Pi Pi Qj qk pk qk pk Qi Qj Qj Qi qk pk qk pk = 0, = ij , = 0. (2.64)

28

That is, for a map to be canonical, the new coords (Q, P ) must themselves satisfy the canonical commutation relations (a.k.a. fundamental Poisson bracket relations) [Qi , Qj ] = [Pi , Pj ] = 0, [Qi , Pj ] = ij , (2.65)

in which the Poisson brackets are understood to be evaluated with respect to the old (q , p) coordinates, as in (2.64). Equation (2.65) is a necessary and sucient condition for (2.62) to be true: a map (q , p) (Q, P ) is canonical if and only if the new coordinates (Q, P ) satisfy the canonical commutation relations (2.65). Invariance of Poisson brackets under canonical maps We can use the condition (2.65) to show that all Poisson brackets are invariant under canonical maps. To simplify notation, we introduce
w= q p

and W =

Q P

(2.66)

in terms of which the relations (2.65) become simply [Wi , Wj ] = Jij , where Jij are the elements of the symplectic matrix (2.32). Then, using expression (2.33) for the Poisson bracket of the functions A(w, t), B (w, t), we have that 2n 2n 2n B A B A Wi W j J = J [A, B ]w = w w W w W w i j j =1 , =1 i=1 , =1 2n 2n 2n A Wi Wj B A B = J = [Wi , Wj ] W w w W W W i j i j i,j =1 i,j =1
2n , =1

(2.67)

A B Jij = [A, B ]W . W W i j i,j =1

2n

So, canonical maps preserve all Poisson brackets. Exercise: We have been cavalier about the choice of signs in the Jacobians in equation (2.63) above. Here is how to show that the signs in that expression are correct. Any pair of n-dimensional vectors (a, b) denes a parallelogram in n-dimensional space. We dene the oriented area of the projection of this parallelogram onto the (xi , xj ) plane to be (dxi dxj )(a, b) = ai bj aj bi . Show that (dxi dxj )(b, a) = (dxi dxj )(a, b) = (dxj dxi )(b, a). Given new coordinates X = X (x), the projection of the (a, b) parallelogram onto the (Xi , Xj ) plane is (dXi dXj )(a, b) = Xi dxk xk Xj dxl xl (a, b) =
kl i (dpi

Xi Xj (dxk dxl )(a, b). xk xl

(2.68)

Hence show that the condition i (dPi dQi )(a, b) = Now read 1216, 1820, 3248 of Arnold.

dqi )(a, b) for all (a, b) implies (2.64).

29

S7 & BT VII: Classical mechanics

Attic A Rigid bodies

A.1 Constraints
Sometimes it is convenient to have some redundancy among the co-ordinates. For example, a circular hoop of mass m rolls without slipping down a rigid wire inclined at an angle to the horizontal. The obvious co-ordinates to describe the system are the distance x of the hoop from its starting point and the angle between the hoops point of contact because with the wire and a reference point P on its rim. But x = R the hoop rolls without slipping, so that x = R + constant, (A.1)

the constant depending on the initial conditions. So, x and are not independent co-ordinates. A holonomic constraint is a relation g (q ; t) = 0, (A.2) where g is a function of the n co-ordinates (q1 , . . . , qn ) and possibly time t. Equation (A.1) is an example of such a constraint. Following 0.2, the procedure for nding the equations of motion of a system with k independent holonomic constraints is to introduce a new generalized co-ordinate i for each constraint gi (q ; t) = 0. Treating these n + k co-ordinates as independent, consider motion in the (n + k )-dimensional augmented conguration space with Lagrangian
k

L ({q , }, q , t) L(q , q , t) +

i gi (q , t).
i=1

(A.3)

Writing down the q and i components of the EL equation for L , the equations of motion are d dt L q L = q
k

i
i=1

gi , q

(A.4)

g1 = = gk = 0. Each constraint gi results in an additional generalized force on the RHS of the EL equation, the size of which is controlled by the co-ordinate i . 1 2 2 mr 2 + 1 Exercise: Consider a system with co-ordinates (r, ), Lagrangian L = 2 2 mr + mgr cos and 2 + mg cos |. More generally, constraint r l = 0. Show that the constraint force has magnitude |mr show that if the co-ordinate q1 is held xed when one solves the EL equations, then the magnitude of the corresponding (generalized) constraint force is given by |L/q1 |.

Returning to our example of a hoop on a wire, the hoops kinetic energy T can be broken down into the energy due to translational motion of its centre of mass, 1 2 , and the rotational energy about the centre 2 mx 1 2 2 of mass, 2 mR . The potential energy V = mgx sin . There is one constraint, g (x, ) = x R = 0. The augmented Lagrangian
2 2 1 L =2 mx 2 + 1 2 mR + mgx sin + (x R),

(A.5)

(A.6)

S7 & BT VII: Classical mechanics for which the EL equations are d mx mg sin = , dt d = R, mR2 dt = 0. x R from the second, we have that Using the third equation to eliminate the rst equation and eliminating , gives 1 g sin . x = 2
d dt mx

30

(A.7)

= . Substituting this into (A.8)

So the hoop rolls down the plane with only half the acceleration it would have in the frictionless case. Not all constraints can be written as g (x, t) = 0. An example of a system with such a non-holonomic constraint is a hoop rolling without slipping down a plane instead of along a wire. Natural co-ordinates to use are the location (x, y ) of the hoops centre, the orientation of a reference a point on the rim of the hoop and another angle giving the orientation of the hoop in the plane. The no-slip conditions mean that the hoops velocity satises sin , cos , x = R y = R (A.9) but these cannot be integrated to give an expression of the form g (x, y, , , t) = 0. To see this, think of rolling the hoop on closed circuits of dierent lengths around the plane, returning to the starting position (x, y ) with the same . The angle at the end depends on the length of the circuit.

A.2 Lagrangian mechanics of rigid bodies

A rigid body is a system of particles having masses mi and positions xi satisfying constraints of the form |xi xj | = rij for all pairs (i, j ) of particles, where each rij is a constant. Exercise: A rigid body moves in an external gravitational potential (x). Show that extremizing the 1 action integral (1.3) with Lagrangian L = T V , where T = 2 i 2 and V = i mi (xi ), subject i mi x 2 2 to the constraints that (xi xj ) = rij , leads to the usual Newtonian equations of motion for a rigid body: d mk xk = mi + constraint forces , (A.10) dt xk
j =k

Exercise: Use (A.10) to show that if there are no external forces acting on the body, then its linear momentum i mi x i and angular momentum i xi mi x i are conserved. See 1.6 later for a more elegant way of obtaining this result. The conguration space of a rigid body is six dimensional. In case this is not obvious, pick any three noncollinear points x1 , x2 and x3 in the body. We need three numbers to specify x1 , another two for x2 (we already know r21 ) and a nal one to x x3 . The positions xi of all the other points in the body are then completely determined by the constraints |xi x1 | = ri1 , |xi x2 | = ri2 and |xi x3 | = ri3 , once weve chosen whether x4 lies above or below the plane dened by (x1 , x2 , x3 ). Let us set up a co-ordinate system that moves with the body, its origin at x1 , its rst basis vector r 1 = (x2 x1 )/|x2 x1 |, the second, r , orthogonal to r but lying in the ( x , x , x ) plane and the third given 2 1 1 2 3 by r 3 = r 1 r 2 . In this frame, the co-ordinates ri of particles in the body do not change with time, r i = 0. In the inertial x frame, the particles co-ordinates
xi = R + B ri ,

where the constraint forces are of the form kj (xk xj ) with kj = jk .

(A.11)

where R = x1 and the rotation matrix B is set by the orientation of ( r1 , r 2 , r 3 ). We use this R and a set of three angles, known as Euler angles, that describe B as our six generalized co-ordinates for the body. The

31

S7 & BT VII: Classical mechanics particles velocities + (xi R) x i = R + B ( ri ), =R

(A.12)

the rst equality following from the denition (1.52) of the (x-frame) angular velocity , the second from the denition of the (r-frame) angular velocity B 1 together with (A.11).

Angular momentum of a rigid body rotating about a xed point Before dening the Euler angles, let us investigate the case of a rigid body rotating about a xed point x = r = 0, so that R = 0. Using the relations (A.11) and (A.12) above, the angular momentum in the x frame
j xi m i x i =

B ri mi B ( ri ) = B

ri mi ( ri ) = B J ,

(A.13)

where the angular momentum vector in the r-frame

J ri mi ( ri ) r ( r ) d3 r =

(A.14) [r2 ( r)r]d3 r,

and we have moved to the continuum limit to avoid a rash of indices in the following. In tensor notation this becomes
3

Ji =
j =1

Iij j , with Iij

d3 r (r2 ij ri rj ),

(A.15)

where r = (X, Y, Z ). Since it is a real symmetric matrix, it has real eigenvalues Ii and eigenvectors bi . If we orient our co-moving axes r 1 , r 2 , r 3 so that r i = bi then I becomes diag(I1 , I2 , I3 ) and the angular momentum Ji = Ii i . The bi are known as the bodys principal or body axes and the Ii its principal moments of inertia. The following table shows Iij for some simple mass distributions, assuming that each has total mass M and that the mass is distributed uniformly. Unless stated otherwise, Iij is measured about the centre of mass. Rod length a (about centre) Rod length a (about end) Ring radius a Disc radius a Spherical shell radius a Sphere radius a Exercise: Verify these! Note that the vectors x, and j live in the x frame, while r, and J live in the r frame. Vectors in dierent frames can meet only through the intercession of the operator B . From (1.46), if we have a vector V that lives in the r frame, then the rate of change of the corresponding x-frame vector, v B V , is given by dv + V ). = B (V dt (A.17)
1 2 12 M a diag(1, 1, 0) 1 2 3 M a diag(1, 1, 0) 1 2 2 M a diag(1, 1, 2) 1 2 4 M a diag(1, 1, 2) 2 2 3 M a diag(1, 1, 1) 2 2 5 M a diag(1, 1, 1)

where the Kronecker delta symbol ij = 1 if i = j and is zero otherwise. Writing out the inertia tensor Iij explicitly, 2 Y + Z2 XY XZ (A.16) I = d3 r (r) XY X2 + Z2 Y Z , XZ Y Z X2 + Y 2

Iij

S7 & BT VII: Classical mechanics We can immediately apply this to j = B J in the case of a free rigid body rotating about a xed point (e.g., centre of mass of a freely falling body). Since there are no external torques j is conserved. So, dj + J) = 0 = B (J dt + J = 0, J (A.18)

32

which is known as Eulers equation. In the principal-axis frame Ji = Ii i and Eulers equation becomes d1 = (I2 I3 )2 3 , dt d2 I2 = (I3 I1 )3 1 , dt d3 I3 = (I1 I2 )1 2 . dt I1

(A.19)

The motion of rigid bodies is interesting because the angular momentum J is not proportional to the angular velocity (unless I1 = I2 = I3 ). Exercise: (The tennis racquet theorem) A rigid body rotates freely about its third principal axis, with J = (0, 0, I3 3 ). It is given a small perturbation, so that 1 and 2 are non-zero, but small. By substituting trial solutions of the form 1 = a1 ekt and 2 = a2 ekt into (A.19) and neglecting second-order terms such as 1 2 , show that the motion is stable (k 2 < 0) if either I3 > max(I1 , I2 ) or I3 < min(I1 , I2 ). Thus rotation about either the short or the long axis of a free rigid body is stable, but rotation about the intermediate axis is unstable. Exercise: Consider a free symmetric top with I1 = I2 = I3 . Show that 3 is a constant of motion and that the angular velocity precesses around the r 3 axis with frequency p = I1 I3 3 . I1 (A.20)

Notice that the precession is retrograde (p < 0) if I3 > I1 (i.e., if the body is oblate). Kinetic energy of a rigid body rotating about a xed point Using x = B ( r), the kinetic energy 1 2 T =2 m x becomes i i i T = =
1 2 1 2

[B ( r)] [B ( r)] d3 r = 2 r2 ( r)2 d3 r.

1 2

( r) d3 r

(A.21)

We can rewrite the contents of the square brackets as 2 r2 ( r)2 = so that T =

1 2 ij 1 T i Iij j = 2 I ,

ij

i j r2 ij

i ri
i j

j rj =
ij

i j (r2 ij ri rj ),

(A.22)

(A.23)

where Iij are the components of the inertia tensor (A.15).

= 0) and we take R to Exercise: Starting from (A.12), show that if we allow translational motion (R be the position of the bodys centre of mass, then the kinetic energy of the body can be split into two parts, 1 2 + 1 T =2 MR i Iij j , (A.24) 2
ij

33

S7 & BT VII: Classical mechanics where M is the total mass of the body. The rst term is the kinetic energy of translational motion of the bodys centre of mass, the second the rotational kinetic energy of the body about its centre of mass. The rest of this section explains how to obtain a Lagrangian for a body moving in a uniform gravitational eld about a xed point (e.g., a top spinning about a point on a table). We rst introduce Euler angles. These describe the rotation matrix B and serve as our generalized co-ordinates. Then we obtain the potential energy and kinetic energy in terms of these angles and the principal moments of inertia of the system. Euler angles The standard method for parametrizing B is as follows. Start with the r 1 , r 2 , r 3 axes coincident with the inertial x 1 , x 2 , x 3 axes. Then apply the following sequence of rotations to the former: 1. Rotate through an angle about the r 3 = x 3 axis. Under this rotation r 3 = x 3 is unchanged and r 1 (temporarily) picks out a new direction known as the line of nodes. Label this direction r N . 2. Rotate through an angle about the line of nodes (the temporary r 1 axis). The line of nodes remains xed, and r 3 comes to its nal position. 3. rotate through an angle about the r 3 axis. Writing out this sequence of operations explicitly, we have that
r 1 cos r 2 = sin 0 r 3

sin cos 0

1 0 00 0 1

0 cos sin

cos 0 sin sin 0 cos

sin cos 0

x 1 0 2 , 0x 1 x 3

(A.25)

or r = Cx , where C is the product of the three rotation matrices above. Since C is orthogonal, we have that x = CT r , or x i = Cji r j . So, a point P with co-ordinates r = ri r i in the r basis can be written as
a = ri r i = ri Cij x j .

(A.26)

In other words, the point P has x co-ordinates x = B r, where the (co-ordinate) rotation matrix B = C T . Exercise: It is perhaps not immediately obvious, so show that any rotation can be represented by some choice (sometimes not unique) of Euler angles (, , ). Potential energy of an axisymmetric top Assume that the top is rotationally symmetric about the r 3 axis. Then V does not change as and are varied, because in the denition of Euler angles the and rotations take place about this axis. Therefore V = mgl cos , where (r1 , r2 , r3 ) = (0, 0, l) is the position of the tops centre of mass in the body frame. Kinetic energy in terms of Euler angles Finally, we need to express the appearing in (A.23) in terms of (, , ) and their derivatives. Between times t and t + dt the orientation of the rigid body changes from (, , ) to ( + d, + d, + d ) and the x co-ordinates of a point r = constant on the body vary from x to x + dx. Since x(t) = B (t)r, we have that (to rst order in dt, d etc) dx = B ( + d, + d, + d )B 1 (, , )x x = dt x, (A.27)

the last equality following from the denition (1.52) of angular velocity. Taking the changes in each angle separately, B ( + d, , )B 1 (, , )x x = dt x B (, , + d )B B (, + d, )B 1 (, , )x x = dt x
1

(A.28)

(, , )x x = dt x,

and the total angular velocity = + + (again, to rst order). Our job is to express , and in terms of either the x or the r basis. Let us do this rst for the special case = = 0. The rst of equations (A.28) is a rotation of d about the x 3 axis, giving x sin r cos r = 3 = 2 + 3 , (A.29)

S7 & BT VII: Classical mechanics because, solving (A.25) for x = (0, 0, 1)T , x 3 = r 2 sin + r 3 cos when = = 0. The second is a rotation of d about the line of nodes r N (= r 1 when = = 0). Therefore r = 1 . The third is a rotation of d about r 3 , so r = 3 .
T

34

(A.30) (A.31)

Gathering these last three equations together, = + + = B (1 , 2 , 3 ) with 1 = , sin , 2 = + cos . 3 = (A.32)

The expression (A.23) for the kinetic energy of our axisymmetric (I1 = I2 ) top becomes
1 2 + 2 sin2 + 1 I3 + cos T =2 I1 2 2

(A.33)

This is assuming that = = 0. But for an axisymmetric top the kinetic energy cannot depend on these cyclic co-ordinates and we can always choose the origin of reference for and to have = = 0. Thus the expression holds for all , and so the Lagrangian for an axisymmetric top moving about a xed point in a uniform eld is
1 2 2 2 L=T V = 1 2 I1 + sin + 2 I3 + cos 2

mgl cos .

(A.34)

Exercise: (optional detour) It is also possible to obtain in terms of (, , ) without assuming that = = 0. Show that for general (, , ), x r r = 3 + N + 3 sin sin + cos ] sin cos sin ] + cos ] = [ r1 + [ r2 + [ r3 , (A.35)

using (A.25) to obtain x3 and rN in terms of the r i . Substitute this into the expression (A.23) for the kinetic energy and show that it reduces to (A.33) when I1 = I2 . Exercise: Free symmetric top revisited We have already seen how the angular velocity of a free symmetric top precesses about r 3 at a rate p = 3 (I1 I3 )/I1 (A.20). Now let us look at it from an inertial frame. By substituting = p t into (A.35) or otherwise explain how one can identify with the precession rate p . Use this together with 3 = + cos from (A.35) to show that 2 = I3 3 /I1 cos and hence that the wobble rate of a uniform disc (I1 = I2 = 1 2 I3 ) satises when 1 , 2 3 . Pinned axisymmetric top in uniform gravitational eld The generalized momenta p = p = L 1 sin2 + I 3 cos2 + I 3 cos , = I L 3 + I 3 cos = I

(A.36)

in (A.34) are clearly constants of motion. So too is the energy E = T + V , which can be written as 2 E=1 2 I1 + This comes from using (A.36) to eliminate = p cos , I3 = p p cos , I1 sin2 (A.38) p2 (p p cos )2 + + mgl cos . 2 2I3 2I1 sin (A.37)

35

S7 & BT VII: Classical mechanics from the expression (A.33) for T in favour of the constants p , p . Therefore the problem reduces to motion in the one-dimensional potential Ve () given by the contents of the square bracket in (A.37). Substituting u = cos , equation (A.37) becomes E= or, rearranging, u 2 = ( u)(1 u2 ) (a bu)2 f (u) with constants a= p , I1 b= p , I1 = 1 I1 2E p I3 , = 2mgl > 0. I1 (A.40) (A.41)
1 2

p2 I1 u 2 (p p u)2 + + + mglu, 2 2 1u 2I1 (1 u ) 2I3

(A.39)

f (u) is a cubic with f (u) as u . Since u = cos , u must lie between -1 and 1. Looking at (A.41) we see that f (1) < 0 unless a = b. Therefore f (u) has two roots u1 , u2 in the interval 1 < u < 1, between which u 2 = f (u) > 0. This means that the inclination nutates (nods) between two values 1 = cos1 u1 and 2 = cos1 u2 . Meanwhile the azimuthal angle precesses at a rate = a bu . 1 u2 (A.42)

changes sign meaning that the curve traced by the top on the (, ) If u = b/a lies between u1 and u2 , sphere has loops see lectures!

B Small oscillations
A mechanical system is in equilibrium if all time derivatives vanish. In particular, if q = q0 is an equilibrium conguration, then we must have q = 0 and, from the EL equation, L/ q = 0 too. To study the behaviour of a system close to equilibrium, the usual rst step is to linearize the equations of motion. This reduces the problem to modelling a coupled set of simple harmonic oscillators, making it easy to test whether the equilibrium is stable or unstable, to calculate the frequencies with which the system rings when knocked, and much more. Expanding L(q , q ) to second order as a Taylor series about (q , q ) = (q0 , 0), ) = L(q0 , 0) + L(q + h, q +h
i

hi

L qi

+
(q0 ,0) i

i h

L q i

(q0 ,0)

1 2 ij

j +h i C T hj + h i Mij h j + O(h3 ), hi Fij hj + hi Cij h ij

(B.1)

T where the constants Fij = Fji 2 L/qi qj , Mij = Mji 2 L/ q i q j and Cij = Cji 2 L/qi q j , all evaulated at (q , q ) = (q0 , 0). Remembering that L/qi = 0 at equilibrium, it is easy to see that none of the rst three terms aect the equations of motion. The linearized EL equation for hk is then d 1 T 1 j j + 1 iC T = 0 hi Cik + 1 Ckj hj + Mkj h Fkj hj + 2 Ckj h h ik 2 2 dt 2 i j j j j i (B.2) i + (Cik Cki )h i Fki hi = 0. Mki h i

The solutions to this homogeneous linear equation are of the form h(t) = Q exp(it), with the vector Q and related through the eigenvalue equation + F ]Q = 0, [ 2 M i C (B.3)

S7 & BT VII: Classical mechanics ij = (Cij Cij ) is the antisymmetric part of C . Taking the determinant of this, the eigenfrequencies where C are given by the roots of + 2 M ) = 0, det(F i C (B.4) The system is (linearly) stable if all the eigenfrequencies are real. For most problems L=T V =
1 2 ij

36

aij (q )q i q j V (q ),

(B.5)

with some symmetric functions aij (q ) = aji (q ) such that the kinetic energy is a positive denite quadratic form in the velocities. For typical cases it turns out that Mij = aij (q0 ), Fij = 2 V /qi qj and Cij = 0. An exception is when there are velocity-dependent forces (e.g., motion in a rotating frame or in an electromagnetic eld). We simply ignore such problems in the following and assume from now on that Cij = 0. It is easy to see that each of the eigenfrequencies is either purely real or purely imaginary. Substituting hi = Qi exp(it) into (B.2), multiplying by Qk , summing over k and rearranging gives 2 = Each of the sums is real, because Mki Qk Qi
ki

Fki Qk Qi /
ki ki

Mki Qk Qi .

(B.6)

=
ki

Mki Qk Qi =
ki

Mik Qk Qi =
ki

Mki Qk Qi ,

(B.7)

by the symmetry of Mki (and Fki ), and swapping labels (i, k ) in the last step. Since the kinetic energy 1 ih j is a positive denite quadratic form, it follows that all 2 > 0 (and therefore the system is T = 2 Mij h stable) if q0 is a local minimum of V . Normal co-ordinates If the eigenfrequencies obtained by solving (B.4) are distinct, then the corresponding eigenvectors Q are orthogonal in the sense that
QT M Q = 0,

if = .

(B.8)

This follows on multiplying the (C = 0) eigenvalue equation (B.4)

2 (F + M )Q = 0

(B.9)

by another eigenvector Q and then using the symmetry of F and M to show that ( )QT M Q = 0. The importance of this is that any small oscillation h(t) that satises the linearized equation of motion (B.2) can be decomposed into a sum of normal modes,
h(t) =

a Q cos( t + ),

(B.10)

where the amplitudes a and phases can be found by premultiplying (B.10) by QT M to obtain
QT M h(t) = a cos( + )

(B.11)

T (assuming the Qi are normalized such that QT M Q = ). Thus for each , Q M h(t) is a combination of the original co-ordinates that oscillates sinusoidally at angular frequency , regardless of how the system was set into motion. A combination of the co-ordinates that inevitably oscillates sinusoidally is called a normal co-ordinate.

37

S7 & BT VII: Classical mechanics Exercise: In terms of generalized co-ordinates (1 , 2 ), a double pendulum has Lagrangian 2 + 1 ml2 2 + ml2 cos(1 2 ) 1 2 + 2mgl cos 1 + mgl cos 2 . L = ml2 1 2 2 Expanding about the equilibrium 1 = 2 = 0 to second order, show that this may be written L 2 + 1 ml2 2 + ml2 1 2 mgl2 1 mgl2 , ml2 1 2 1 2 2 2 (B.13) (B.12)

and that the EL equations in matrix form are = g l 2 2 1 2

,

(B.14)

where = (1 , 2 )T . Find the normal modes. [Ans: eigenmodes (1, 2)T ei t and (1, 2)T ei+ t , with 2 eigenfrequencies = (g/l)(2 2).]

J. Magorrian, MT 2009 (most problems taken from J. J. Binneys 2006 course)

S7: Classical mechanics problem set 1

1.

A particle is conned to move under gravity along a smooth wire that passes through two rings at (x, y, z ) = (0, 0, h) and (X, 0, 0). The particle starts at rest from the rst, upper ring. Using conservation of energy, show that the time for the particle to travel from the upper to lower ring is given by
X

T [z (x)] =
0

1+z2 2g (h z )

1/2

dx,

(1-1)

where z (x) is the height of the wire as a function of horizontal position x. Find the shape z (x) that extremizes T [z (x)]. [Hint: integrals of the form 1 /2 At dt (1-2) B+t can be solved by substituting t = A (B + A) sin2 .]
Ans: By conservation of energy, the particle moves with speed v 2 (z ) = 2g (h z ). We then have that
2 + z 2 = v2 = x dx dt
2

1+z

= 2g (h z ),

(1-3)

where z = dz/dx. Taking the square root of both sides and rearranging,
1+z2 2g (h z )
1/2

dt =

dx,

(1-4)

so that the time taken to travel from x = 0 to x = X along path z (x) is given by [z (x)] =
1+z2 2g (h z )
1/2

X 0

L(z, z , x) dx with

L(z, z , x) =

(1-5)

This is a functional of the form weve encountered in lectures, but with t replaced by x and x(t) by z (x). Since L does not depend explicitly on x, the path z (x) that extremizes [z (x)] satises L L 1 L= = z 1+z2 2g (h z )(1 + z 2 )
1/2

constant = z Rearranging and integrating,

(1-6)

z (x)

hz A+z

1/2

dz = x,

(1-7)

where A is a constant and we have used the bc z (x = 0) = h. Substituting z = h (A + h) sin2 and integrating, x = ( A + h) Writing 2 this becomes
h [ sin ] , x = A+ 2 A+h z = h 2 [1 cos ] , 1 2

sin 2 .

(1-8)

(1-9)

which is a cycloid with (x, z ) = (0, h) at its cusp. The constant A is determined by the condition that the curve pass through the endpoint (X, 0).

2.

Write down the Lagrangian for the motion of a particle of mass m in a potential (R, ) and obtain the equations of motion in plane-polar co-ordinates (R, ). Show that if does not explicitly depend on then is a constant of the motion and interpret this result physically. the generalized momentum p L/
2 ]. The Lagrangian L = T V = , so T = 1 m[R 2 + R2 and the angular speed R Ans: The radial speed is R 2 2 2 2 1 m[R + R ] m(R, ), for which the EL equations become 2 d 2 m , ) = mR (mR dt R d 2 (mR ) = m . dt

(2-1)

is conserved if does not depend on ; The second of these shows that the generalized momentum p mR2 the angular momentum of the particle is constant if there are no torques acting on it.

Obtain the Lagrangian in terms of the variables u 1/R and . Show that if (R) = /R the EL equations give u() = A cos( 0 ) + B, (2-2) where A, B and 0 are arbitrary constants. Show that the orbit is an ellipse if B > A and a parabola or hyperbola otherwise.
= u/u 2 and the Lagrangian becomes Ans: For R = 1/u, R 2 2 u + 2 4 u u

1 m L= 2

m(u).

(2-3)

The EL equation for ,

d dt u2 =0

= h, a constant. u2

(2-4)

The EL equation for u is

d dt

u u4

+ 2m

2 2 u + m 3 m = 0. 5 u u

(2-5)

= hu2 and dt = d/hu2 to obtain We can eliminate t from this by substituting d d

hu2

m 2 du hu u4 d

2m

u5

h2 u 4

du d

+ mh u m = 0.

(2-6)

Expanding the derivative in the rst term and simplifying, d2 u +u = 2, d2 h for which the general solution is u() = A cos( 0 ) + /h2 ,
2

(2-7)

(2-8)

where A and 0 are constants. If B = /h > A then u > 0 and so the orbit is bound. To nd the general shape of the orbit, let x = r cos( 0 ) and y = r sin( 0 ). Then, from the general solution above with B = /h2 ,
1 =A

x + 2 r h

1 Ax = Br

(1 Ax) = B

x2 + y 2 .

(2-9)

Rearranging gives (B 2 A2 )x2 + 2Ax + By 2 = 1, which is the equation of a conic section an ellipse if B > A, hyperbola if B < A or parabola if B = A.

3.

A particle of mass m slides inside a smooth straight tube OA. The particle is connected to point O by a light spring of natural length a and spring constant mk/a. The system rotates in a horizontal plane with constant angular velocity about a xed vertical axis through O. Find the distance r of the particle from O at time t for the case when 2 < k/a, if r = a and r = 0 at t = 0. Show also for this case that the maximum value of the reaction of the tube on the particle is 2ma 3 /b, where b2 = k/a 2 .
1 2 + 2 r 2 ) and V (r ) = 1 m(r (mk/a)(r a)2 , so the Lagrangian Ans: T = 2 2

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

1 2

mk 2 (r a) . a

(3-1)

The equation of motion is mk d m 2 r + mr (r a) = 0 dt a

+ 2 r = k, r (3-2)

with 2 k/a 2 (= b2 in the question). The general solution is r = A cos t + B sin t + k

2

(3-3)

= 0 at t = 0 gives Choosing the integration constants A and B to satisfy r = a and r

r = a

k
2

cos t +

k
2

(3-4)

The angular momentum mr2 of the particle varies as it oscillates in and out. Therefore the particle feels a torque f r, where f is the reaction of the tube on the particle and fr =
d mr2 = 2mrr dt

f = 2m r.

(3-5)

is maximum at t = /2 with value r max = (a k/2 ), so fmax = 2ma 3 /. r

Ans: (alternative) In terms of (non-rotating) polar co-ordinates (r, ), the particle has Lagrangian
) = 1 m(r 2) 2 + r2 L(r, , r, 2
1 2

mk 2 (r a) , a

(3-6)

but is subject to the holonomic constraint t = 0. So, the augmented Lagrangian L = L + ( t)

=
1 2

2) 2 + r2 m(r

1 2

mk 2 (r a) + ( t), a

(3-7)

for which the EL equations are

d 2 + mk (r a) = 0 mr mr dt a d = mr2 dt t = 0

(r ) () ().
(3-8)

The rst and last of these taken together reduce to equation (3-2) above and the derivation of r(t) proceeds as from there. To obtain the reaction force, use the fact that the (generalized) constraint force is given in this case d by L/ = . By the second and third equations of (3-8), = d mr2 , a torque, and the reaction force follows t from (3-5).

4.

Write down the Euler equations for a free rigid body in terms of its principal moments of inertia I1 , I2 , I3 and the angular velocity in the body frame. If the body is rotationally symmetric about its z axis show that 3 is a constant of the motion and that precesses about the z axis with angular frequency 3 (I3 I1 )/I1 . What is the period of this precession for the earth, which has (I3 I1 )/I1 = 0.00327?
+ J = 0, where J and are the angular momentum and angular velocity Ans: Eulers equation is J referred to axes that co-rotate with the body. If we choose these axes to be the bodys principal axes, then J = (I1 1 , I2 2 , I3 3 ) and Eulers equation becomes d1 = (I2 I3 )2 3 , dt d2 I2 = (I3 I1 )3 1 , dt d3 I3 = (I1 I2 )1 2 . dt

I1

(4-1)

Recall that I1 = I2 = I3 = (x)(y 2 + z 2 ) d3 x (x)(x2 + z 2 ) d3 x (x)(x2 + y 2 ) d3 x,

(4-2)

3 = 0 and the rst so I1 = I2 if the body is rotationally symmetric about the z axis. Then (4-1) implies that two of (4-1) can be rewritten 1 = p 2 , 2 = p 1 , (4-3)

with p 3 (I3 I1 )/I1 . The general solution is

1 = A cos(p + 0 ) 2 = A sin(p + 0 ), (4-4)

where the amplitude A and phase 0 are constants of integration. So, (and therefore J ) precesses about the bodys z axis. For the earth, 3 = (1 day)1 , so that the precession frequency should be p = 0.00327 day1 , corresponding to a period of about 300 days. In fact, the observed period is 433 days.

5.

A heavy symmetric top rotating about a xed point has Lagrangian

1 2 2 2 L= 1 2 I1 + sin + 2 I3 + cos 2

mgl cos ,

(5-1)

where I1 = I2 and I3 are its principal moments of inertia and (, , ) are the usual Euler angles. Write = Ve /, where down two conserved momenta and hence show that obeys the equation I1 Ve () = (p p cos )2 + mgl cos . 2I1 sin2 (5-2)

= 0 and = 0 Suppose that the top is released with initial conditions = 0 , 0 /I1 . it nutates about 0 with frequency I3

mglI1 /I3 . Show that

Ans: Since and are cyclic co-ordinates (L/ = L/ = 0), the conserved momenta are clearly L (I1 sin2 + I3 cos2 ) + I 3 cos , = L + cos ). p = = I3 ( p =
can be written in terms of these momenta as The velocity =

(5-3)

p p cos . I1 sin2

(5-4)

There are at least two ways of proceeding from here. The rst way is to use the EL equation for to obtain I1 ), then eliminate and in that expression in favour of p and p , nally showing that this in terms of (, , gives something equal to (minus) the derivative of the given Veff . This is long! The second, simpler, way is to nd the Hamiltonian corresponding to the given L, noting that L is of the V (q ), where the matrix elements Aij = Aji are functions of the generalized coordinates A q q form 1 2 ij i ij j 1 L = 2 j , so that H = p q i = + V (q ), in which q i is A q q = (, , ). Then pi = L/ q Aij q q j ij i ij j understood to be a function of q and p. Therefore 2 + 2 sin2 + 1 I3 + cos H (, p , p , p ) = 1 I 2 1 2
2

+ mgl cos (5-5)

p2 p2 (p p cos )2 + + mgl cos , = + 2 2I1 2I1 sin2 2I3

, (5-3) for ( + cos ) and the denition p L/ = I1 for . Hamiltons equation using (5-4) for = H/ when written out in terms of becomes I1 = dVeff /d . p = 0, = 0 0 I3 cos 0 , p = 0 I3 , so that For the initial conditions = 0 , mglI1 /I3 we have p =
2 2 I3 0 (cos 0 cos )2 , 2I1 sin2

Veff ()

(5-6)

0 the eect of the mgl cos term being negligible as mglI1 /I3 . Since the local minimum of Veff is very close to 0 , we may expand Veff as a Taylor series about 0 ,
2 2 I3 0 (cos 0 cos 0 cos + sin 0 sin )2 2I1 (sin 0 cos + cos 0 sin )2

Veff (0 + ) Veff (0 )

1 2

0 I3 I1

( ) .

(5-7)

0 /I1 . So, for small departures from equilibrium, oscillates about 0 with frequency I3

6.

A particle of mass m1 hangs by a light string of length l from a rigid support, and a second mass, m2 , hangs by an identical string from m1 . The angles with the (downward) vertical of the strings supporting m1 and 1 , 2 ) of the system. Hence show that m2 are 1 and 2 , respectively. Write down the Lagrangian L(1 , 2 , the frequencies of the two normal modes of oscillation about the equilibrium 1 = 2 = 0 are
2 =

g m1 + m2 1 l m1

m2 . m1 + m2 m2 , and (b) m2 m1 .

(6-1)

Describe the motion in each of the normal modes in the cases (a) m1

1 and has potential energy m1 gl cos 1 . Mass m2 has position (x, y ) = Ans: Mass m1 moves with speed l (l sin 1 + l sin 2 , l cos 1 + l cos 2 ), so its potential potential energy is m2 gl(cos 1 + cos 2 ) and, for 1 , 2 0, its 1 + l 2 . Expanding L = T V about 1 = 2 = 0, speed is dominated by its horizontal component of velocity, l
1 2 + 1 m2 l2 ( 1 + 2 )2 1 m1 gl 2 1 m2 gl 2 1 m2 gl 2 . m1 l2 L 2 1 1 1 2 2 2 2 2

(6-2)

The linearized equations of motion are then

d 1 + m2 l2 ( 1 + 2 ) + (m1 + m2 )gl1 = 0, m1 l2 dt d 1 + 2 ) + m2 gl2 = 0. m2 l2 ( dt

(6-3)

Substituting i = i eit , these become the eigenvalue equation 2 (m1 + m2 )l2 + (m1 + m2 )gl 2 m2 l2 Taking the determinant,
(m1 + m2 )(gl l )m2 (gl l ) = (m2 l )
2 2 2 2 2 2 2

2 m2 l2 m2 l2 + m2 gl
2

1 2

= 0.

(6-4)

m1 + m2 2 2 2 2 (gl l ) = l m2 gl = 2 l2 2 = g l 1
1 1
m2 m1 +m2

m2 m1 + m2
=

(6-5) 1 1
m2 m1 +m2 m2 m1 +m2

g l

g m1 + m2 1 l m1

m2 . m1 + m2

Case (a), m1 m2 . Both frequencies g/l because the upper, heavier mass swings without disturbance from the second. Case (b), m2 m1 . To rst order in m1 /m2 , m2 m1 = 1 m1 + m2 m1 + m2
1/2

m1
2m2

(6-6)

One frequency is now very high (the light mass m1 on the taught string) and the other is string of length 2l).

g/2l (mass on a

7.

A circular hoop of mass m and radius a hangs from a point on its circumference and is free to oscillate in its own plane. A bead of mass m can slide without friction around the hoop. Choose a set of generalized co-ordinates and write down the Lagrangian for the system. Show that the natural frequencies of small oscillations about equilibrium are 1 = 2g/a and 2 = g/2a.
Ans: See diagram below for co-ordinate system. The hoops moment of inertia about its centre of mass I0 = ma2 . By the parallel axis theorem, its moment of inertia about the pivot point is Ip = I0 + ma2 = 2ma2 . The kinetic 2 = ma2 2 and its PE is mga cos . The beads horizontal oset from the pivot energy of the hoop is then 1 I 2 p 1 + )2 and is a sin + a sin and its vertical oset is a cos + a cos . So for , 0 the bead has KE ma2 ( 2 PE mga cos mga cos . To second order in , , the Lagrangian of bead+hoop is
2 + ( + )2 + 1 mga(2 2 + 2 ). L= 1 ma2 2 2 2 (7-1)

The equations of motion are

d 2 2 ) + 2mga = 0, + ma ( + 2ma dt d + ) + mga = 0. ma2 ( dt (7-2)

Substituting = eit , = eit gives the eigenvalue equation

g 3 2 + 2 a 2

2 2 +

g a

= 0.

(7-3)

Taking the determinant of both sides,

2

g 3 2 a

g 2 4 = 0 a

2 5

g g +2 a a 2 a

=0 (7-4) = 0.

g 2a

2g

8.

The (X, Y, Z ) frame rotates with angular speed = k. A particle of mass m moves in the potential
2 2 2 2 2 2 V (X, Y, Z ) = 1 2 m(X X + Y Y + Z Z ).

(8-1)

By solving for the frequencies of the particles normal modes about the equilibrium X = Y = Z = 0, show that the motion is unstable if X < < Y .
Ans:
1 + k r ] 2 V (r ) L= 2 m [r

1 2

Y )2 + (Y + X )2 + Z 2 1 m 2 X 2 + 2 Y 2 + 2 Z 2 . m (X X Y Z 2

(8-2)

= 0, so the matrix Cij in the lectures is non-zero. Nevertheless, we can proceed using Notice that 2 L/X Y the same ideas as before. The equations of motion are d Y ) m(Y + X ) + m 2 X = 0, m(X X dt d + X ) + m(X Y ) + m 2 Y = 0, m(Y Y dt d + m 2 Z = 0. mZ Z dt

(8-3)

The Z motion decouples from the (X, Y ) motion, so one normal frequency is Z . Letting X = X0 eit , Y = Y0 eit , the equations of motion become
2 2 2 + X 2i

2i 2 2 2 + Y

X0 Y0

= 0.

(8-4)

The determinant
( + X )( + Y ) 4 = 0,
2 2 2 2 2 2 2 2

(8-5)

which, after rearranging and using the usual formula for quadratics, has solution
2 = 2 + X + Y =B
2 2 2 2 2 2 2 2 2 ) )( 2 Y ) 4( 2 X + Y (2 2 + X

(8-6)

2 2 B 2 4( 2 X )( 2 Y ).

2 2 If X < and < Y , then 4( 2 X )( 2 Y ) > 0 and one of the two roots of 2 is negative and so the motion is unstable.

9.

What is meant by the terms symmetry principle and conservation law as used in classical dynamics? Give simple examples to illustrate the symmetries underlying the conservation of linear and angular momenta. A system with three degrees of freedom described by co-ordinates q1 , q2 , q3 has Lagrangian
2 2 2 2 2 1 2 L= 2 (q 1 + q 2 +q 3 ) 1 2 (q1 + q2 + q3 ) (q2 q3 + q3 q1 + q1 q2 ),

(9-1)

where 0 < < 1 2 . Show that L is invariant under innitesimal rotations about the (1, 1, 1) axis in q -space, and hence nd a constant of motion other than the total energy. Verify from the equation of motion that it is indeed constant.
Ans: The rst two terms in the Lagrangian are clearly invariant under rotations about any axis, so we merely have to show that L = q2 q3 + q3 q1 + q1 q2 is invariant under rotations about n = (1, 1, 1). The change in L under such a rotation L = L (q + n q ) L (q ) L = (n q ) + O( 2 ) q = (q3 + q2 , q3 + q1 , q2 + q1 ) (q3 q2 , q1 q3 , q2 q1 ) 2 2 2 2 2 2 = (q3 q2 ) + (q1 q3 ) + (q2 q1 )
= 0.

(9-2)

By Noethers theorem, the corresponding constant of the motion C= L 1 , q 2 , q 3 ) (q3 q2 , q1 q3 , q2 q1 ) (n q ) = ( q q 1 (q3 q2 ) + q 2 (q1 q3 ) + q 3 (q2 q1 ). =q

(9-3)

To show that C is indeed constant, dierentiate (9-3) with respect to time,

dC 1 (q3 q2 ) + q 2 (q1 q3 ) + q 3 (q2 q1 ) + q 1 (q 3 q 2 ) + q 2 (q 1 q 3 ) + q 3 (q 2 q 1 ) =q dt 1 (q3 q2 ) + q 2 (q1 q3 ) + q 3 (q2 q1 ), =q i from the equations of motion, and take q 1 + q1 + (q3 + q2 ) = 0, q 2 + q2 + (q1 + q3 ) = 0, q 3 + q3 + (q2 + q1 ) = 0, q

(9-4)

(9-5)

to obtain
dC = (q1 + (q3 + q2 ))(q3 q2 ) + (q2 + (q1 + q3 ))(q1 q3 ) + (q3 + (q2 + q1 ))(q2 q1 ) dt = 0.

(9-6)

10.

A particle with position co-ordinates r moves in a central potential V (r). Find all potential functions V (r) and corresponding functions (r) for which the vector
K=r (r r ) + (r )r

(10-1)

is conserved.
= 0. So, Ans: K conserved K (r r ) + r 0=r d rr ) + , (r r r + r dt r (10-2)

/r . where d/dr and we have used dr/dt = r r d ) = 0 and from the equations of motion we have that The potential is spherically symmetric, so d (r r t = V / r = V r /r . Then r 0=

V ) + )r + r r (r r (r r r r V )r + r )r r 2 r + = (r r (r r r r

(10-3)

Multiplying by r and writing out the ith component of the equation in tensor notation,
i + j ri r 2 r rj r
j j

0 = V

j ri + r r i rj r

= V
j

j ri rj r
j 2

j ij + r2 r
j

j ri + rj r
j

j ij rr

(10-4)

=
j

j V ri rj + r V ij + ri rj + rij . r

j in our ICs, so the contents of the square bracket must vanish. We are also free We are free to choose each r to choose each ri , so if the [ ] is to vanish for any trajectory the factors multiplying rij and ij must vanish separately. So,

r2 V + r = 0 where A is a constant.

and

V + =0

dr

= A/r

V = A/r,

(10-5)

Find also the potentials V (r) and functions (r) for which the components of the matrix Qij r i r j + (r)ri rj (10-6)

are constants of the motion, where ri , r i (i = 1, 2, 3) are the components of position and velocity of the particle along any three independent xed axes.
Ans:
rr ij = r i r j + r i r j + i rj + ri r j ) 0=Q ri rj + ( r

r V ri rj + ( r j + r i rj ) + i rj + ri r j ) = (ri r rr r r

(10-7)

= and introduce factors ik or jk into the other terms to pull out a common r k factor. Now expand r r r r k k k k in ICs gives 0 = V /r and 0 = , so that V = 1 Independence of each ri and r r2 , where is a constant. 2

10

J. Magorrian MT 2009, borrowing from J. J. Binneys 2006 course

S7: Classical mechanics problem set 2

1.

Show that if the Hamiltonian is indepdent of a generalized co-ordinate q0 , then the conjugate momentum p0 is a constant of motion. Such co-ordinates are called cyclic co-ordinates. Give two examples of physical systems that have a cyclic co-ordinate.
i = H/qi its obvious that if H doesnt depend explicitly on, say, q1 , then p1 is conserved. Ans: Using p Examples: p is conserved in axisymmetric potential V (R, z ); pz is conserved for motion in a magnetic eld ; px , py , pz are conserved for free particles, etc. B = Bk

2.

A dynamical system has generalized co-ordinates qi and generalized momenta pi . Verify the following properties of the Poisson brackets: [qi , qj ] = [pi , pj ] = 0,
Ans: From denition of PB
[qi , pj ] =
k

[qi , pj ] = ij .

(2-1)

qi pj qk pk
k

qi pj = pk qk
k

ik jk 0 = ij .

(2-2)

Similarly, [qi , qj ] and [pi , pj ] are obviously zero.

If p is the momentum conjugate to a position vector r and L = r p, evaluate [Lx , Ly ], [Ly , Lx ] and [Lx , Lx ].
Ans: By antisymmetry of PBs, [Lx , Lx ] = 0 and [Lx , Ly ] = [Ly , Lx ]. So we need only calculate [Lx , Ly ] = [ypz zpy , zpz xpz ]. One way to do this is to use the linearity, antisymmetry and chain rule for PBs to reduce the expression to something involving the canonical commutation relations (see lectures). Another is to apply the denition of PB directly:
[Lx , Ly ] =

(ypz zpy ) (zpz xpz ) (ypz zpy ) (zpz xpz ) x px px x + (ypz zpy ) (zpz xpz ) (ypz zpy ) (zpz xpz ) y py py y (ypz zpy ) (zpz xpz ) (ypz zpy ) (zpz xpz ) + z pz pz z = (py )(x) ypx = Lz .

(2-3)

This is the overly cautious way of writing out. A more sensible answer would point out that Lx is independent of x and px and Ly is independent y and py , so the rst two lines above must clearly be zero and we need only consider the third.

The Lagrangian of a particle of mass m and charge e in a uniform magnetic eld B and electrostatic potential is 1 1 L = mr 2 + er (B r) e. (2-4) 2 2 Derive the corresponding Hamiltonian and verify that the rate of change of mr equals the Lorentz force.
Ans: Momentum p
: Hamiltonian is Legendre transform w.r.t. r L = mr 2 + 1 L H =pr e(B r ) r 2 =
1 2

L + 1 e(B r ). = mr 2 r

(2-5)

2 + e = mr

1 2m

p 1 e(B r ) 2

+ e.

(2-6)

is given by the Lorentz force, start with Hamiltons equation p = H/ r : To verify that the rate of change of mr

(LHS) (RHS)

= p

d + 1 ) mr e(B r 2 dt

H 1 1 = m p 2 e(B r ) 1 e(B r ) e 2 r r r

(2-7)

We can expand the RHS using the following (remember that r , p are independent): [p (B r )] = [r (p B )] = p B , (B r ) = r r r 2 1 (B r ) (B r ) = 2 (B r ) = 1 B 2 r 2 (B r )2 = B 2 r (B r )B 2 r r r = (B r ) B p
= H/ r gives + 1 e(B r ) and (2-8) to expand p Using p = mr 2 = p 1 e e2 d 1 + e(B r ) = + e(B r ) B mr (B r ) B e , mr dt 2 2m 2 4m r
d dt

(2-8)

(2-9)

which simplies to

= er B e(/ r ) as expected. mr

Show that the momentum component along B and the sum of the squares of the momentum components are all constants of motion when = 0. Find another constant of motion associated with time translation symmetry.
. We take the latter. Dot (2-9) with B : Ans: Notice ambiguity! Momentum could mean either p or mr d = eB pB =Bp =0 dt r

if constant.

(2-10)

= p 1 So component of p along B is conserved. So too is component of mr e(B r ). To show other two 2 : components are constant, assume = 0 and dot Lorentz force equation with mr 1 d d 2 2 ) = mr = (er B) r = 0. (m r mr 2 dt dt (2-11)

)2 is conserved. Since B mr is along conserved, must have that sum-square of other two components of So (mr (but not p) is conserved too. mr The constant of motion associated with time translation symmetry is H itself.

3.

Let p and q be canonically conjugate co-ordinates and let f (p, q ) and g (p, q ) be functions on phase space. Dene the Poisson bracket [f, g ]. Let H (p, q ) be the Hamiltonian that governs the systems dynamics. Write down the equations of motion of p and q in terms of H and the Poisson bracket.
Ans: Denition of PB:
[f, g ] = = [q, H ], p = [p, H ]. Hamiltons equations are q

f g f g . q p p q

(3-1)

In a galaxy the density of stars in phase space is f (q , p, t), where q and p each have three components. When evaluated at the location (q (t), p(t)) of any given star, f is time-independent. Show that f consequently satises f + [f, H ] = 0, (3-2) t where H is the Hamiltonian that governs the motion of every star.
Ans: f is constant along orbits so df /dt = 0. Equation (3-2) follows on using the chain rule to write df /dt = (f /q ) + p (f /q ) and then substituting for (q, p ) from Hamiltons equations. f /t + q

Consider motion in a circular razor-thin galaxy in which the potential of any star is given by the function V (R), where R is a radial co-ordinate. Express H in terms of plane polar co-ordinates (R, ) and their conjugate momenta, with the origin coinciding with the galaxys centre. Hence, or otherwise, show that in this system f satises the equation f pR f p f + + t m R mR2 where m is the mass of the star.
Ans: Standard procedure: Write down Lagrangian L in terms of plane-polar co-ordinates; this L denes ; take Legendre transform of L to get H . momenta conjugate to (R, ) through p L/ q The Lagrangian L=
1 2 V (R), 2 + R2 m R 2 (3-4)

p2 V R mR3

f = 0, pR

(3-3)

. Taking the Legendre transform, and p = mR2 so that the momenta pR = mR m(R 2 ) + V (R ) + p 2 + R2 H = pR R = 1 2m 1 2

p2 R +

p2 R2

(3-5) + V (R),

have been expressed in terms of the phase-space co-ordinates (R, , pR , p ). and where the generalized velocities R To obtain (3-3) start from (3-2), but writing out the [f, H ] explicitly:

0=

f f f H f H f H f H + [f, H ] = + + , t t R pR pR R p p

(3-6)

from which the required result follows.

4.

Show that in spherical polar co-ordinates the Hamiltonian of a particle of mass m moving in a potential V (x) is p2 1 p2 + V (x). (4-1) + p2 + H= r 2m r2 r2 sin2 Show that p = constant when V / 0 and interpret this result physically.
2 +r2 2 2 = r 2 +r 2 sin2 Ans: Starting from x = r sin cos etc, we can show that the particles velocity satises x and so the Lagrangian

L= Using pi L/qi , the momenta

1 1 2 V. 2 + r2 2 V = m r 2 + r 2 sin2 mx 2 2

(4-2)

pr = mr,

p = mr2 ,

p = mr2 sin2 .

(4-3)

Taking the Legendre transform of L,

2 + mr 2 sin2 2 L L = mr 2 + mr 2 H =pq = 1 2 + r2 2 + V (r, , ), 2 + r 2 sin2 m r 2 (4-4)

) in terms of phase-space , from which (4-1) follows on using (4-3) to express the generalized velocities (r, co-ordinates (r, , , pr , p , p ). = H/ = V / = 0 and so p (the angular momentum about the z If V does not depend on then p axis) is conserved.
2 2 Given that V depends only on r, show that [H, K ] = 0, where K p2 + p / sin . By expressing K as a and interpret this result physically. function of

Ans: One way of showing [H, K ] = 0 is by writing out the six terms in the Poisson bracket explicitly. Alterna2 tively, note that the Hamiltonian (4-1) can be written H = p2 r /2m + K/2mr + V and so
[H, K ] = 1 1 2 [p r , K ] + 2m m

K , K + [V, K ]. r2

(4-5)

The rst term vanishes because K does not depend on r. Similarly, the nal term vanishes because V = V (r) and K does not depend on pr . Using the chain rule for PBs, the middle term
[K/r , K ] = [K, K ]
2

r2

+ [1/r , K ]K,

(4-6)

but clearly [K, K ] = 0 and [1/r2 , K ] = 0 since K is independent of pr . Therefore all terms in (4-5) vanish and so [H, K ] = 0, meaning that K is a constant of motion. , Writing K out in terms of generalized velocities ,
)2 + (r sin )2 = m2 r 2 v 2 K = m2 r2 (r tangential , (4-7)

which is the square of the total angular momentum. It vanishes because the potential is spherically symmetric.

Consider circular motion with angular momentum h in a spherical potential V (r). Evaulate p () when the orbits plane is inclined by to the equatorial plane. Show that p = 0 when sin = cos and interpret this result physically.
2 2 Ans: The orbit is inclined at an angle , so p = h cos . Using h2 = K = p2 + p / sin , we have that

2 p2 = h

cos2 sin2

(4-8)

which tends to zero as sin cos the particle is at its turning point in the (R, z ) plane, which is where are zero. both p and

5.

Oblate spheroidal co-ordinates (u, v, ) are related to regular cylindrical polars (R, z, ) by R = cosh u cos v ; z = sinh u sin v. (5-1)

For a particle of mass m show that the momenta conjugate to these co-ordinates are pu = m2 (cosh2 u cos2 v )u, pv = m2 (cosh2 u cos2 v )v, 2 2 2 p = m cosh u cos v . Hence show that the Hamiltonian for motion in a potential (u, v ) is H=
2 p2 p2 u + pv + + . 2m2 (cosh2 u cos2 v ) 2m2 cosh2 u cos2 v

(5-2)

(5-3)

1 )2 + z 2 + (R 2 ] . Have that Ans: Start from L = 2 m[R

R = cosh u cos v z = sinh u sin v So

=u sinh u cos v v cosh u sin v R =u cosh u sin v + v sinh u cos v z

(5-4)

2 + z 2 [cosh2 u sin2 v + sinh2 u cos2 v ] 2 [sinh2 u cos2 v + cosh2 u sin2 v ] + 2 v 2 = 2 u R 2 2 2 [cosh2 u(1 cos2 v ) + sinh2 u cos2 v ] [(cosh2 u 1)(1 sin2 v ) + cosh2 u sin2 v ] + 2 v = u 2 2 2 2 2 +v (cosh u cos v ) = u

(5-5)

and
1 2 . 2 + v 2 ) + cosh2 u cos2 v L= 2 m2 (cosh2 u cos2 v )(u

(5-6)

) we have etc. Taking the Legendre transform of L w.r.t. (u, v, The momenta (5-2) drop out using pu L/ u that L= + pv v + p H = pu u

p2 p2 p2 v u + + 2 2 2 2 2 2 m (cosh u cos v ) m (cosh u cos v ) m cosh2 u cos2 v

2

m2

1 2

2 p2 u + pv + (m2 )2 (cosh2 u cos2 v )

p m2 cosh u cos v

(5-7)

+ ,

= pu /m2 (cosh2 u cos2 v ) etc. Simplifying gives the required expression. using (5-2) to obtain u

Show that [H, p ] = 0 and hence that p is a constant of motion. Identify it physically.
Ans:
[H, p ] =

H p p H H p p H H p p H + + , u pu u pu v pv v pv p p
0 0 0 0 1 0

(5-8)

all but one of the terms being zero since wi /wj = ij , where w (u, v, , pu , pv , p ). (Remember that phase-space co-ordinates are independent of one another!) The remaining term involves H/ which vanishes, since the potential and therefore H does not depend on . The conserved momentum, p , is the angular momentum about the symmetry axis.

6.

A particle of mass m and charge Q moves in the equatorial plane = /2 of a magnetic dipole. Given that the dipole has vector potential sin e , (6-1) A= 4r2 evaluate the Hamiltonian H (pr , p , r, ) of the system.
QA) Ans: You might be tempted to use H = (p + Q, but thats for Cartesian p. One possibility is to make 2m a canonical map to new polar co-ords, but its simpler to go back to basics and rederive H from L. 2

2 + Q(x A ) L= 1 mx 2 =
1 2

2 ] + Q(r ) 2 + r2 m[r

4r 2

. Q
4r

(6-2)

pr = mr,

+ p = mr2

Q
4r

= p

mr2 .

(6-3)

Take LT of L,
L = 1 m(r 2) + p 2 + r2 H = pr r 2 = 1 2m 1

p2 r +

r2

Q 4r

(6-4)

The particle approaches the dipole from innity at speed v and impact parameter b. Show that p and the particles speed are constants of motion.
Ans: H does not depend explicitly on , so p = const. We know that H is conserved, but from (6-4) H = 1 mv 2 2 and so the speed v is constant.

Show further that for Q > 0 the distance of closest approach to the dipole is D= where a2 Q/mv .
) and H = 1 mv 2 . H is constant along the Ans: From the ICs we have p = mbv (sign depends on initial 2 orbit. Therefore, equating H at innity with H at pericentre (radius r0 for which pr = 0), we nd that 1 1 mv 2 = 2 2 2mr0
1 2

b + b2 a 2 b + b2 + a 2

for for

> 0, < 0,

(6-5)

mbv a2 v 4r0

Q 4r0

v=

r0

bv
1 4

(6-6)

2 2 r0 = br0 a ,

using the fact that v > 0 when going from the rst to the second line. < 0. Then (6-6) becomes r 2 = | br0 1 a2 | = br0 + 1 a2 . The solutions to this First let us investigate the case 0 4 4 2 2 are given by 2r0 = b b + a . We choose the larger of the two solutions because we seek the turning point 1 < 0. for a particle that approaches from innity. Therefore r0 = 2 (b + b2 + a2 ) when the initial 2 1 2 The case > 0 is slightly more complicated. We need to solve r0 = |br0 4 a |. The solutions are 2r0 = 1 2 1 2 a ) and 2r0 = b + b2 + a2 (for br0 < 4 a ). The largest value of r0 is therefore b b2 a2 (assuming br0 > 4 2 1 2 2 ( b + b a ) , provided a < 4 br . 0 2

7.

An axisymmetric top has Lagrangian

1 2 2 2 2 L= 1 2 I1 ( sin + ) + 2 I3 ( cos + ) mga cos ,

(7-1)

where (, , ) are the usual Euler angles. Show that the tops Hamiltonian H= p2 p2 (p p cos )2 + + + mga cos . 2I1 2I3 2I1 sin2 (7-2)

Using Hamiltons equations or otherwise show that the top will precess steadily at xed inclination to the vertical provided satises (p p cos )(p cos p ) 0 = mga + . (7-3) I1 sin4
Ans: The Hamiltonian is derived in the lecture notes. For the top to precess steadily at xed inclination we , =p = 0. Using Hamiltons equation for the rate of change of p I1 require that
0=

H (p p cos ) (p p cos )2 = p sin cos mga sin 2 I1 sin I1 sin3 2 (p p cos )[p sin (p p cos ) cos ] = mga sin I1 sin3 (p p cos )[p p cos ] = mga sin . I1 sin3

(7-4)

8.

A point charge q is placed at the origin in the magnetic eld generated by a spatially conned current distribution. Given that q r (8-1) E= 4 0 r 3 and B = A with A = 0, show that the elds momentum
P
0

E B d3 x = q A(0).

(8-2)

Use this result to intrepret the formula for the canonical momentum of a charged particle in an electromagnetic eld. [Hint: use B = A and then index notation (easy) or vector identities (not so easy) to expand E B into a sum of two terms. To each term apply the tensor form of Gausss theorem, which states that d3 xi T = d2 Si T , no matter how many indices the tensor T carries. In one term you can make use of A = 0 and in the other 2 r1 = 4 3 (r).]
Ans: Easy to show that P = Permutation tensor (Levi-Civita symbol)
123 213

q
4

d r

( A).

(8-3)

even perm (1, 2, 3) = 321 = 132 = 1 odd perm (1, 2, 3) all other = 0.
=
231

312

=1

(8-4)

Handy because (summation convention)

(a b)i = ( a)i =
ijk

aj bk j ak =
ijk

ijk

ak xj

(8-5)

Useful identity (contract over middle index)

ijk klm

= il jm jl im

(8-6)

Then

( A)
i

ijk

klm

l Am )
1

= (il jm jl im ) j =

l Am j Ai .

(8-7)

i Aj j

Integrating by parts (strictly, Gauss theorem)

integrate 3 A=0

d x j

i Aj =
= 0.

d Sj

i Aj

d x i j Aj

(8-8) d x j
3

j Ai =
integrate

d Sj

Ai

d x

j j

Ai

2 r 1 =4 (r )

= 4Ai (0). + q A... To interpret, recall that the canonical momentum for a particle in a magnetic eld p = mr

J. Magorrian, HT 2011

S7 and BT VII: Classical mechanics problem set 3

1.

Starting from the denition of a canonical map in terms of the Poincar e integral invariant, explain how functions of the form F1 F1 1 (a) F1 (q, Q, t) generate mappings p = F q , P = Q , K = H + t ; F2 F2 F2 (b) F2 (q, P, t) generate mappings p = q , Q = P , K = H + t ; F3 F3 3 (c) F3 (p, Q, t) generate mappings q = F p , P = Q , K = H + t ; F4 F4 4 (d) F4 (p, P, t) generate mappings q = F p , P = P , K = H + t . Cases (c) and (d) are not covered in the notes, but you can derive them by analogy with how (b) is obtained from (a). Ans: We require that (P dQ K dt) = (pdq H dt) for any loop in extended phase space. For this to hold the integrands must be equal up to a total derivative: P dQ K dt + dS = pdq H dt, (1-1)

where S = S (P, Q, t). In the following well assume that all co-ordinate transformations are invertible. (a) Assume that P can be expressed as P = P (q, Q, t). Then we can take S = F1 (q, Q, t), so that dS = Substitute into (1-1) and rearrange: P+ F1 Q dQ + K + F1 + H dt + t F1 p dq = 0. q (1-3) F1 F1 F1 dq + dQ + dt. q Q t (1-2)

The result follows on noting that (dq, dQ, dt) are allowed to vary independently were using (q, Q, t) as independent co-ordinates of points on our loop in extended phase space. (b) The long way: apply a Legendre transform to S = F1 (q, Q, t), replacing Q by P = (F1 )/Q. This gives a new function F2 (q, P, t) = QP (F1 (q, Q, t)) in which Q(q, P, t) is given implicitly by P = F1 /Q. Rearranging, S = F1 = QP + F2 (q, P, t). Therefore dS = P dQ QdP + F2 F2 F2 dq + dP + dt, q P t (1-4)

in which Q (and therefore dQ) is to be regarded as a function of (q, P, t). Substituting into (1-1) and rearranging we have that K + F2 + H dt + t F2 F2 p dq + Q + q P dP = 0, (1-5)

from which the required result follows on noting that (dq, dP, dt) vary independently. Quicker way of deriving the correct S is to note that simply taking S = F2 (q, P, t) wont work because that would leave nothing to kill the rst term P dQ of (1-1). So try S = QP + F2 (q, P, t) in which Q is regarded as a function Q(q, P, t). (c) Take S = qp + F3 (p, Q, t), assuming q = q (p, Q, t). (d) S = qp QP + F4 (p, P, t), assuming q = q (p, P, t) and Q = Q(p, P, t).

2.

1 2 A mechanical system has Hamiltonian H = 2 (p + 2 q 2 ). By rst eliminating f (P ), or otherwise, nd a generating function F1 (q, Q) for new phase-space co-ordinates (Q, P ) in terms of which

p = f (P ) cos Q, q= f (P ) sin Q. (2-1)

State any conditions that you need to apply to f (P ). What is the Hamiltonian in the new co-ordinates (Q, P )? What are Hamiltons equations for (Q, P )? Ans: Substituting f (P ) = q/ sin Q from the second equation into the rst, p = q cos Q F1 = . sin Q q (2-2)

2 Therefore F1 (q, Q) = 1 2 q cot Q + G(Q) for some function G(Q). But we must also satisfy

P =

F1 1 G =1 q 2 2 . 2 Q sin Q Q

(2-3)

Substituting q = (f (P )/ ) sin Q into this, we see that G = 0 and f (P ) = 2P . The transformation has no explicit time dependence, so F1 /t = 0. Then K (Q, P ) = H (q (Q, P ), p(Q, P )) = Hamiltons equations are simply = 0, P = . Q (2-5) This (P, Q) are actionangle co-ordinates for the SHO Hamiltonian and a more usually written (J, ).
1 2

(f (P ) cos Q)2 + 2

f (P ) sin Q

2 2 =1 2 f (P ) = P.

(2-4)

3.

Find a generating function F2 (q, P ) for the mapping found in the previous question. (Do not expect your answer to be pretty.) Ans: Brief statement of the problem: we need to nd an F2 (q, P ) for which the implicit relations p = F2 /q , Q = F2 /P produce the mapping p= q= 2P cos Q, 2P sin Q. (3-1)

The least painful approach is to recall how F2 is related to F1 . From Q1 we know that F1 = S = P Q + F2 . So, F2 = P Q + F1 (3-2) 1 = PQ + 2 q 2 cot Q, taking the expression for F1 from Q2. We can use the second of (3-1) to obtain sin Q = q 2P
2 1 /2

cos Q = 1 q 2P from which we have that F2 = P sin1 q 2P q P 2

(3-3) ,

q 2 2P

1/2

(3-4)

An alternative is to solve the PDEs p = F2 /q , Q = F2 /P for F2 directly: 2 p = 2P 1 q 2P Substituting q /2P = sin u and integrating gives F2 = 2P 1 2 q 2P
1/2 1 /2

F2 . q

(3-5)

dq + G(P )

= 2P =P

cos2 u du + G(P ) (1 + cos 2u)du + G(P ) (3-6)

= Pu + 1 2 P sin 2u + G(P ) = P sin1 q 2P q P 2 1 q 2 2P

1/2

+ G(P ).

Finally, we need to choose G(P ) to make F2 /P = Q. Checking that G(P ) = 0 satises this condition is left as an exercise...

4.

Show that the transformation

p = eQ q = P eQ

(4-1)

is canonical. Ans: Look at pdq P dQ = eQ P eQ dQ eQ dP P dQ = dP, (4-2) which is a total derivative. Therefore the transformation is canonical. An alternative is to check Poisson brackets. Clearly we have that [q, q ] = [p, p] = 0. The only step remaining is to check whether [q, p] = 1. We have that [q, p] = p q q p = P eQ 0 eQ eQ = 1. Q P Q P (4-3)

5.

A particle has Lagrangian L(x, x , t) = 1 2 V (x, t). Write down a Hamiltonian for the particle in terms of 2x (x, px , t). The particles co-ordinates in another frame are given by x = x (t). Write down a Lagrangian L(x , x , t) and use it to construct a Hamiltonian in the new co-ordinates and momenta. Find a generating function F2 (x , p, t) for the mapping from x to x . Verify that under this mapping the original H (x, px , t) transforms to the Hamiltonian you constructed from L(x , x , t). Ans:
2 H (x, px , t) = 1 2 px + V (x, t).

(5-1)

In the accelerated co-ordinates, we have 2 V (x = x + , t), + ) L(x , x , t) = 1 2 (x and so from which px = x + H (x , px , t) = x px L(x , x , t) 1 2 + V (x = x + , t). =2 px px The mapping generated by a function F2 (x, px , t) is x = F2 /px , px = F2 /x. So, try x =x= F2 px F2 (x, px , t) = (x )px . (5-4) (5-3) (5-2)

Check: px = F2 /x = px , which is ne. The transformed Hamiltonian K (x , px , t) = H x(x , px , t), px (x , px , t), t +

2 =1 2 px + V (x = x + , t) px .

F2 t

(5-5)

6.

A particle moving in three dimensions has Hamiltonian H (x, px , t) = p2 x /2m + V (x, t). A mapping to new phase-space co-ordinates (r, pr ) is generated by the function
T F2 (x, pr , t) = pT r B x,

(6-1)

where B (t) is a time-dependent rotation matrix (BB T = I ). Show that the Hamiltonian in the new coordinates is given by K (r, pr , t) =
p2 r

2m

T + V (B r, t) + pT r B x.

(6-2)

T Show further that pT r B x can be written as pr ( r ) and explain how to obtain . Ans: It is safest to carry out the following manipulations using index notation. We have that F2 = pri Bji xj and that the transformation is given implicitly by rk = pxk F2 = ik Bji xj = Bjk xj prk F2 = = pri Bji jk = pri Bki . xk

(6-3)

Therefore r = B T x (so x = B r) and px = B pr . The transformed Hamiltonian K=H+ F2 2 = 21 m px + V (x, t) t 2 T T = 21 m (B pr ) + V (B r , t) + pr B x 2 T T = 21 m pr + V (B r , t) + pr B x,

(6-4)

as required. To express the nal term above in something more friendly, recall from the notes that r + Br x =B T x + Br = BB = x + Br ,
T T T T T T T because 0 = d dt (BB ) = BB + B B = BB + (BB ) , meaning that BB is skew symmetric and T T )T x = B B T x = x. so BB x can be written as x for some choice of . Therefore (BB T T x = x by B gives Premultiplying B B

(6-5)

T x = B T ( x ) = B T B T x = r , B

(6-6)

T where = B T . Thefore pT r B x = pr ( r ), as required. T x = x, which can be written B r = B ( r), The simplest way of nding is from the relation BB T or B B r = r. Then can be found by writing out the RHS as a skew-symmetric matrix acting on r, 0 3 2 r1 0 1 r2 . (6-7) r = 3 2 1 0 r3 explicitly and comparing to this. The components of follow by writing out B T B