Classical Mechanics and Electrodynamics

Classical Mechanics and Electrodynamics
Lecture notes FYS 3120

Jon Magne Leinaas Department of Physics, University of Oslo December 2009
Preface
These notes are prepared for the physics course FYS 3120, Classical Mechanics and Electrodynamics, at the Department of Physics, University of Oslo. The course consists of three parts, where Part I gives an introduction to Analytical Mechanics in the form of Lagrange and Hamilton theory. In Part II the subject is Special Relativity, where four-vector notation for vectors and tensors are introduced and applied to relativistic kinematics and dynamics. Finally in Part III electrodynamics is discussed from the point of view of solutions of Maxwells equations, with special focus on relativistic transformations and the radiation phenomenon. Department of Physics, University of Oslo, February 2009. December 2009 Jon Magne Leinaas
Contents
I
1
Analytical mechanics
Generalized coordinates 1.1 Physical constraints and independent variables . . . 1.1.1 Examples . . . . . . . . . . . . . . . . . . 1.2 The conguration space . . . . . . . . . . . . . . . 1.3 Virtual displacements . . . . . . . . . . . . . . . . 1.4 Applied forces and constraint forces . . . . . . . . 1.5 Static equilibrium and the principle of virtual work
9
13 13 15 18 20 21 22 25 25 29 33 34 36 38 40 41 44 44 45 45 45 47 49 49 51 52 53 56 57 60 62
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
Lagranges equations 2.1 DAlemberts principle and Lagranges equations . . . . . . . . . 2.1.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Symmetries and constants of motion . . . . . . . . . . . . . . . . 2.2.1 Cyclic coordinates . . . . . . . . . . . . . . . . . . . . . 2.2.2 Example: Point particle moving on the surface of a sphere 2.2.3 Symmetries of the Lagrangian . . . . . . . . . . . . . . . 2.2.4 Example: Particle in rotationally invariant potential . . . . 2.2.5 Time invariance and energy conservation . . . . . . . . . 2.3 Generalizing the formalism . . . . . . . . . . . . . . . . . . . . . 2.3.1 Adding a total time derivative . . . . . . . . . . . . . . . 2.3.2 Velocity dependent potentials . . . . . . . . . . . . . . . 2.4 Particle in an electromagnetic eld . . . . . . . . . . . . . . . . . 2.4.1 Lagrangian for a charged particle . . . . . . . . . . . . . 2.4.2 Example: Charged particle in a constant magnetic eld . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
Hamiltonian dynamics 3.1 Hamiltons equations . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Example: The one-dimensional harmonic oscillator . . . . . . 3.2 Phase space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Hamiltons equations for a charged particle in an electromagnetic eld 3.3.1 Example: Charged particle in a constant magnetic eld . . . 3.4 Calculus of variation and Hamiltons principle . . . . . . . . . . . . . 3.4.1 Example: Rotational surface with a minimal area . . . . . . . 5
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
CONTENTS
II
4
Relativity
The four-dimensional space-time 4.1 Lorentz transformations . . . . . . . . . . . 4.2 Rotations, boosts and the invariant distance 4.3 Relativistic four-vectors . . . . . . . . . . 4.4 Minkowski diagrams . . . . . . . . . . . . 4.5 General Lorentz transformations . . . . . . Consequences of the Lorentz transformations 5.1 Length contraction . . . . . . . . . . . . 5.2 Time dilatation . . . . . . . . . . . . . . 5.3 Proper time . . . . . . . . . . . . . . . . 5.4 The twin paradox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
73 73 75 77 80 82 85 85 87 90 91 95 95 95 96 96 97 98 100 100 102 105 105 109 112 114 116 117 119 122 125 125 127 128 131
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
The four-vector formalism and covariant equations 6.1 Notations and conventions . . . . . . . . . . . . . . . . . . . 6.1.1 Einsteins summation convention . . . . . . . . . . . 6.1.2 Metric tensor . . . . . . . . . . . . . . . . . . . . . . 6.1.3 Upper and lower indices . . . . . . . . . . . . . . . . 6.2 Lorentz transformations in covariant form . . . . . . . . . . . 6.3 General four-vectors . . . . . . . . . . . . . . . . . . . . . . 6.4 Lorentz transformation of vector components with lower index 6.5 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.6 Vector and tensor elds . . . . . . . . . . . . . . . . . . . . . Relativistic kinematics 7.1 Four-velocity and four-acceleration . . . . . . . . . . 7.1.1 Hyperbolic motion through space and time . 7.2 Relativistic energy and momentum . . . . . . . . . . 7.3 The relativistic energy-momentum relation . . . . . . 7.3.1 Space ship with constant proper acceleration 7.4 Doppler effect with photons . . . . . . . . . . . . . 7.5 Conservation of relativistic energy and momentum . 7.6 The center of mass system . . . . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
Relativistic dynamics 8.1 Newtons second law in relativistic form . . . . . . . . . . . . . . . . . . . . . . . . 8.1.1 The Lorentz force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1.2 Example: Relativistic motion of a charged particle in a constant magnetic eld 8.2 The Lagrangian for a relativistic particle . . . . . . . . . . . . . . . . . . . . . . . .
III
9
Electrodynamics
139
Maxwells equations 143 9.1 Charge conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 9.2 Gauss law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
CONTENTS 9.3 9.4 9.5 Amp` eres law . . . . . . . . . . . . . . . . . . . . . . . . . . . Gauss law for the magnetic eld and Faradays law of induction Maxwells equations in vacuum . . . . . . . . . . . . . . . . . 9.5.1 Electromagnetic potentials . . . . . . . . . . . . . . . . 9.5.2 Coulomb gauge . . . . . . . . . . . . . . . . . . . . . . Maxwells equations in covariant form . . . . . . . . . . . . . . The electromagnetic 4-potential . . . . . . . . . . . . . . . . . Lorentz transformations of the electromagnetic eld . . . . . . . 9.8.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . 9.8.2 Lorentz invariants . . . . . . . . . . . . . . . . . . . . . Example: The eld from a linear electric current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7 146 147 148 149 150 152 154 155 157 157 158 163 163 165 170 173 174 177 177 179 182 183 184 187 189 189 192 193 195 197 199 200 202
9.6 9.7 9.8
9.9
10 Dynamics of the electromagnetic eld 10.1 Electromagnetic waves . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2 Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.3 Electromagnetic energy and momentum . . . . . . . . . . . . . . . . . 10.3.1 Energy and momentum density of a monochromatic plane wave 10.3.2 Field energy and potential energy . . . . . . . . . . . . . . . . 11 Maxwells equations with stationary sources 11.1 The electrostatic equation . . . . . . . . . . . . . . 11.1.1 Multipole expansion . . . . . . . . . . . . 11.1.2 Elementary multipoles . . . . . . . . . . . 11.2 Magnetostatics . . . . . . . . . . . . . . . . . . . 11.2.1 Multipole expansion for the magnetic eld 11.2.2 Force on charge and current distributions .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
12 Electromagnetic radiation 12.1 Solutions to the time dependent equation . . . . . . . . . . . . 12.1.1 The retarded potential . . . . . . . . . . . . . . . . . . 12.2 Electromagnetic potential of a point charge . . . . . . . . . . . 12.3 General charge and current distribution: The elds far away . . . 12.4 Radiation elds . . . . . . . . . . . . . . . . . . . . . . . . . . 12.4.1 Electric dipole radiation . . . . . . . . . . . . . . . . . 12.4.2 Example: Electric dipole radiation from a linear antenna 12.5 Larmors radiation formula . . . . . . . . . . . . . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
CONTENTS
Part I
Analytical mechanics
Introduction
The form of classical mechanics we shall discuss here is often called analytical mechanics. It is essentially the same as the mechanics of Newton, but brought into a more abstract form. The analytical formulation of mechanics was developed in the 18th and 19th century mainly by two physicists, Joseph Louis Lagrange (1736-1813) and William Rowan Hamilton (1805-1865). The mathematical formulation given to mechanics by these two, and developed further by others, is generally admired for its formal beauty. Although the formalism was developed a long time ago, it is still a basic element of modern theoretical physics and has inuenced much the later theories of relativity and quantum mechanics. Lagrange and Hamilton formulated mechanics in two different ways, which we refer to as the Lagrangian and Hamiltonian formulations. They are equivalent, and in principle we may make a choice between the two, but instead it is common to study both these formulations as two sides of the analytic approach to mechanics. This is because they have different useful properties and it is advantageous to able to apply the method that is best suited to solve the problem at hand. One should note, however, a certain limitation in both these formulations of mechanics, since they in the standard form assume the forces to be conservative. Thus mechanical systems that involve friction and dissipation are generally not handled by this formulation of mechanics. We refer to systems that can be handled by the Lagrangian and Hamiltonian formalism to be Hamiltonian systems. In Newtonian mechanics force and acceleration are central concepts, and in modern terminology we often refer to this as a vector formulation of mechanics. Lagrangian and Hamiltonian mechanics are different since force is not a central concept, and potential and kinetic energy instead are functions that determine the dynamics. In some sense they are like extensions of the usual formulation of statics, where a typical problem is to nd the minimum of a potential. As a curious difference the Lagrangian, which is the function that regulates the dynamics in Lagranges formulation, is the difference between kinetic and potential energy, while the Hamiltonian which is the basic dynamical function in Hamiltons formulation, is usually the sum of kinetic and potential energy. The Hamiltonian and Lagrangian formulations are generally more easy to apply to composite systems than the Newtonian formulation is. The main problem is to identify the physical degrees of freedom of the mechanical system, to choose a corresponding set of independent variables and to express the kinetic and potential energies in terms of these. The dynamical equations, or equations of motion, are then derived in a straight forward way as differential equations determined either by the Lagrangian or the Hamiltonian. Newtonian mechanics on the other hand expresses the dynamics as motion in three-dimensional space, and for all students that have struggled with the use of the vector equations of linear and angular momentum knows that the for a composite system such a vector analysis is not always simple. However, as is generally common when a higher level of abstraction is used, there is something to gain and something to loose. A well formulated abstract theory may introduce sharper tools for analyzing a physical system, but often at the expense of more intuitive physical interpretation. That is the case also for analytical mechanics, and the vector formulation of 11
12 Newtonian mechanics is often indispensable for the physical interpretation of the theory. In the following we shall derive the basic equations of the Lagrangian and Hamiltonian mechanics from Newtonian mechanics. In this derivation there are certain complications, like the distinction between virtual displacements and physical displacements, but application of the derived formalism does not depend on these intermediate steps. The typical problem of using the Lagrangian or Hamiltonian formalism is based on a simple standardized algorithm with the following steps: First determine the degrees of freedom of the mechanical system, then choose an independent coordinate for each degree of freedom, further express the Lagrangian or Hamiltonian in terms of the coordinates and their velocities and then express the dynamics either in form of Lagranges or Hamiltons equations. The nal problem is then to solve the corresponding differential equations with the given initial conditions, but that is the purely mathematical part of the problem.
Chapter 1
Generalized coordinates
1.1 Physical constraints and independent variables
In the description of mechanical systems we often meet constraints, which means that the motion of one part of the system strictly follows the motion of another part. In the vector analysis of such a system there will be unknown forces associated with the constraints, and a part of the analysis of the system consists in eliminating the unknown forces by applying the constraint relations. One of the main simplications of the Lagrangian and Hamiltonian formulations is that the dynamics is expressed in variables that from the beginning take these constraints into account. These independent variables are known as generalized coordinates and they are generally different from Cartesian coordinates of the system. The number of generalized coordinates correspond to the number of degrees of freedom of the system, which is equal to the remaining number of variables after all constraint relations have been imposed.
z m g r2 R r1 y r m
Figure 1.1: Two small bodies connected by a rigid rod As a simple example let us consider two small bodies (particles) of equal mass m attached to the end points of a thin rigid, massless rod of length l that moves in the gravitational eld, as shown in Fig. 1. In a vector analysis of the system we write the following equations m r1 = m g + f |r1 r2 | = l 13 m r2 = m g f (1.1)
14
CHAPTER 1. GENERALIZED COORDINATES
The two rst equations are Newtons equation applied to particle 1 and particle 2 with f denoting the force from the rod on particle 1. The third equation is the constraint equation which expresses that the length of the rod is xed. The number of degrees of freedom d is easy to nd, d=3+31=5 (1.2)
where each of the two vector equations gives the contribution 3, while the constraint equation removes 1. As a set of generalized coordinates corresponding to these 5 degrees of freedom we may chose the center of mass vector R = (X, Y, Z ) and the two angles (, ) that determine the direction of the rod in space. Expressed in terms of these independent coordinates the position vectors of the end points of the rod are l l sin cos )i + (Y + sin cos )j + (Z + 2 2 l l r2 = (X sin cos )i + (Y sin cos )j + (Z 2 2 r1 = (X + l cos ) 2 l cos ) 2
(1.3)
For the kinetic and potential energy, written as functions of the independent (generalized) coordinates, this gives, T V 1 1 2 2 2 2 2 2 2 2 m( r2 1+r 2 ) = m(X + Y + Z ) + ml ( + sin ) 2 4 = mg (z1 + z2 ) = 2mgZ =
(1.4)
with the z -axis dening the vertical direction. These functions, which depend only on the 5 generalized coordinates are the input functions in the the Lagrange and Hamilton equations, and the elimination of constraint relations means that the unknown constraint force does not appear in the equations. Let us now make a more general formulation of the transition from Cartesian to generalized coordinates. Following the above example we assume that a general mechanical system can be viewed as composed of a number of small bodies with masses mi , i = 1, 2, ..., N and position vectors ri , i = 1, 2, ..., N . We assume that these cannot all move independently, due to a set of constraints that can be expressed as a functional dependence between the coordinates fj (r1 , r2 , ..., rN ; t) = 0 j = 1, 2, ..., M (1.5)
One should note that such a dependence between the coordinates is not the most general possible, for example the constraints may also depend on velocities. However, the possibility of time dependent constraints are included in the expression. Constraints that can be written in the form (1.5) are called holonomic (or rigid constraints in simpler terms), and in the following we will restrict the discussion to constraints of this type. The number of variables of the system is 3N , since for each particle there are three variables corresponding to the components of the position vector ri , but the number of independent variables is smaller, since each constraint equation reduces the number of independent variables by 1. This reduction follows since a constraint equation can be used to express one of the variables as a function of the others and thereby removing it from the set of independent variables. The number of degrees of freedom of the system is therefore d = 3N M (1.6)
1.1. PHYSICAL CONSTRAINTS AND INDEPENDENT VARIABLES
15
with M as the number of constraint equations, and d then equals the number of generalized coordinates that is needed to give a full description of the system. We denote in the following such a set of coordinates by {qk , k = 1, 2, ..., d}. Without specifying the constraints we cannot give explicit expressions for the generalized coordinates, but that is not needed for the general discussion. What is needed for the discussion is to realize that when the constraints are imposed, the 3N Cartesian coordinates can in principle be written as functions of the smaller number of generalized coordinates, ri = ri (q1 , q2 , ..., qd ; t) , i = 1, 2, ..., N (1.7)
The time dependence in the relation between the Cartesian and generalized coordinates reects the possibility that the constraints may be time dependent. For convenience we will use the notation q = {q1 , q2 , ..., qd } for the whole set so that (1.7) gets the more compact form ri = ri (q, t) , i = 1, 2, ..., N (1.8)
Note that the set of generalized coordinates can be chosen in many different ways, and often the coordinates will not all have the same physical dimension. For example some of them may have dimension of length, like the center of mass coordinates of the example above, and others may be dimensionless, like the angles in the same example. We shall illustrate the points we have made about constraints and generalized coordinates by some further examples.
1.1.1
Examples
A planar pendulum We consider a small body with mass m attached to a thin, massless rigid rod of length l that can oscillate freely about one endpoint, as shown in Fig. 1.2 a). We assume the motion to be limited to a two-dimensional plane. There are two Cartesian coordinates in this case, corresponding to the components of the position vector for the small massive body, r = xi + y j. The coordinates are restricted by one constraint equation, f (r) = |r| l = 0. The number of degrees of freedom is d = 2 1 = 1, and therefore one generalized coordinate is needed to describe the motion of the system. A natural choice for the generalized coordinate is the angle indicated in the gure. Expressed in terms of generalized coordinate the position vector of the small body is r() = l(sin i cos j) (1.9)
and we readily check that the constraint is automatically satised when r is written in this form. We may further use this expression to nd the kinetic and potential energies expressed in terms of the generalized coordinate 1 2 2 ml 2 V () = mgl cos ) = T (
(1.10)
16
y x

y x
l1
g
l1
1
m1
l2 2
m2
a)
b)
Figure 1.2: A planar pendulum a) and a double pendulum b) A planar double pendulum A slightly more complicated case is given by the double pendulum shown in Fig. 1.2 b). If we use the same step by step analysis of this system, we start by specifying the Cartesian coordinates of the two massive bodies, r1 = x1 i + y1 j and r2 = x2 i + y2 j. There are 4 such coordinates, x1 , y1 , x2 and y1 . However they are not all independent due to the two constraints, f1 (r) = |r| l1 = 0 and f2 (r1 , r2 ) = |r1 r2 | l2 = 0. The number of degrees of freedom is therefore d = 4 2 = 2, and a natural choice for the two generalized coordinates is the angles 1 and 2 . The Cartesian coordinates are now expressed in terms of the generalized coordinates as r1 (1 ) = l1 (sin 1 i cos 1 j)
r2 (1 , 2 ) = l1 (sin 1 i cos 1 j) + l2 (sin 2 i cos 2 j) This gives for the kinetic energy the expression 1 , 1 ) = T (1 , 2 , = 1 m(r1 2 + r2 2 ) 2 1 1 2 2 2 2 (m1 + m2 )l1 1 + m2 l2 2 + m2 l1 l2 cos(1 2 )1 2 2 2
(1.11)
(1.12) and for the potential energy V (1 , 2 ) = m1 gy1 + m2 gy2 = (m1 + m2 )gl1 cos 1 m2 gl2 cos 2 Rigid body As a third example we consider a three-dimensional rigid body. We may think of this as composed of a large number N of small parts, each associated with a position vector rk , k = 1, 2, ..., N . These vectors are not independent since the distance between any pair of the small parts is xed. This corresponds to a set of constraints, |rk rl | = dkl with dkl xed. However, to count the number of (1.13)
1.1. PHYSICAL CONSTRAINTS AND INDEPENDENT VARIABLES
17
independent constraints for the N parts is not so straight forward, and in this case it is therefore easier nd the number of degrees of freedom by a direct argument. The Cartesian components of the center-of-mass vector obviously is a set of independent variables R = Xi + Y j + Zk (1.14)
When these coordinates are xed, there is a further freedom to rotate the body about the center of mass. To specify the orientation of the body after performing this rotation, three coordinates are needed. This is easily seen by specifying the orientation of the body in terms of the directions of the axes of a body-xed orthogonal frame. For these axes, denoted by (x , y , z ), we see that two angles (, ) are needed to x the orientation of the z axis, while the remaining two axes are xed by a rotation with an angle in the x , y plane. A complete set of generalized coordinates may thus be chosen as q = {X, Y, Z, , , } The number of degrees of freedom of a three-dimensional rigid body is consequently 6. (1.15)
Time dependent constraint
s m g v
Figure 1.3: A body is sliding on an inclined plane while the plane moves with constant velocity in the horizontal direction.
We consider a small body sliding on an inclined plane, Fig. 1.3, and assume the motion to be restricted to the two-dimensional x, y -plane shown in the gure. The angle of inclination is , and we rst consider the case when the inclined plane is xed (v = 0). With x and y as the two Cartesian coordinates of the body, there is one constraint equation y = x tan (1.16)
and therefore one degree of freedom for the moving body. As generalized coordinate we may conveniently choose the distance s along the plane. The position vector, expressed as a function of this generalized coordinate, is simply r(s) = s(cos i sin j) (1.17)
Let us next assume that the inclined plane to be moving with constant velocity v in the x-direction. The number of degrees of freedom is still one, but the position vector is now a function of both the generalized coordinate s and of time t, r(s, t) = s(cos i sin j) + vti (1.18)
18
This corresponds to a time-dependent constraint equation y = (x vt) tan (1.19)
In the general discussion to follow we will accept that the constraints may depend on time, since this possibility can readily be taken care of by the formalism. Non-holonomic constraint
y
v x
Figure 1.4: Example of a non-holonomic constraint. The velocity v of a skate moving on ice is related to the
direction of the skate, here indicated by the angle . However there is no direct relation between the position coordinates (x, y ) and the angle .
Even if, in the analysis to follow, we shall restrict the constraints to be holonomic, it may be of interest to to consider a simple example of a non-holonomic constraint. Let us study the motion of one of the skates of a person who is skating on ice. As coordinates for the skate we may choose the two Cartesian components of the position vector r = xi + y j together with the angle that determines the orientation of the skate. There is no functional relation between these three coordinates, since for any position r the skate can have an arbitrary angle . However, under normal skating there is a constraint on the motion, since the direction of the velocity will be the same as the direction of the skate. This we may write as = v (cos i + sin j) r which gives the following relation y =x tan (1.21) (1.20)
This is a non-holonomic constraint, since it is not a functional relation between coordinates alone, but between velocities and coordinates. Such a relation cannot simply be used to reduce the number of variables, but should be treated in a different way.
1.2
The conguration space
To sum up what we have already discussed: A three-dimensional mechanical system that is composed of N small parts and which is subject to M rigid (holonomic) constraints has a number of d = 3N M
1.2. THE CONFIGURATION SPACE
19
degrees of freedom. For each degree of freedom an independent generalized coordinate qi can be chosen, so that the time evolution is fully described by the time dependence of the set of generalized coordinates q = {q1 , q2 , ..., qd } (1.22)
The set q can be interpreted as the set of coordinate of a d-dimensional space (a manifold) that is referred to as the conguration space of the system. Each point q corresponds to a possible conguration of the composite system, which species the positions of all the parts of the system in accordance with the constraints imposed on the system. In the Lagrangian formulation the time evolution in the conguration space is governed by the Lagrangian, which is a function of the generalized coordinates q , of their velocities q and possibly of time t (when the constraints are time dependent). The normal form of the Lagrangian is given as the difference between the kinetic and potential energy L(q, q, t) = T (q, q, t) V (q, t) (1.23)
In the following we shall derive the dynamical equation expressed in terms of the Lagrangian. In this derivation we begin with the vector formulation of Newtons second law applied to the parts of the system and show how this can be reformulated in terms of the generalized coordinates. For the discussion to follow it may be of interest to give a geometrical representation of the constraints and generalized coordinates. Again we assume the system to be composed of N parts, the position of each part being specied by a three-dimensional vector, rk , k = 1, 2, ..., N . Together these vectors can be thought of as a vector in a 3N dimensional space, R = {r1 , r2 , ..., rN } = {x1 , y1 , z1 , ..., xN , yN , zN } (1.24)
which is a Cartesian product of N copies of 3-dimensional, physical space, where each copy corresponds to one of the parts of the composite system. When the vector R is specied that means that the positions of all parts of the system are specied. The constraints impose a restriction on the position of the parts, which can be expressed through a functional dependence of the generalized coordinates R = R(q1 , q2 , ..., qd ; t) (1.25)
When the generalized coordinates q are varied the vector R will trace out a surface (or submanifold) of dimension d in the 3N dimensional vector space. This surface1 , where the constraints are satised, represents the conguration space of the system, and the set of generalized coordinates are coordinates on this surface, as schematically shown in the gure. Note that the conguration space will in general not be a vector space like the the 3N dimensional space. When the constraints are time independent the d-dimensional surface is a xed surface in the 3N vector space. Let us assume that we turn on the time evolution, so that the coordinates become time dependent, q = q (t). Then the composite position vector R describes the time evolution of the system in the form of a curve in R3N that is constrained to the d-dimensional surface, R(t) = R(q (t))
1
(1.26)
Such a higher-dimensional surface is often referred to as a hypersurface.
20
q2 V
q1
Figure 1.5: Geometrical representation of the conguration space as a hypersurface in the 3N dimensional
vector space dened by the Cartesian coordinates of the N small parts of the physical system. The points R that are conned to the hypersurface are those that satisfy the constraints and the generalized coordinates dene a coordinate system that covers the surface (yellow lines). The time evolution of the system describes a curve denes at all times a R(t) on the surface (blue line). If the surface is time independent, the velocity V = R tangent vector to the surface.
is a tangent vector to the surface, as shown in the gure. and the velocity vector V = R When the constraints are time dependent, and the surface therefore changes with time, the velocity vector V will in general no longer be a tangent vector to the surface at any given time, due to the motion of surface itself. However, for the discussion to follow it is convenient to introduce a type of displacement which corresponds to a situation where we freeze the surface at a given time and then move the coordinates q q + q . The corresponding displacement vector R is a tangent vector to the surface. When the constraints are time dependent such a displacement can obviously not correspond to a physical motion of the system, since the displacement takes place at a xed time. For that reason one refers to this type of change of position as a virtual displacement.
1.3
Virtual displacements
We again express the position vectors of each part of the system as functions of the generalized coordinates, rk = r(q1 , q2 , ..., qd ; t) (1.27)
where we have included the possibility of time dependent constraints. We refer to this as an explicit time dependence, since it does not come from the change of the coordinates q during motion of the system. A general displacement of the positions, which satises the constraints, can then be decomposed in a contribution from the change of general coordinates q at xed t and a contribution from change of t with xed q ,
d
drk =
j =1
rk rk dqj + dt qj t
(1.28)
1.4. APPLIED FORCES AND CONSTRAINT FORCES
21
In particular, if we consider the dynamical evolution of the system, the velocities can be expressed as
d
k = r
j =1
rk rk q j + qj t
(1.29)
The motion in part comes from the time evolution of the generalized coordinates, q = q (t), and in part from the motion of the surface dened by the constraint equations. Note that in the above expression for the velocity we distinguish between the two types of time derivatives, referred to as explicit time derivative, t and total time derivative d = dt q j
j =1
+ qj t
The rst one is simply the partial time derivative, which is well dened when acting on any function that depends on coordinates q and time t. The total time derivative, on the other hand, is meaningful only when we consider a particular time evolution, or path, expressed by time dependent coordinates q = q (t). It acts on variables that are dened on such a path in conguration space. A virtual displacement corresponds to a displacement qi at xed time t. This means that it does not correspond in general to a real, physical displacement, which will always take a nite time, but rather to an imagined displacement, consistent with the constraints for a given instant. Thus a change caused by virtual displacements measures the functional dependence of a variable on the generalized coordinates q . For the position vectors rk , the change under a virtual displacement can be written as
d
rk =
j =1
rk qj qj
(1.30)
There is no contribution from the explicit time dependence, as it is for the general displacement (1.28).
1.4
Applied forces and constraint forces
The total force acting on part k of the system can be thought of as consisting of two parts, Fk = Fa k + fk (1.31)
where fk is the generally unknown constraint force, and Fa k is the so-called applied force. The constraint forces can be regarded as a response to the applied forces caused by the presence of constraints. As a simple example, consider a body sliding on an inclined plane under the action of the gravitational force. The forces acting on the body are the gravitational force, the normal force from the plane on the body and nally the friction force acting parallel to the plane. The normal force is counteracting the normal component of the gravitational force and thus preventing any motion in the direction perpendicular to the plane. This is the force we identify as the constraint force, and the other forces we refer to as applied forces.
22
Ff
N r mg
Figure 1.6: A body on an inclined plane. The applied forces are the force of gravity and the friction. The
normal force is a constraint force. It can be viewed as a reaction to other forces that act perpendicular to the plane and neutralizes the component of the forces that would otherwise create motion in conict with the constraints. The direction of virtual displacements r is along the inclined plane. This is so even if the plane itself is moving since a virtual displacement is an imagined displacement at xed time.
We assume now that a general constraint force is similar to the normal force, in the sense of being orthogonal to any virtual displacement of the system. We write this condition as f R = 0 (1.32)
where we have introduced an 3N dimensional vector f = (f1 , f2 , ..., fN ) for he constraint forces in the same way as for the position vector R = (r1 , r2 , ..., rN ) for all the N parts of the system. The condition (1.32) means that in the 3N dimensional space the constraint force is a normal force. It acts perpendicular to the surface dened by the constraints and can be viewed as a reaction to other forces that act perpendicular to the surface. For the motion on the hypersurface, however, they make no contribution, and the main idea is to eliminate the effects of constraint forces by changing from Cartesian to generalized coordinates. The orthogonality condition (1.32) can be re-written in terms of three-dimensional vectors as fk rk = 0 (1.33)
and we note that the expression can be interpreted as the work performed by the constraint forces under the displacement rk . Thus, the work performed by the constraint forces under any virtual displacement vanishes. One should note that this does not mean that the work done by a constraint force under the the time evolution will always vanish, since the real displacement drk may have a component along the constraint force if the constraint is time dependent.
1.5
Static equilibrium and the principle of virtual work
Let us assume the mechanical system to be in static equilibrium. This means that there is a balance between the forces acting on each part of the system so that there is no motion, Fa k + fk = 0 , k = 1, 2, ..., N (1.34)
1.5. STATIC EQUILIBRIUM AND THE PRINCIPLE OF VIRTUAL WORK
23
f R
Figure 1.7: The constraint force f is a force that is perpendicular to the virtual displacements, and therefore to
the hypersurface that denes the conguration space.
Since the virtual work performed by the constraint forces always vanishes, the virtual work done by the applied forces will in a situation of equilibrium also vanish, Fa k rk = fk rk = 0 (1.35)
This form of the condition for static equilibrium is often referred to as the principle of virtual work. This condition can be re-expressed in terms of the 3N-dimensional vectors as Fa R = 0 (1.36)
Geometrically this means that in a point of equilibrium on the d dimensional surface in R3N , the applied force has to be orthogonal to the surface. This seems easy to understand: If the applied force has a non-vanishing component along the surface this will induce a motion of the system in that direction. That cannot happen in a point of static equilibrium. Let us reconsider the virtual work and re-express it in terms of the generalized coordinates. We have W =
k
Fk r k Fa k rk Fa k rk qj qj (1.37)
=
k
=
k j
=
j
Qj qj
where, at the last step we have introduced the generalized force, dened by Qj =
k
Fa k
rk qj
(1.38)
24
We note that the generalized force depends only on the applied forces, not on the constraint forces. At equilibrium the virtual work W should vanish for any virtual displacement q , and since all the coordinates qi are independent, that means that the coefcients of qi have all to vanish Qj = 0 , j = 1, 2, ..., d (equilibrium condition) (1.39)
Thus at equilibrium all the generalized forces have to vanish. Note that the same conclusion cannot be drawn about the applied forces, since the coefcients of rk may not all be independent due to the constraints.
equilibrium point
V equipotential lines
Figure 1.8: Equilibrium point. At this point the derivatives of the potential with respect to the generalized
coordinates vanish and the gradient of the potential is perpendicular to the surface dened by the conguration space.
In the special cases where the applied forces can be derived from a potential V (r1 , r2 , ...), Fa k = k V (1.40)
with k is the gradient with respect to the coordinates rk of part k of the system, the generalized force can be expressed as a gradient in conguration space, Qj = The equilibrium condition is then simply V , qj j = 1, 2, ..., d (equilibrium condition) (1.42) k V V rk = qj qj (1.41)
which means that the the potential has a local minimum (or more generally a stationary point) on the d dimensional surface that represents the conguration space of the system.
Chapter 2
Lagranges equations
2.1 DAlemberts principle and Lagranges equations
The description of the equilibrium condition discussed in the previous section can be extended to a general description of non-equilibrium dynamics, if we follow the approach of dAlembert1 . For each part of the system Newtons second law applies, mk r k = Fk = Fa k + fk and for a virtual displacement that implies rk ) rk = 0 (Fa k mk (2.2) (2.1)
which is referred to as DAlemberts principle. The important point is, like in the equilibrium case, that by introducing the virtual displacements in the equation one eliminates the (unknown) constraint forces. The expression is in fact similar to the equilibrium condition although the force which appears in this expression, Fa rk , is not simply a function of the positions rk , but also of k mk the accelerations. Nevertheless, the method used to express the equilibrium condition in terms of the generalized coordinates can be generalized to the dynamical case and that leads to Lagranges equations. In order to show this we have to rewrite the expressions. The rst part of Eq. (2.2) is easy to handle and we write it as before as Fa k rk = Qj qj
j
(2.3)
with Qj as the generalized force. Also the second part can be expressed in terms of variations in the generalized coordinates and we rewrite DAlemberts principle as (Qj mk rk rk )qj = 0 qj (2.4)
Since this should vanish for arbitrary virtual displacements, the coefcients of qj have to vanish, and this gives mk rk rk = Qj , qj j = 1, 2, ..., d (2.5)
k
1
Jean le Rond dAlembert (1717 1783) was a French mathematician, physicist and philosopher.
25
26
CHAPTER 2. LAGRANGES EQUATIONS
This can be seen as a generalized form of Newtons second law, and the objective is now to re-express the left hand side in terms of the generalized coordinates and their velocities. To proceed we split the acceleration term in two parts, mk rk d rk = ( qj dt k mk r rk ) qj k mk r d rk ) ( dt qj (2.6)
and examine each of these separately. Two re-write the rst term we rst note how the velocity vector depends on the generalized coordinates and their velocities, k r d rk = dt rk rk q j + qj t (2.7)
The expression shows that whereas the position vector depends only on the generalized coordinates, and possibly on time (if there is explicit time dependence), rk = rk (q, t) (2.8)
, which depends also on the time derivative q that is not the case for the velocity vector r . At this point we make an extension of the number of independent variables in our description. We simply consider the generalized velocities q j as variables that are independent of the generalized coordinates qj . Is that meaningful? Yes, as long as we consider all possible motions of the system, we know that to specify the positions will not also determine the velocities. So, assuming all the positions to be specied, if we change the velocities that means that we change from one possible motion of the system to another. In the following we shall therefore consider all coordinates q = q1 , q2 , ..., qd , all velocities q = q 1 , q 2 , ..., q d , and time t to be independent variables. Of course, when we consider a particular time evolution, q = q (t) then both qj and q j become dependent on t. So the challenge is not to mix these two views, the rst one when all 2d+1 variables are treated as independent, and the second one when all of them are considered as time dependent functions. However, the idea is not much more complicated than with the space and time coordinates (x, y, z, t), which in general can be considered as independent variables, but when applied to the motion of a particle, the space coordinates of the particle become dependent of time, x = x(t) etc. As already discussed these two views are captured d in the difference between the partial derivative with respect to time, t and the total derivative dt . The latter we may now write as d = dt ( qj
j
+q j )+ q j qj t
(2.9)
since we have introduced q j as independent variables. From Eq.(2.7) we deduce the following relation between partial derivatives of velocities and positions k r rk = q j qj which gives k r rk r 1 2 k k = =r r qj q j 2 q j k (2.11) (2.10)
2.1. DALEMBERTS PRINCIPLE AND LAGRANGES EQUATIONS This further gives mk

k
27
d d rk )= ( ( rk dt qj dt q j
1 d 2 mk r k) = 2 dt
T q j
(2.12)
with T as the kinetic energy of the system. This expression simplies the rst term in the right-hand side of Eq.(2.6). The second term we also re-write, and we use now the following identity d rk ) = ( dt qj = = which shows that the order of differentiations k mk r 2 rk rk ) q l + ( qj ql t qj rk rk q l + ) ql qj (2.13)
qj
( qj k r qj
d dt
and
can be interchanged. This gives k mk r k r qj
d rk ( ) = dt qj = =
( qj T qj
1 2 mk r k) 2 (2.14)
We have then shown that both terms in Eq. (2.6) can be expressed in terms of partial derivatives of the kinetic energy. By collecting terms, the equation of motion now can be written as d dt T q j T = Qj , qj j = 1, 2, ..., d (2.15)
In this form the position vectors rk has been eliminated from the equation, which only makes reference to the generalized coordinates and their velocities. The equation we have arrived at can be regarded as a reformulation of Newtons 2. law. It does not have the usual vector form. Instead there is one independent equation for each degree of freedom of the system. We will make a further modication of the equations of motion based on the assumption that the applied forces are conservative. This means that the generalized forces Qj (as well as the true forces Fk ) can then be derived from a potential function, Qj = V qj (2.16)
and the dynamical equation can therefore be written as d dt T q j T V = , qj qj j = 1, 2, ..., d (2.17)
28
We further note that since the potential only depends on the coordinates qj and not on the velocities q j the equation can be written as d dt (T V ) q j (T V ) = 0, qj j = 1, 2, ..., d (2.18)
This motivates the the introduction of the Lagrangian, dened by L = T V . In terms of this new function the dynamical equation can be written in a compact form, known as Lagranges equation, d dt L q j L = 0, qj j = 1, 2, ..., d (2.19)
Lagranges equation gives a simple and elegant description of the time evolution of the system. The dynamics is specied by a single, scalar function - the Lagrangian -, and the dynamical equation has a form that shows similarities with the equation which determines the equilibrium in a static problem. In that case the coordinate dependent potential is the relevant scalar function. In the present case it is the Lagrangian, which will in general depend on velocities as well as coordinates. It may in addition depend explicitly on time, in the following way L(q, q, t) = T (q, q, t) V (q, t) (2.20)
where explicit time dependence appears if the Cartesian coordinates are expressed as time dependent functions of the generalized coordinates (in most cases due to time dependent constraints). Note that the potential is assumed to only depend on coordinates, but not on velocities, but the formalism has a natural extension to velocity dependent potentials. Such an extension is particularly relevant to the description of charged particles in electromagnetic elds, where the magnetic force depends on the velocity of the particles. We will later show in detail how a Lagrangian can be designed for such a system. Lagranges equation motivates a general, systematic way to analyze a (conservative) mechanical system. It consist of the following steps 1. Determine a set of generalized coordinates q = (q1 , q2 , ..., qd ) that ts the system to be analyzed, one coordinate for each degree of freedom. 2. Find the potential energy V and the kinetic energy T expressed as functions of coordinates q , velocities q and possibly time t. 3. Write down the set of Lagranges equations, one equation for each generalized coordinate. 4. Solve the set of Lagranges equations for the given initial conditions. Other ways to analyze the system, in particular the vector approach of Newtonian mechanics, would usually also, when the unknown forces are eliminated, end up with a set of equations, like in point. 4. However, the method outlined above is in many cases more convenient, since it is less dependent on a visual understanding of the action of forces on different parts of the mechanical system. In the following we illustrate the Lagrangian method by some simple examples.
2.1. DALEMBERTS PRINCIPLE AND LAGRANGES EQUATIONS
29
2.1.1
Examples
Particle in a central potential, planar motion We consider a point particle of mass m which moves in a rotationally invariant potential V (r). For simplicity we assume the particle motion to be constrained to a plane (the x, y -plane). We follow the schematic approach outlined above. 1. Since the particle can move freely in the plane, the system has two degrees of freedom and a convenient set of (generalized) coordinates are, due to the rotational invariance, the polar coordinates (r, ), with r = 0 as the center of the potential. 2. The potential energy, expressed in these coordinates, is simply the function V (r), while the 2 ), and the Lagrangian is kinetic energy is T = 1 2 + r2 2 m(r 1 2 ) V (r ) 2 + r2 L = T V = m(r 2 (2.21)
3. There are two Lagranges equations, corresponding to the two coordinates r and . The r equation is d dt and the equation is d dt From the last one follows = mr2 (2.24) L L =0 d ) = 0 (mr2 dt (2.23) L r L =0 r 2 + V = 0 mr mr r (2.22)
with as a constant. The physical interpretation of this constant is the angular momentum of the particle. , and inserted in (2.22) this gives the following differential 4. Eq.(2.24) can be used to solve for equation for r(t)
2
mr
mr3
V =0 r
(2.25)
To proceed one should solve this equation with given initial conditions, but since we are less focussed on solving the equation of motion than on applying the Lagrangian formalism, we stop the analysis of the system at this point. For the case discussed here Newtons second law, in vector form, would soon lead to the same equation of motion, with a change from Cartesian to polar coordinates. The main difference between the two approaches would then be that with the vector formulation, this change of coordinates would be made after the (vector) equation of motion has been established, whereas in Lagranges formulation, this choice of coordinates may be done when the Lagrangian is established, before Lagranges equations are derived.
30
I y m1 g l-y
m2
Figure 2.1: Atwoods machine with two weights.
Atwoods machine We consider the composite system illustrated in the gure. Two bodies of mass m1 and m2 are interconnected by a cord of xed length that is suspended over a pulley. We assume the two bodies to move only vertically, and the cord to roll over the pulley without sliding. The pulley has a moment of inertia I . We will establish the Lagrange equation for the composite system. 1. The system has only one degree of freedom, and we may use the length of the cord on the left-hand side of the pulley, denoted y , as the corresponding (generalized) coordinate. This coordinate measures the (negative) height of the mass m1 relative to the position of the pulley. The corresponding position of the mass m2 is d y , with d as the sum of the two parts of the cord on both sides of the pulley. With R as the radius of the pulley, the angular position of this can be related to the coordinate y by y = R. 2. The potential energy, expressed as a function of y is V = m1 gy m2 (d y )g = (m2 m1 )gy m2 d where the last term is an unimportant constant. For the kinetic energy we nd the expression 1 1 1 2 1 I 2 T = m1 y 2 + m2 y 2 + I = (m1 + m2 + 2 )y 2 2 2 2 R This gives the following expression for the Lagrangian 1 I 2 L = (m1 + m2 + 2 )y + (m1 m2 )gy + m2 d 2 R (2.28) (2.27) (2.26)
It is the functional dependence of L on y and y that is interesting, since it is the partial derivative of L with respect to these two variables that enter into Lagranges equations. 3. The partial derivatives of the Lagrangian, with respect to coordinate and velocity, are L = (m1 m2 )g , y L I = ((m1 + m2 + 2 )y y R (2.29)
2.1. DALEMBERTS PRINCIPLE AND LAGRANGES EQUATIONS
31
and for the Lagrange equation this gives d dt L =0 y I (m1 + m2 + 2 ) y + (m2 m1 )g = 0 R m1 m2 y = g I m1 + m2 + R 2 L y
(2.30)
This equation shows that the weights move with constant acceleration, and with specied initial data the solution is easy to nd. Pendulum with accelerated point of suspension As discussed in the text, the Lagrangian formulation may include situations with explicit time dependence. We consider a particular example of this kind. Consider a planar pendulum that performs oscillations in the x, y -plane, with y as the vertical direction. The pendulum bob has mass m and the pendulum rod is rigid with xed length l and is considered as massless. It is suspended in a point A which moves with constant acceleration in the x-direction, so that the coordinates of this point are 1 xA = at2 , 2 yA = 0 (2.31)
with a as the (constant) acceleration. We will establish the equation of motion of the pendulum.
xA=at x
m
Figure 2.2: Pendulum with accelerated point of suspension 1. The pendulum moves in a plane with a xed distance to the point of suspension. This means that the system has one degree of freedom, and we choose the angle between the pendulum rod and the vertical direction as generalized coordinate. Expressed in terms of the coordinates of the
32 pendulum bob are
1 x = xA + l sin = l sin + at2 2 y = l cos with the corresponding velocities cos + at x = l sin y = l 2. The potential energy is V = mgl cos and the kinetic energy is T = = 1 m(x 2 + y 2) 2 1 2 + 2atl cos + a2 t2 ) m(l2 2
(2.32)
(2.33)
(2.34)
(2.35)
This gives the following expression for the Lagrangian 1 2 + 2atl cos + a2 t2 ) + mgl cos L = m(l2 2 (2.36)
and also explicitly on time t. As expected it depends on the generalized coordinate , its velocity The time dependence follows from the (externally determined) motion of the point of suspension. 3. Lagranges equation has the standard form d dt L L =0 (2.37) and can be expressed as a differential equation for by evaluating the partial derivatives of L with , respect to and L L This gives + ml(g sin + a cos ) = 0 ml2 (2.39) sin mg l sin = ma t l + ma t l cos = ml2 (2.38)
disappears from the equation. It is convenient to re-write We note that the term which is linear in the equation by introduce a xed angle 0 , dened by g= g 2 + a2 cos 0 , a = g 2 + a2 sin 0 (2.40)
2.2. SYMMETRIES AND CONSTANTS OF MOTION The equation of motion is then + ml ml2 g 2 + a2 sin( 0 ) = 0
33
(2.41)
and we recognize this as a standard pendulum equation, but with oscillates about the rotated direction = 0 rather than about the vertical direction = 0, and with a stronger effective acceleration of gravity g 2 + a2 . Again we leave out the last point which is to solve this equation with given boundary conditions. We only note that the form of the equation of motion is in fact what we should expect from general reasoning. If we consider the motion in an accelerated reference frame, which follows the motion of the point of suspension A, we eliminate the explicit time dependence caused by the motion of the point A. However, in such an accelerated frame there will be be a ctitious gravitational force caused by the acceleration. The corresponding acceleration of gravity is a and the direction is opposite of the direction of acceleration, which means in the negative x-direction. In this frame the effective gravitational force therefore has two components, the true gravitational force in the negative y -direction and the ctitious gravitational force in the negative x-direction. The effective acceleration of gravity is therefore g 2 + a2 and the direction is given by the angle 0 . The pendulum will perform oscillations about the direction of the effective gravitational force.
2.2
Symmetries and constants of motion
There is in physics a general and interesting connection between symmetries of a physical system and constants of motion. Well known examples of this kind are the relations between rotational symmetry and spin conservation and between translational symmetry and conservation of linear momentum. The Lagrangian formulation of classical mechanics gives a convenient way to derive constants of motion from symmetries in a direct way. A general form of this connection was shown in eld theory by Emmy Noether (Noethers theorem) in 1918. In a simpler form it is valid also for systems with a discrete set of variables, as we discuss here. One of the important consequences of nding constants of motion is that they can be used to reduce the number of variables in the problem. And even if the equations of motion cannot be fully solved, the conserved quantities may give important partial information about the motion of the system. Before discussing this connection between symmetries and constants of motion, it may be of interest with some general comments about symmetries in physics. Symmetry may have slightly different meanings depending on whether we consider a static or a dynamical situation. A body is symmetric under rotations if it looks identical when viewed from rotated positions. Similarly a crystal is symmetric under a group of transformations that may include rotations, translations and reections, if the lattice structure is invariant under these transformations. These are static situations, where the symmetry transformations leave unchanged the body or structure that we consider.2 In a dynamical situation we refer to certain transformations as symmetries when they leave the equations of motion invariant rather than physical bodies or structures. In general the equations of motion take different forms depending on the coordinates we use, but in some cases a change of coordinates will introduce no change in the form of the equations. A well-known example is the case of inertial frames, where Newtons 2. law has the same form whether we use the coordinates of one
2 The symmetries we consider are often restricted to space-transformations (or space-time transformations), but more general types of symmetry transformations may be considered, which involve mappings of one type of particles into another, changing the color of a body etc.
34
inertial reference frame or another. It is this form of dynamical symmetry which is of interest for the further discussion. Let us describe the time evolution of system by the set of coordinates q = {qi , i = 1, 2, ..., d}, where d is the number of degrees of freedom of a system. A particular solution of the equations of motion we denote by q = q (t). A coordinate transformation is a mapping q q = q (q, t) , (2.42)
where we may regard the new set of coordinates q as a function of the old set q (and possibly of time t). This transformation is a symmetry transformation of the system if any solution q (t) of the equation of motions is mapped into a new solution q (t) of the same equations of motion. We shall here focus on symmetries that follows from invariance of the Lagrangian under the coordinate transformation, in the sense L(q , q , t) = L(q, q, t) (2.43)
Note that since velocities and coordinates are considered as independent variables of the Lagrangian we need to specify how the coordinate transformations act on the velocities. This we do by assuming the coordinate transformation (2.42) to act on time dependent coordinates q = q (t). For such paths in conguration space the velocity can be expressed as the total time derivative of the coordinates q i =
j
qi q q j + i qj t
(2.44)
and this species how the coordinate transformations act on the velocities. As we shall discuss in Sect. 2.2.2 below, if the Lagrangian is invariant under a transformation (2.42) of coordinates and (2.44) of velocities, then it follows that the transformation is a symmetry transformation in the dynamical sense discussed above. At the same time this invariance gives rise to a constant of motion. In this way the Lagrangian gives a direct link between symmetries and constants of motion of the system. However, before discussing this general connection between invariance of the Lagrangian, symmetry of the equations of motion and the presence of conserved quantities, we shall consider the simpler case where constants of motion follow from the presence of cyclic coordinates.
2.2.1
Cyclic coordinates
We consider a Lagrangian of the general form L = L(q, q, t) (2.45)
with q = (q1 , q2 , ..., qd ) as the set of generalized coordinates. We further assume that the Lagrangian is independent of one of the coordinates, say q1 . This means L =0 q1 and we refer to q1 as a cyclic coordinate. From Lagranges equation then follows d dt L q 1 =0 (2.47) (2.46)
2.2. SYMMETRIES AND CONSTANTS OF MOTION This means that the physical variable p1 L q 1
35
(2.48)
which we refer to as the conjugate momentum3 to the coordinate q1 , is a constant of motion. Thus, for every cyclic coordinate there is a constant of motion. The presence of a cyclic coordinate can be used to reduce the number of independent variables from d to d 1. The coordinate q1 is already eliminated, since it does not appear in the Lagrangian, but q 1 is generally present. However also this can be eliminated by using the fact that p1 is a constant of motion. Let us write this condition in the following way, L = p1 (q2 , ..., qd ; q 1 , q 2 , ..., q d ; t) = k q 1 (2.49)
with k as a constant. In this equation we have written explicitly the functional dependence of p1 on all coordinates and velocities, except for q1 . This equation can in principle be solved for q 1 , q 1 = f (q2 , ...qd ; q 2 , ...q d ; k ; t) (2.50)
with the function f as the unspecied solution. In this way both q1 and q 1 are eliminated as variables, and the number of independent equations of motion are reduced from d to d 1. Note, however, that the d 1 Lagrange equations will not only depend on the d 1 remaining coordinates and their velocities, but also on the constant of motion k . The value of this constant is determined by the initial conditions. In the previous example of motion of a particle in a central potential the angular variable was cyclic and the corresponding conjugate momentum that was identied as the angular momentum was therefore conserved. In that case the equations of motion could be reduced to one equation, the radial equation, in a form that depended on the conserved angular momentum l. As stated above, the velocity q 1 can in principle be eliminated by solving Eq.(2.49). This may suggest that in practice to invert the expressions for p1 and q 1 may not be so simple. However, in reality this is rather straight forward, due to the general form of the Lagrangian, as we shall see. We start with the expression for the kinetic energy of a mechanical system, expressed in terms of the Cartesian coordinates T =
k
1 2 mk r k 2
(2.51)
where rk are the position vectors of the individual, small (pointlike) parts of the full system. Due to constraints, the number of degrees of freedom d will generally be smaller than the number 3N of Cartesian coordinates. We express these in the usual way as functions of a set of generalized coordinates, rk = rk (q1 , q2 , ..., qd ) (2.52)
For simplicity we assume no explicit time dependence. The expression for the velocity vectors is, k = r
i
3
rk q i qi
(2.53)
The conjugate momentum is also referred to as generalized momentum or canonical momentum.
36
which means that the velocity vectors are linear in q i . Therefore the kinetic energy is quadratic in q i , T =
ij k
1 1 rk rk ) ( mk 2 qi qj 2
gij q i q j
ij
(2.54)
with gij =
k
rk rk qi qj
(2.55)
The symmetric matrix gij only depends on the coordinates, with the cyclic coordinate q1 excluded, gij = gij (q2 , ..., qd ) The Lagrangian has a similar dependence on the velocities, L=T V = 1 2 gij (q ) q i q j V (q ) (2.57) (2.56)
ij
where V (q ) as well as gij (q ) is independent of q1 . The expression for the corresponding conjugate momentum is p1 = g11 q 1 +
i=1
g1i q i
(2.58)
which shows that p1 is a linear function of q 1 . With p1 as a constant k , this gives for q 1 the equation g11 q 1 +
i=1
g1i q i = k
(2.59)
which is easily solved for q 1 , q 1 = 1 (k g11 g1i q i )

i=1
(2.60)
2.2.2
Example: Point particle moving on the surface of a sphere
We consider a point particle of mass m that moves without friction on the surface of a sphere, under the inuence of gravitation. The gravitational eld is assumed to point in the negative z -direction. This system has two degrees of freedom, since the three Cartesian coordinates (x, y, z ) of the particle are subject to one constraint equation r = x2 + y 2 + z 2 = const. As generalized coordinates we chose the polar angles (, ), so that the Cartesian coordinates are x = r cos sin y = r sin sin z = r cos with r as a constant. The corresponding velocities are sin sin ) x = r(cos cos + cos sin ) y = r(sin cos z = r sin (2.62) (2.61)
2.2. SYMMETRIES AND CONSTANTS OF MOTION The potential energy is V = mgz = mgr cos with g as the acceleration of gravitation, and the kinetic energy is 1 2 1 2 + sin2 2) T = (x +y 2 + z 2 ) = mr2 ( 2 2 This gives the following expression for the Lagrangian 1 2 + sin2 2 ) mgr cos L = mr2 ( 2 Clearly is a cyclic coordinate, L =0 and therefore Lagranges equation for this variable reduces to L sin2 = l = mr2 with l as a constant. Lagranges equation for the variable is 2 sin cos + mgr sin = 0 mr2 mr2
37
(2.63)
(2.64)
(2.65)
(2.66)
(2.67)
(2.68)
in terms of the constant To eliminate the variable from the equation, we express, by use of (2.67), of motion l, = Inserted in (2.68) this gives mr2 l2 cos + mgr sin = 0 mr2 sin3 (2.70) l mr2 sin2 (2.69)
This illustrates the general discussion of cyclic coordinates. In the present case the elimination of the coordinate has reduced the equations of motion to one, and the only remaining trace of the coordinate is the appearance of the conserved quantity l in the equation. There is one point concerning cyclic coordinates which is of interest to comment on. That is the connection between cyclic coordinates and symmetries of the Lagrangian. Clearly the independence of L under changes in the variable means that the Lagrangian is invariant under rotations around the z -axis, which represents the direction of the gravitational eld. The rotational symmetry is therefore linked to the presence of the cyclic coordinate , and the cyclic coordinate is further related to the presence of the constant of motion l. It is straight forward to show that this constant has the physical interpretation as the z -component of the orbital angular momentum of the particle. The connection between symmetries of the Lagrangian and constants of motion is in fact more general than indicated by the presence of cyclic coordinates. To illustrate this let us assume that we can (articially) turn off the gravitational eld in the example above, so that the Lagrangian is identical to the kinetic energy only. The coordinate is cyclic as before, and the z -component of the
38
angular momentum is still a constant of motion. However, now there is invariance under all rotations in three dimensions, and the z -axis is in no way a preferred direction. This means that also the x and y -components of the angular momentum have to be conserved, not only the z -component. However, just by inspecting the cyclic coordinates of the Lagrangian this is not obvious, since we cannot choose any coordinate system so that there is one independent cyclic coordinate for each component of the angular momentum. In a more general setting each independent symmetry of the Lagrangian, even if this is not represented by a cyclic coordinate, will give rise to a conserved quantity. We shall next examine this point.
2.2.3
Symmetries of the Lagrangian
The existence of a cyclic coordinate can be viewed as expressing a symmetry of the Lagrangian in the following way. We consider a coordinate transformation of the form q1 q1 = q1 + a (2.71)
where a is a parameter that can be continuously be varied. The transformation describes a continuous set of translations in the cyclic coordinate. In the previous example that corresponds to rotations about the z axis. The fact that the coordinate is cyclic means that the Lagrangian is invariant under these translations, and from that follows that if q (t) is a solution of Lagranges equation so is the transformed coordinate set q (t). A cyclic coordinate thus corresponds to a symmetry of the system. We shall now discuss more generally how invariance of the Lagrangian under a coordinate transformation is related on one hand to a symmetry of the equations of motion and on the other hand to the presence of a constant of motion. In the general case there may be no cyclic coordinate corresponding to the symmetry transformation. We consider then a continuous set of time independent coordinate transformations q q = q (q ) , L(q , q ) = L(q, q ) . (2.72)
and assume this to be symmetry transformation in the sense that it leaves the Lagrangian invariant, (2.73)
This equation means that under a change of variables q q the Lagrangian will have the same functional dependence of the new and old variables. Since the Lagrangian determines the form of the equations of motion, this implies that the time evolution of the system, described by coordinates q (t) and by coordinates q (t) will satisfy the same equations of motion. We will demonstrate this explicitly. A change of variables will give a change in partial derivatives of the Lagrangian in the following way L qm L q m =
k
L qk L q k + ) qk qm q k qm (2.74)
=
k
L q k q k q m
Note that in the last expression there is no term proportional to qk / q m , since in a coordinate transformation the old coordinates q will not depend on the new velocities q , but only on the new coordinates q . The relation between the velocities is q k =
m
qk q qm m
(2.75)
2.2. SYMMETRIES AND CONSTANTS OF MOTION which implies q k qk = qm q m This allows a reformulation of the partial derivative of L with respect to velocities L q m We are interested in the total time derivative d L ( ) = dt q m where the last term can be rewritten as d qk ( ) = dt qm = = 2 qk 2 qk q l + ql qm t qm qk qk q + ) ql l t d L qk ( ) + dt q k qm L d qk ( ) q k dt qm =
k
39
(2.76)
L qk q k qm
(2.77)
(2.78)
( qm q k qm
(2.79)
We nally collect expressions from (2.74), (2.78) and (2.79), which give the following relation d L L ( ) = dt q m qm d L L ( ) dt q k qk qk . qm (2.80)
This demonstrates explicitly that if q (t) satises Lagranges equations, and thereby the right-hand side of (2.80) vanishes, then the transformed coordinates q (t) will also satisfy the same set of Lagranges equation. Thus a coordinate transformation that is a symmetry transformation in the sense that it leaves the Lagrangian invariant will also be a symmetry transformation in the sense that it maps solutions of the equations of motion into new solutions. Note, however, that the opposite may not always be true. There may be coordinate transformations that map solutions of the equations of motion into new solutions without leaving the Lagrangian unchanged. We will next show that when the Lagrangian is invariant under a continuous coordinate transformation4 this implies the presence of a constant of motion, and we shall nd an expression for this constant. In order to do so we will focus on transformations qi = qi + qi , with the change of coordinates qi taken to be arbitrarily small, and we therefore assume that terms that are higher order in qi can be neglected. As an example of such continuous transformations we may take the rotations about a given axis, where any rotation may be built up by a continuous change from the identity. We consider the Lagrangian L evaluated along the transformed path q (t) and relate it to L evaluated along the original path q (t) by expanding to rst order in q , L(q , q ) = L(q, q) +
k
4
L L qk + q k ) . qk q k
(2.81)
Continuous transformation here means that the transformation depends on a parameter that can be changed continuously, like the rotation angle in the case of rotational symmetry.
40 Invariance of the Lagrangian then implies (

k
L L qk + q k ) = 0 , qk q k
(2.82)
which we may re-write as L d L d L qk ( )qk + ( qk ) = 0 . qk dt q k dt q k (2.83)
We will assume that q (t) satises Lagranges equations, and the two rst terms therefore cancel. This gives d L ( qk ) = 0 . dt q k The following quantity is then a constant of motion K =
k
(2.84)
L qk . q k
(2.85)
With qk as an innitesimal change of the coordinates, it can be written as qk = Jk (2.86)
where Jk is a nite parameter characteristic for the transformation. The innitesimal parameter can be ommitted and that gives the following expression for the nite (non-innitesimal) constant of motion associated with the symmetry K=
k
L Jk . q k
(2.87)
To summarize, if we can identify a symmetry of the system, expressed as invariance of the Lagrangian under a coordinate transformation, we can use the above expression to derive a conserved quantity corresponding to this symmetry.
2.2.4
Example: Particle in rotationally invariant potential
In order to illustrate the general discussion we examine a rotationally invariant system with kinetic and potential energies 1 2 T = mr , 2 V = V (r) , (2.88)
which gives the following Lagrangian in Cartesian coordinates 1 L = m(x 2 + y 2 + z 2 ) V ( x2 + y 2 + z 2 ) , 2 and in polar coordinates 1 2 + r2 sin2 2 ) V (r) , L = m(r 2 + r2 2 (2.90) (2.89)
2.2. SYMMETRIES AND CONSTANTS OF MOTION
41
The system is obviously symmetric under all rotations about the origin (the center of the potential), but we note that expressed in Cartesian coordinates there is no cyclic coordinate corresponding to these symmetries. In polar coordinates there is one cyclic coordinate, . The corresponding conserved quantity is the conjugate momentum p = L , = mr2 sin2 (2.91)
and the physical interpretation of p is the z -component of the angular momentum )z = m(xy . (mr r yx ) = mr2 sin2 (2.92)
Clearly also the other components of the angular momentum are conserved, but there are no cyclic coordinates corresponding to these components. We use the expression derived in the last section to nd the conserved quantities associated with the rotational symmetry. First we note that an innitesimal rotation can be expressed in the form r r = r + r or r = r , (2.94) (2.93)
where the direction of the vector species the direction of the axis of rotation and the absolute value species the angle of rotation. We can explicitly verify that to rst order in the transformation (2.93) leaves r 2 unchanged, transforms in the same way (by time derivative of (2.93)) also r 2 is invariant and since the velocity r under the transformation. Consequently, the Lagrangian is invariant under the innitesimal rotations (2.93), which are therefore symmetry transformations of the system. By use of the expression (2.84) we nd the following expression for the conserved quantity associated with the symmetry transformation,
3
K=
k=1
L r = m(r r ) . xk = mr x k
(2.95)
Since this quantity is conserved for arbitrary values of the constant vector , we conclude that the vector quantity l = mr r (2.96)
is conserved. This demonstrates that the general expression we have found for a constant of motion reproduces, as expected, the angular momentum as a constant of motion when the particle moves in a rotationally invariant potential.
2.2.5
Time invariance and energy conservation
We consider a Lagrangian L = L(q, q ) that has no explicit time dependence, so that L =0 t (2.97)
42
This functional independence of t we note to be similar to the functional independence of q1 , when this is a cyclic coordinate. Time is certainly not a coordinate in the same sense as qi , and in particular there is no conjugate momentum to t. Nevertheless, there is a conserved quantity that can be derived from the time independence of L. To show this we consider the total time derivative of L when evaluated for a path q (t) that satises the equations of motion. The total time derivative picks up contributions both from the explicit dependence of L on time t and from the dynamical time dependence of L on the generalized coordinates qi (t) , dL = dt (
i
L L L q i + q i ) + q i qi t
(2.98)
We re-write this equation and make use of the fact that the time dependence of qi is determined by Lagranges equation, dL dt =
i
d L ( q i ) dt q i
L L d L ( q i + ) dt q i qi t (2.99)
=
i
d L L ( q i ) + dt q i t
This shows that the following quantity H=

i
L q i L q i
(2.100)
which is called the Hamiltonian of the system, satises the equation dH L = dt t (2.101)
This means that if L has no explicit time dependence, L t = 0, then H is time independent under the full time evolution of the system and is therefore a constant of motion. When the Lagrangian has the standard form L = T V , and when the constraints are time independent, the Hamiltonian corresponds to the sum of kinetic and potential energy, H = T + V . In that case conservation of H means that the total energy is conserved. It is easy demonstrate this by using the general expression for L (see Eq. (2.57)), which is valid for time independent constraints L= This gives L = q i and therefore H =
j
1 2
ij
gij (q ) q i q j V (q )
(2.102)
gij (q ) q j
j
(2.103)
gij (q ) q i q j (
1 2
ij
gij (q ) q i q j V (q ))
1 2
gij (q ) q i q j + V (q )
ij
= T +V
(2.104)
2.2. SYMMETRIES AND CONSTANTS OF MOTION
43
The energy conservation can be understood in the following way. We know that if energy is not conserved, the reason for this must be that there are external forces that perform a non-vanishing work on the system, either by extracting energy from or adding energy to the system. However, the assumption here is that all the applied forces (external or internal) are conservative. And the work done by conservative forces does not lead to a change of the total energy, but only a shift of energy from kinetic energy T to the potential energy V . Therefore the only external forces that can change the total energy are the non-conservative forces, and the only forces that in the present case may be nonconservative are the constraint forces. However, the constraint forces satisfy the principle of virtual work. That means that they do not perform any work under virtual displacements. If the constraints are time independent, that implies that the work under real displacements vanishes in the same way as under virtual displacements. Therefore the total energy H = T + V is conserved. However, if the constraints are time dependent there is a real difference between virtual and real displacements. In that case the the constraint forces may perform a non-vanishing work under real displacements even if the virtual work vanishes, and consequently the total energy may not be conserved. Note that the Hamiltonian is dened by Eq. (2.100) also when the constraints are time dependent, but in that case the Hamiltonian is generally not equal to the sum of kinetic and potential energy. This is seen by using the general expression k = r
i
rk r q i + qi t
(2.105)
The last term, which does not depend on q gives a more general expression for the kinetic energy energy, of the form T = 1 2 gij (q, t) q i q j +
ij i
hi (q, t)q i + f (q, t)
(2.106)
The additional terms lead to an expression for the Hamiltonian, which is in general different from T + V . One should note that even if the constraints are time dependent, the Lagrangian may in some cases be time independent, provided the functions gij , hi , f and V are all independent of time. In that case the Hamiltonian H is a constant of motion, but it is not identical to the total energy of the system. (For a particular example see Problem 2.1c in the Exercise collection.) In a similar way as the Lagrangian L is the fundamental quantity in Lagranges description of the dynamics of a physical (conservative) system, the Hamiltonian H is the fundamental quantity in Hamiltons description. However, one should note that it is not the value of the physical quantity L that is important in Lagranges formulation, but rather its functional dependence on the generalized coordinates q and their velocities q . This is so since the partial derivatives of L with respect to these variables enter in Lagranges equation. In a similar way it is not the value of H that is important in Hamiltons formulation, but rather its functional dependence on the basic variables in Hamiltons description. But whereas L is considered as a function of q and q , the Hamiltonian is instead considered as a function of q and p, where p = (p1 , p2 , ..., pd ) denotes the set of canonical momenta, pi = L q i (2.107)
This change of basic variables is important, since it is the partial derivatives of H with respect to q and p that enter into Hamiltons equation. Note, however, that we do not reserve the symbol H for this function of q and p. In the usual physicist tradition we shall use the symbol H for the physical quantity, whether this quantity is written as a function of q and p or of q and q .
44
2.3
2.3.1
Generalizing the formalism

Adding a total time derivative
A change of the the Lagrangian L(q, q, t) L (q, q, t) (2.108)
will usually lead to a change in the corresponding equations of motions, but not always. Let us consider a change given by L (q, q, t) = L(q, q, t) + d f (q, t) dt (2.109)
where f (q, t) is a differentiable function of the coordinates qi , but not of the velocities q i . The additional term, which can be written as a total time derivative, does not change the (Lagrange) equations of motion, as we can easily demonstrate. We dene the additional term as g (q, q, t) d f= dt f f q i + qi t (2.110)
and consider the contribution to the Lagrange equation from this additional term, L d L L d g g d L ( ) = ( ) + ( ) dt q i qi dt q i qi dt q i qi We have g = qi and d f d g ( )= ( )= dt q i dt qi 2f 2f q m + qm qi tqi (2.113) 2f 2f q m + qi qm qi t (2.112) (2.111)
and since we assume the function f to be well behaved, so the order of differentiation can be interchanged, these two expressions are equal. This means that the contribution to Lagranges equation vanishes, d g g ( ) =0 dt q i qi (2.114)
Therefore, two Lagrangians that differ by a total time derivative, like in (2.109), are equivalent in the sense that they give rise to the same equations of motion. In particular, if the Lagrangian is given by the standard expression L = T V , this implies that an equally valid Lagrangian for the same system, is obtained by adding (or subtracting) a total time derivative to the expression T V . This observation is sometimes useful in order to simplify the expression for the Lagrangian. One should also note, that even if a symmetry of a physical system will often correspond to invariance of the Lagrangian under a given transformation, invariance up to a total time derivative would more generally give rise to a symmetry of the equations of motion. Also in this case, when the Lagrangian is invariant up to the addition of a total time derivative, there is a constant of motion corresponding to the symmetry. This can be shown in essentially the same way as we have done for the case of an invariant Lagrangian.
2.4. PARTICLE IN AN ELECTROMAGNETIC FIELD
45
2.3.2
Velocity dependent potentials
We return to the equation of motion, in the form it had before we assumed the applied forces to be conservative (Eq.(2.15), d dt T q j T = Qj , qj j = 1, 2, ..., d (2.115)
V Lagranges equation was derived from this by writing the generalized force as Qj = q and assumj ing V to be velocity independent. However, there is an obvious possibility of extending the formalism by assuming the potential to be velocity dependent, written as U = U (q, q, t), with the generalized force depending on U as
Qi =
d U U ( ) = dt q i qi
2U q j + qj q i
2U 2U U q j + q j q i t qi qi
(2.116)
In this case the equation of motion (2.5) can be written in the standard Lagrangian form, if the Lagrangian is now dened as L(q, q, t) = T (q, q, t) U (q, q, t) (2.117)
This generalized form of Lagranges equation has an important application in the description of charged particles in electromagnetic elds, as we shall see. In that case the potential U depends linearly on the velocity and this dependence on the velocity gives rise to the magnetic force that acts on the particles.
2.4
2.4.1
Particle in an electromagnetic eld

Lagrangian for a charged particle
We consider the motion of a charged particle in an electromagnetic eld, and since there are no constraints the Cartesian coordinates of the particle are used as the generalized coordinates. The equation of motion is ma = e(E(r, t) + v B(r, t)) F(r, v, t) (2.118)
with e as the charge of the particle, E as the electric eld and B as the magnetic eld. Only in the electrostatic case, with B = 0, this equation of motion can be derived from a Lagrangian of the standard form L = T V , with V = e as the electrostatic potential. However, as we shall see, in the general case the force can be expressed in terms of a velocity dependent potential as Fi = d U U ( ) dt x i xi (2.119)
and therefore the equation of motion can be derived from the Lagrangian L = T U . In order to show this, we introduce the electromagnetic potentials E = A , t B=A (2.120)
46 and express the force in terms of the potentials F = e[
A + v ( A)] t A = e[ + (v A) v A] t
(2.121)
In component form this is Fi = e( = Ai A v Ai ) +v xi t xi d [e ev A] (eAi ) dt xi
(2.122)
If the velocity dependent potential U is now dened as U = e ev A that gives U = eAi x i (2.124) (2.123)
and it is clear from (2.122) that the Lorentz force F is related to U by Eq. (2.119). The Lagrangian of the charged particle in the electromagnetic eld is therefore 1 2 e(r, t) + er A(r, t) L = T U = mr 2 (2.125)
Let us further examine the form of the conjugate momentum and the Hamiltonian in this case. We have pi = which gives = p eA mr (2.127) L = mx i + eAi x i (2.126)
This shows that the canonical momentum p in this case is not identical to the mechanical momentum mv of the charged particle. The Hamiltonian is now L H = pr 1 = v (mv + eA) v2 + e ev A 2 1 = mv2 + e 2 1 = (p eA)2 + e 2m
(2.128)
We note that this is different from T + U , but is identical to the total energy T + V , with V = e as the potential energy of the charge in the electromagnetic eld. According to the previous discussion H should be a constant of motion if the Lagrangian has no explicit time dependence. In the present case this can be related to a more direct argument for
2.4. PARTICLE IN AN ELECTROMAGNETIC FIELD
47
conservation of energy in the following way. We rst note that time independence of L means that the potentials and therefore the electric and magnetic elds are time independent. The electric part of the force in (2.118) is Fe = e. This is a conservative force, that does not change the total energy, but only shift energy from the kinetic to the electrostatic part. The magnetic part of the force, Fm = ev B, acts in a direction perpendicular to the direction of motion, and therefore performs no work on the particle, so the total energy is left unchanged. If the potentials on the other hand are time dependent, the electric force is no longer conservative and the interaction of the particle with the electric elds will change the total energy. There is one point about the Lagrangian that is worthwhile noting. It is not gauge invariant, even if the equation of motion is gauge invariant. A gauge transformation is a modication of the potentials of the form = , t A A = A + (2.129)
with = (r, t)) as an arbitrary differentiable function of space and time. The elds E and B are left unchanged by this transformation, and usually gauge transformations are therefore considered as not corresponding to any physical change. The question is whether the non-invariance of the Lagrangian is consistent with this view. The transformation induces the following change of the Lagrangian L L = L + e( d + v ) = L + e t dt (2.130)
So we see that the gauge transformation adds a term to the Lagrangian that can be written as a total time derivative. As already discussed Lagrangians that differ by a total time derivative are equivalent, so in this sense no essential change is made under the gauge transformation.
2.4.2
Example: Charged particle in a constant magnetic eld
We assume the electromagnetic potentials are = 0, 1 A= rB 2 (2.131)
with B constant. It is straight forward to check that B = A, so that B represents a constant magnetic eld. We use the established expression for the Lagrangian of a charged particle, 1 1 1 L = mv2 + ev A = mv2 v (r B) 2 2 2 (2.132)
and will check that the corresponding Lagrange equation is consistent with the known expression for the equation of motion of a charged particle in a magnetic eld. The partial derivatives with respect to coordinates and velocities are L e = (v B)i xi 2 and L e = mvi (r B)i vi 2 (2.134) (2.133)
48 The latter gives
e d L ) = mai (v B)i ( dt vi 2 Lagranges equation, in the standard form L d L ) =0 ( dt vi xi then gives mai e(v B)i = 0 or in vector form ma = ev B
(2.135)
(2.136)
(2.137)
(2.138)
The left hand side is the well known Lorentz force which acts on a charged particle in a magnetic eld. We nd the Hamiltonian H = vpL
1 1 = v (mv + eA) mv2 + v (r B) 2 2 1 = mv 2 2 1 = (p eA)2 2m
(2.139)
and note that this is identical to the kinetic energy. This is conserved, as follows from the fact that the Lagrangian has no explicit time dependence, and the energy conservation is consistent with the fact that the magnetic force can only change the direction of the velocity but not its absolute value.
Chapter 3
Hamiltonian dynamics
3.1 Hamiltons equations
In Lagranges formulation the Lagrangian L(q, q, t) acts as a dynamical steering function of the physical system. It determines the motion of the system through its partial derivatives with respect to the variables qi and q i . Hamiltons formulation of the dynamics of a physical system can be viewed as derived from Lagranges formulation by a change of the steering function from the Lagrangian to the Hamiltonian, L(q, q, t) H (q, p, t) (3.1)
where this transformation is combined with a change of fundamental variables, from the set of generalized coordinates and velocities (q, q ), to the set of coordinates and conjugate momenta (q, p). This type of transformation is referred to as a Legendre transformation. The reason for combining the change of fundamental variables with the change in the dynamical function is that the equations of motion are expressed through the partial derivatives of this function with respect to the fundamental variables. Similar types of transformations are known from thermodynamics, where the thermodynamical variables p, T, V, S, ... are related through partial derivatives of the relevant thermodynamical potential. There is a certain freedom in the choice of fundamental and derived variables, and a change in this choice is accompanied by a change of thermodynamic potential so that the derived variables can also after the transformation be expressed through partial derivatives of the potential. To be more specic we return to the denition of the Hamiltonian H=
i
pi q i L ,
pi =
L q i
(3.2)
As already discussed, we may invert the relation between the conjugate momentum and velocity in the expression for pi , to give the velocity as a function of momentum and coordinates (and possibly time), q i = q i (p, q, t) (3.3)
Thereby we may express the Hamiltonian as a function of q, p and t. To see how Lagranges equation can be reformulated in terms of partial derivatives of H , we consider rst the variation in H under an 49
50
CHAPTER 3. HAMILTONIAN DYNAMICS
innitesimal change in the variables of the system. From the denition of H follows dH =
i
(dpi q i + pi dq i ) dL (dpi q i + pi dq i ) (pi L L L dqi dq i dt qi q i t
=
i
=
i
L L L )dq i + q i dpi dqi dt q i qi t L L dqi dt qi t (3.4)
=
i
q i dpi
and the important point to notice is that the differential dq i has disappeared in the nal expression due to the denition of the canonical momentum pi . This means that only the differentials for a set of independent variables (q, p) appear on the right-hand-side of the equation. The coefcients in front of these can be interpreted as partial derivatives of H with respect to the corresponding variables. With H as a function of q , p and t, the general expression for the change in H due to a change in the fundamental variables is dH =
i
H dpi + pi
H H dqi + dt qi t
(3.5)
and by comparing with (3.4), we nd the following relations q i = H , pi H L = , qi qi L H = t t (3.6)
One should note that, at this point, no dynamics is involved in these equations. They are simply consequences of the denitions of the canonical momenta and the Hamiltonian. However, at the next step we make use of Lagranges equation, which may be written as p i = L qi (3.7)
By use of this the two rst of the above equations (3.6) can be written as q i = H , pi p i = H qi (3.8)
These equations, which are known as Hamiltons equations, can be viewed as equivalent to Lagranges equations, in the sense that they constitute a complete set of equations of motion for the physical system. As already shown, Hamiltons equations follow from Lagranges equations, and in a similar way one can from Hamiltons equations re-derive Lagranges equation. Hamiltons equations (3.8) can be supplemented by a third equation H dH = dt t (3.9)
This identity follows from (3.4) by use of Hamiltons equations for q and p . This shows directly that H L if there is no explicit time dependence, which means t = t = 0, then the total time derivative of H vanishes and therefore the Hamitonian is a constant of motion.
3.1. HAMILTONS EQUATIONS
51
In the derivation of Hamiltons equation it was noticed that only the equations for p i were dynamical, in the sense that only these equations depended on Lagranges equation to be satised. However, after Hamiltons equation have been established, there is no reason for treating the equations for q and p differently. The standard way to view the equations is that both equations are parts of the full set of equations of motion for the system, with the coordinates and momenta being represented in a symmetric way. Compared to Lagranges formulation, it seems that we have doubled the set of equations, since now there are two equations for each degree of freedom, whereas in Lagranges formulation there is only one. However, the two Hamiltons equations are rst order in time derivatives, whereas Lagranges equation is second order. The two rst order differential equations can be replaced by a single second order differential equation, and we shall demonstrate this in a simple example.
3.1.1
Example: The one-dimensional harmonic oscillator
In this case there is no constraint (except for the reduction to one dimension) and we use the linear coordinate x as generalized coordinate. For kinetic and potential energy we have the expressions 1 T = mx 2 , 2 The Lagrangian is therefore 1 1 2 kx2 L = mx 2 2 and the canonical momentum conjugate to x is p= The Hamiltonian is dened by H = px L= and from this follows Hamiltons equations H p = p m H p = = kx x Position and momentum are therefore coupled through the two equations p , p = kx x = m From these equations p can be eliminated to give x = x + k x=0 m 1 2 1 2 p + kx = T + V 2m 2 (3.13) L = mx x (3.11) 1 V = kx2 2 (3.10)
(3.12)
(3.14)
(3.15)
(3.16)
which is the standard harmonic oscillator equation, with = k/m as the circular frequency of the oscillator. This is the equation we would have derived directly from the Lagrangian through Lagranges equation, and the reduction from the two Hamiltons equations to the single Lagranges equation has been obtained by eliminating the momentum p. Although this is a very simple example, it illustrates the way in which Hamiltons equations are used, and how these equations relate to Lagranges equations.
52
3.2
Phase space
At an earlier stage we introduced the conguration space of the physical system as the d dimensional space described by the generalized coordinates q = (q1 , q2 , ..., qd ). These d coordinates, one for each degree of freedom of the system, are all independent variables. Later, in the discussion of the Lagrangian formulation, we extended this set to a larger set of 2d variables, by treating the velocities q = (q1 , q 2 , ..., q d ) as independent of the coordinates. To treat the velocities as independent of the coordinates may initially look strange, especially when they are expressed as the time derivatives of the coordinates. However, if instead of focussing on the time evolution of the coordinates q (t) for a given trajectory in conguration space, we take q to mean a possible conguration of the system at a given instant, then the coordinates do not determine the velocities at the same instant. As an example, assume a particle has a position r at a given time t. The velocity v at the same time is not determined by the position, and can therefore be treated as an independent variable. This simply means that many particle orbits, with different velocities, may pass through the same point in space. To vary v with r xed then means to change from one orbit to another. In this sense coordinates and velocities of the Lagrangian may be treated as independent. Only after Lagranges equations has been solved, with given initial conditions, will the coordinates and velocities be linked together to determine a unique trajectory in conguration space. In Hamiltons formulation coordinates and momenta are treated on equal footing. Therefore the d dimensional conguration space seems less important than the 2d dimensional phase space. This space is the one where coordinates and velocities are all treated as independent variables. However, more commonly than using coordinates and velocities, one takes coordinates and momenta as the independent variables in phase space, since these are the standard variables in Hamiltons equations.
(q(t), p(t)) q2 q(t) p
q1 a) b)
Figure 3.1: Motion in conguration space a) and phase space b). In conguration space the trajectory is determined by the time dependent coordinates q (t), and many different trajectories, with different initial conditions, may pass through the same point. In phase space the trajectory is specied by the time dependent coordinates and momenta (q (t), p(t)). In this case only one trajectory will pass through a given point, and all dynamical trajectories (those that satisfy the equations of motion) will together form a ow pattern through phase space.
One of the interesting features of the phase space description becomes apparent when one considers the time evolution with given initial conditions. We know that 2d initial data are needed to give a unique trajectory. In the Lagrangian formulation this is because there are d second order differential equations to determine the motion, and in the Hamiltonian formulation since there are 2d rst order equations. In conguration space this means that through a given point (determined by the d coordi-
3.2. PHASE SPACE
53
nates) there are many possible trajectories, as we have already discussed. However, in phase space the number of coordinates needed to determine a point is 2d and that is also the number of data needed to determine uniquely a trajectory. This means that through a point in phase space there will pass only one dynamical trajectory (i.e., a trajectory that satises the equations of motion). This situation is illustrated in Fig. 3.1 for the case of a two-dimensional phase space. Through each point passes one and only one trajectory, specied by the initial conditions. If we continuously change these conditions, the trajectory will be deformed in such a way that, when we consider all possible motions of the system at the same time, the trajectories will form a ow pattern through phase space. These paths will be distinct, so that two paths will never cross (except at some singular, isolated points, which we shall discuss in an example to follow). This description of the dynamics, as a ow pattern in phase space, is particularly important in statistical mechanics, where one does not consider sharply dened initial conditions but rather a time evolution of the system with a statistical distribution over many initial data. As we shall see in examples, the phase space description is also sometimes useful to obtain a qualitative understanding of the motion of the system, without actually solving the equations of motion. Thus, if we nd the special points of the ow, corresponding to points of equilibrium, and use the general properties of the phase space ow, we can derive a good qualitative picture of the full ow pattern, and thereby the motion of the system.
3.2.1
Examples
Phase space for the harmonic oscillator We write the Hamiltonian of a one-dimensional harmonic oscillator in the following form H= 1 2 (p + m2 2 x2 ) 2m (3.17)
with as the circular frequency of the oscillator. Since the Hamiltonian has no explicit time dependence, the total time dependence of H vanishes dH H = =0 dt t The energy H = E is therefore a constant of motion. This implies p2 + m2 2 x2 = 2mE (3.19) (3.18)
and we recognize this as the equation for an ellipse in the two-dimensional plane with x and p as coordinates, which is the phase space of the harmonic oscillator. Since x and p have different physical dimensions, the eccentricity of the ellipse has no physical signicance, and we can rescale one of the coordinates, for example by redening the x coordinate, x = mx (which gives x it the same physical dimension as p), so that the ellipse becomes a circle, p2 + x 2 = 2mE (3.20)
The radius of the circle is determined by the energy and increases as E with energy. Since the energy is a constant of motion these circles of constant energy are the trajectories of the harmonic oscillator in phase space.
54
p
Figure 3.2: Phase space ow for the one dimensional harmonic oscillator. The time evolution dene circles
of constant energy with motion in the clockwise direction. The curves of constant energy are here plotted with constant energy differences.
We further have Hamiltons equations = m H = p x p H = x p = x
(3.21)
which show that the system moves in the clockwise direction along a circle of constant energy. The initial conditions determine the energy and thereby the circle which the oscillator follows. We may consider the Hamiltonian H (x, p) as dening a phase space potential. Hamiltons equations show that the system moves in the direction orthogonal to the gradient of the potential, which means motion along one of the equipotential curves. As illustrated in Fig. 3.2 these (directed) curves of constant energy determine the phase space ow of the harmonic oscillator.
The pendulum Let us next consider the phase space motion of a planar pendulum. With l as the length of the pendulum rod, m as the mass of the pendulum bob, and the angle of displacement chosen as the generalized coordinate, we nd the following expression for the Lagrangian 1 2 + mgl cos L = ml2 2 The canonical momentum conjugate to is p= L = ml2 q (3.23) (3.22)
3.2. PHASE SPACE
55
and we nd the following expression for the Hamiltonian L H = p 1 2 2 = ml mgl cos 2 p2 = mgl cos 2ml2
(3.24)
Again there is no explicit time dependence, which means that the energy H = E is a constant of motion. From this follows that a trajectory of the pendulum in phase space is given by p2 + 4m2 gl3 sin2 For small oscillations it simplies to p2 + m2 gl3 2 = 2ml2 E (3.26) = 2ml2 E 2 (3.25)
It has the same form as the phase space equation of the harmonic oscillator, which we have already discussed, although the coordinates are different. In the present case p has the dimension of angular momentum rather than linear momentum, and is is a dimensionless variable. But that is not important for the phase space motion, and by a proper scaling of the variables it can also here be given the form of equation of a circle, with radius determined by the energy, 2 = 2ml2 E , p2 +
p
= m
gl3
(3.27)
-2
Figure 3.3: Phase space ow for the pendulum. There are two types of motion, where the closed curves
represent oscillations of the pendulum about the stable equilibrium and the open curves represent full rotations. The dashed curves are limit curves that separate the two types of motion. The singular crossing points of these curves are the points of unstable equilibrium. They are not real crossing points of the particle trajectories, since the pendulum velocity at these points vanishes.
When we include motion also for larger angles, we rst note that that the Hamiltonian H (p, ) is a periodic function of , and the equipotential curves in the , p-plane therefore will show a periodic
56
behavior under a shift + 2 . Therefore the point of stable equilibrium will be periodically repeated at angles (, p) = (2n, 0) with n as an integer. If we increase the energy and therefore the amplitude of oscillations, the motion is represented by circles of increasing radii around each point of stable equilibrium. Due to the periodic structure these closed curves will necessarily get deformed for sufciently large amplitudes, and at some point there is a singular situation is reached when the closed curves belonging to neighboring equilibrium points will touch. This we interpret as corresponding to the situation where the pendulum reaches the upper point of unstable equilibrium. If the energy is increased even further, the motion is not bounded in the angular variable, but describes full rotations in the angle. This qualitative picture is in full agreement with the plot of phase space trajectories shown in the gure. There are solutions of bounded motion, corresponding to oscillations of the pendulum around the point of stable equilibrium, but there are also solutions of unbounded motion. The transition between these two different types of motion is represented by equipotential curves that intersect in singular points. These represent the point of unstable equilibrium, with the pendulum rod at rest in an upright vertical position. We see from this discussion that we can reach a rather complete, qualitative understanding of the phase space motion by using the knowledge of what happens for small oscillations together with implications of periodicity of the motion.
3.3
Hamiltons equations for a charged particle in an electromagnetic eld
We have in an earlier section established the form of the Lagrangian for a charged particle in an electromagnetic eld 1 L = mv2 e + ev A 2 (3.28)
with and A as the electromagnetic potentials, m as the mass and e as the charge of the particle. The corresponding canonical momentum is p = mv + eA and the Hamiltonian is H= 1 (p eA)2 + e 2m (3.30) (3.29)
This classical Hamiltonian has the same form as its quantum counterpart, and it represents the total energy of the system. If the potentials are time independent, the Hamiltonian H is also time independent and the energy is conserved. We take the Cartesian coordinates of the particle as generalized coordinates, and write these as xi , i = 1, 2, 3, with x1 = x, x2 = y and x3 = z in the usual way. Hamiltons equations in this case give x i = p i H 1 = (pi eAi ) pi m H e A = = (p eA) e xi m xi xi
(3.31)
3.3. HAMILTONS EQUATIONS FOR A CHARGED PARTICLE IN AN ELECTROMAGNETIC FIELD57 We will check that these two equations reproduce the well known form of Newtons second law applied to the charged particle in the electromagnetic eld. We do this by eliminating p from the equations, mx i = p i e dAi dt
j
= p i e( = e m
Ai Ai x j + ) xj t Aj e e( xi xi (
j
(pj eAj )
Ai Ai x j + ) xj t (3.32)
= e(
Ai + )+e xi t
Aj Ai )x j] xi xj
This we can write in a more familiar form by use of the expressions for the electric and magnetic elds Ei = ( with
ijk
Ai + ), xi t
Bk =
ij
kij
Aj xi
(3.33)
as the antisymmetric Levi-Civita symbol. The last equation can be inverted to give Aj Ai = xi xj
kij Bk k
(3.34)
and therefore the equation of motion (3.32) can be written as mx i = eEi + e

jk
j ijk Bk x
(3.35)
In vector form it gives the standard (non-relativistic) equation of motion for a charge particle in the electromagnetic eld, ma = e(E + v B) (3.36)
This again demonstrates that Hamiltons (as well as Langranges) equations have a different form, but are equivalent to Newtons second law when applied to the same system. We shall next see how Hamiltons equations can be used in a direct way to nd the motion of a charged particle in a constant magnetic eld.
3.3.1
Example: Charged particle in a constant magnetic eld
We assume the particle to be moving in a constant magnetic eld with direction along the z -axis, B = B k. The vector potential we write as 1 A= rB 2 with components 1 Ax = eBy , 2 1 Ay = eBx , 2 Az = 0 (3.38) (3.37)
58
It is straight forward to check that the curl of this vector potential reproduces the correct magnetic eld. The scalar potential vanishes, = 0. The Hamiltonian (3.30) then gets the form H= 1 1 1 1 (p eA)2 = [(px + eBy )2 + (py eBx)2 + p2 z] 2m 2m 2 2 (3.39)
We note that z is a cyclic coordinate1 , and it follows directly from Hamiltons equations that p z = H =0 z (3.40)
so that pz = mz is a constant of motion. Thus, the motion in the z -direction has the simple form of motion with constant velocity z = z 0 + vz 0 t (3.41)
with vz 0 = pz /m and z0 as constants determined by the initial conditions. From this follows that the motion in the x, y -plane (the plane orthogonal to the magnetic eld) is decoupled from the motion in the z -direction. We write Hamiltons equations for this motion, 1 1 H = (px + eBy ) px m 2 H eB 1 p x = = (py eBx) x 2m 2 H 1 1 = (py eBx) y = py m 2 H eB 1 p y = = (px + eBy ) y 2m 2 x =
(3.42)
By inspecting the right-hand-side of the equations we see that they can be grouped in pairs that are essentially identical. By combining these the following equations are established, 1 p x eB y = 0 2 1 p y + eB x = 0 2 which means that there are two constants of motion 1 Kx = px eBy 2 1 Ky = py + eBx 2 Combined into a vector, this vector is 1 K = p er B 2 1 = mv + eA er B 2 = mv er B
1
(3.43)
(3.44)
(3.45)
If H does not depend on z , it is clear from the denition of the Hamiltonian that also the Lagrangian is independent of
z.
3.3. HAMILTONS EQUATIONS FOR A CHARGED PARTICLE IN AN ELECTROMAGNETIC FIELD59 and it is easy to verify directly from the equation of motion (3.36) that this vector is conserved. We consider next the linear combinations of the equations (3.42) with opposite signs of those in (3.43), 1 eB 1 p x + eB y = (py eBx) 2 m 2 eB 1 1 = (px + eBy ) p y eB x 2 m 2 1 mv = p eA = p + er B 2 They get the form x = y eB y m eB = x m y + 2 y = 0
(3.46)
These equations can be expressed in terms of components of the mechanical momentum vector (3.47)
(3.48)
which implies that each component satises a harmonic oscillator equation x + 2 x = 0 , (3.49)
with = eB/m as the circular frequency. This is known as the cyclotron frequency. The solutions to the equations have the form x = A cos t , y = A sin t (3.50)
where A is a constant to be determined by the initial conditions, and where a convenient choice of time t = 0 has been chosen. These expressions may be combined with the expressions for the components of the conserved vector K, and we focus rst on the x-component, 1 px + eBy = A cos t , 2 By combining these we nd y = 1 (A cos t Kx ) eB y0 + R cos t Kx , eB A eB 1 px eBy = Kx 2 (3.51)
(3.52)
where, in the last expression, we have introduced the constants y0 = Similarly we have 1 py eBx = A cos t , 2 which gives x = 1 (A sin t + Ky ) eB x0 + R sin t 1 py + eBx = Ky 2 (3.54) R= (3.53)
(3.55)
60
The solutions for the components of the position vector show that the particle moves with constant speed on a circle of radius R about a point in the x, y -plane with coordinates (x0 , y0 ). These coordinates, as well as the radius R are determine by the initial conditions. The circular frequency = eB/m is xed by the strength of the magnetic eld and the charge, and is independent of the energy of the particle. The direction of circulation in the circular orbit is determined by the sign of eB , so that negative sign corresponds to positive orientation of the motion in the x, y -plane. When the circular motion in the x, y -plane is combined with the linear motion along the z -axis, this gives a spiral formed orbit of the particle around a magnetic ux line. The radius of the circular part is determined by the contribution to the kinetic energy of the particle from the motion in the x, y plane, 1 T = m 2 R2 2 (3.56)
A well known realization of this type of motion is for electrons and protons in the magnetic eld of the earth. For these particles there is an additional effect, which is due to the convergence of the magnetic eld lines towards the magnetic poles. This convergence induces a slow down of the component of the motion along the lines and, eventually a reversal of the motion. In this way the electrons may be trapped in a spiral like motion between the two poles with points of reection above the atmosphere. The van Allen radiation belts are formed by charged particles from the sun, which are captured in this type of orbits.
3.4
Calculus of variation and Hamiltons principle
The motion in the conguration space of a physical system is described by the time dependent generalized coordinates q (t). A specic time evolution may be determined by solving the equations of motion with initial conditions specied by the coordinates q (t0 ) and velocities q (t0 ) for a given initial time t = t0 . For a d dimensional conguration space, these 2d initial data uniquely species the evolution of the system. However, the solution may be specied also in other ways, in particular by xing the coordinates at two different times, q (t1 ) and q (t2 ). Again such a set of 2d data will specify a unique solution2 . Even if the two ways to specify a solution, either by initial data at a single time t0 or by endpoint data at two different times t1 and t2 , are equivalent, they may give rise to different points of view concerning the dynamics of the system. We consider the following problem motivated by choosing the latter type of boundary conditions: When considering all possible paths q = q (t) that satisfy the boundary conditions q (t1 ) = q1 and q (t2 ) = q2 , with q1 and q2 as two given sets of coordinates, what characterizes the dynamical path (the one that satises the equations of motion), in comparison to other continuous paths between the given end points? Hamilton formulated an answer to this question in the form of a variational problem, called Hamiltons principle. The principle is formulated by use of the action integral of paths between the end points. The denition of the action is
t2
S [q (t)] =
t1
2
L(q (t), q (t), t)dt
(3.57)
In exceptional cases there may be more than one solution.
3.4. CALCULUS OF VARIATION AND HAMILTONS PRINCIPLE
61
It is well dened for any continuous, differentiable path q (t) between the end points, not only the one that satises the equation of motion. The action is a functional of the path, which means that it is a function of the function q (t). Hamiltons principle refers to variations in the value of the action S [q (t)] under small variations in in the path q (t): The path q (t) between the xed end points q (t1 ) = q1 and q (t2 ) = q2 , which describes the dynamical evolution of the physical system, is characterized by the action being stationary under small variations in the path, q (t) q (t) + q (t), with q (t1 ) = q (t2 ) = 0. We write the condition as S = 0 (3.58)
where the meaning of this equation is that the change in S vanishes to rst order in the variation q (t), for the path q (t) that is followed by the system between the the specied initial and nal points. We may say that Hamiltons principle expresses a global view on the evolution of the system in conguration space, with the correct, dynamical path being specied as the solution of a variational problem. Lagranges equations, on the other hand, gives a local condition for the dynamical evolution, in the form of a differential equation that should be satised at any time t during the evolution. These two ways of describing the motion of the system are not in conict, but are instead equivalent, as we shall demonstrate. In order to show that Lagranges equations and Hamiltons principle are two equivalent ways to describe the dynamics of the system, we examine how the change in the action S for a small variation in the coordinates around a given path can be expressed in terms of the Lagrangian. To rst order in the variations in the coordinates we have
t2
S =
t1 t2
L(q (t), q (t), t)dt (

t1 k
L L qk + q k )dt qk q k
(3.59)
The integral can be manipulated in the following way

t2
S =
t1 t2 k
L d L d L qk + ( qk ) ( )qk ]dt qk dt q k dt q k L qk q k
t2
=
t1 t2 k
L d L [ qk ( )qk ]dt + qk dt q k [ L d L ( )]qk dt qk dt q k
t1
=
t1 k
(3.60)
where in the last step we have used the condition that the end points should be xed during the variations in the coordinates, so that q (t1 ) = q2 = 0. The expression we have derived for the change in the action shows that S indeed vanishes to rst order in variations of the coordinates for a path that satises Lagranges equations. We note that the implication also works the other way, in the sense that if S vanishes for arbitrary variations in the generalized coordinates, this implies that Lagranges equations have to be satised for the path q (t). As pointed out, Hamiltons principle gives an interesting, different view on the evolution of the system. It is a global view on the dynamical path in conguration space, and this view may add
62
something interesting to the understanding of the evolution of the system. However, in most cases, the equations of motion, expressed in Lagranges or Hamiltons form, will give the most convenient way to actually determining the time evolution of the system. Variational problems are met in many elds of physics. It is interesting to note that the relation we have discussed between Hamiltons principle and Lagrangess equations may be useful for such problems more generally. Consider a problem where a physical quantity should be stationary under variation in some variables (typically a minimum or maximum problem), and where this quantity can be expressed as an integral, similar to the action integral S = Ldt. In that case there is a relation between the variational problem and the set of differential equations, which correspond to Lagranges equations. This reformulation of the variational problem as a set of differential equations may be useful for solving the problem, and we shall next illustrate this by an example where a variational problem is solved in this way.
3.4.1
Example: Rotational surface with a minimal area
We consider the following problem: Two points (x1 , y1 ) and (x2 , y2 ) in a the x, y plane are selected. We want to determine the curve y (x) in the plane which links the two points and which gives rise to a surface of minimal area when the curve is rotated in 3-dimensional space around the x axis. This is a typical variational problem where we want to determine a curve y (x) with xed endpoints y (x1 ) = x1 , The area to be minimized can be written as
x2
y (x2 ) = x2
(3.61)
A[y (x)] =
x1
2y
1 + y 2 dx
(3.62)
where we have used the notation y = dy/dx. This expression for the area is found by considering the contribution from an innitesimal section of width dx in the x direction, dA = 2y dx2 + dy 2 = 2y 1 + y 2 dx (3.63)
and then integrate this along the x axis. The variational problem can be written as A = 0 (3.64)
for variations y (x) with y (x1 ) = y (x2 ) = 0. The problem is seen to be of precisely the same form as in Hamiltons principle although the variables are different and the interpretation of the problem also. To exploit the formal correspondence we write the area functional as
x2
A = 2
x1
L(y, y )dx
(3.65)
with L(y, y ) as the function corresponding to the Lagrangian. We note that here x has taken the place of t in Hamiltons principle, and y has taken the place of q with y as the equivalent of a generalized coordinate. (For convenience we have pulled out the constant factor 2 .) The correspondence makes
63
it easy to write the differential equation that is equivalent to the variational problem. It has the form of Lagranges equation d L L ) ( =0 dx y y We calculate the partial derivatives, L = y and get the differential equation d dx yy 1+y2 1+y2 =0 (3.68) 1+y2, L = y yy 1+y2 (3.67) (3.66)
By doing the differentiation with respect to x and simplifying the equation we get yy y 2 = 1 (3.69)
which is a non-linear differential equation that is second order in derivatives. Usually a non-linear differential cannot be solved by analytic methods, but in the present case it can. We will in this case make a complete discussion of the problem by showing how to solve the differential equaton. In order to do so we change to a new variable u in the following way u= This gives u = 1 (yy y 2 ) y2 (3.71) y y (3.70)
By applying the equation (3.69) which y should satisfy, we nd u = which gives u = 2 y = 2uu y3 (3.73) 1 y2 (3.72)
This means that u should satisfy the differential equation u + 2uu = 0 (3.74)
Since the expression on the left-hand side can be written as a derivative with respect to x, the equation can immediately be integrated once to give u + u2 = k 2 (3.75)
where k is a constant. (Note that we can write the integration constant in (3.74) as a positive constant k 2 , since Eq.(3.72) shows that u is positive.)
64
We have now a rst order differential equation to solve, and we do this by integrating the equation in the following way, k2 u =1 u2 k2 du =x+C u2 (3.76)
with C as an unspecied integration constant. The integral, which determines u as a function of x can be solved, and we do this by the following substitution (the result is also listed in standard integration tables) u = k tanh w By differentiating the expression we nd du = and by combining this with k 2 u2 = k 2 (1 tanh2 w) = we nd du cosh2 w k 1 = dw = dw 2 2 2 2 k u k k cosh w This means that the integral in (3.76) is reduced to the simple form dw = k (x + C ) with solution w = kx + w0 where w0 is an integration constant. The expression for u is then found to be u = k tanh(kx + w0 ) with derivative u = For y this nally gives the solution y= 1 cosh(kx w0 ) k (3.85) k2 1 = 2 2 y cosh (kx + w0 ) (3.84) (3.83) (3.82) (3.81) (3.80) k2 cosh2 w (3.79) k dw cosh2 w (3.78) (3.77)
where the two integration constants k and w0 are (implicitly) determined from the boundary conditions, y (x1 ) = y1 , y (x2 ) = x2 ,
cosh(kx1 w0 ) = ky1 ,
cosh(kx2 w0 ) = ky2
(3.86)

(x1,y1) y (x2,y2)
65
Figure 3.4: Minimal rotational surface derived from the variational problem, here shown in blue. The yellow
curves indicates the collapsed surface, which may have even a smaller area .
The above expressions solve the variational problem. However, some further comments may be appropriate. In the case of Hamiltons principle we note that any solution to the variational problem gives a solution of the equations of motion. It is not important whether the solution corresponds to a minimum, a maximum or a saddle point of the action. In the present case, on the other hand, we are specically interested in nding the minimum. By nding the variation in the area for innitesimal variations in the function, y (x), calculated to second order, we can decide whether the solution we have found is a local minimum. This is similar to deciding whether a function has a minimum in a point where the derivative vanishes, by calculating and checking the sign of the second derivative of the function. It is straight forward to check in this way that the solution we have found is in fact a local minimum. Another question is if we have found the global minimum. In fact, it is almost obvious that it is so only when the two points (x1 , y1 ) and (x2 , y2 are not too far apart, in the sense that (x2 x1 ) is not too large compared to y1 and y2 . Therefore, if we separate x1 and x2 with y1 and y2 xed it is clear that the area of the surface which is generated by the curve we have found will increase with the separation between the two points. At some point it will become preferable to collapse the surface in the following way: Close to each boundary point the curve y (x) falls abruptly to 0, and between the two points differs only innitesimally from 0, to form a narrow cylinder of vanishing area. Such 2 + y 2 ) independent of the distance between x and x . The a surface will have the area A = (y1 1 2 2 reason we do not see this surface in our analysis is that it corresponds to a curve in the x, y plane that is not differentiable. It can in this sense be excluded, but the point is that close to this curve there are differentiable curves with almost the same area. (The situation is similar to the one when we search a minimum of a function in a bounded region. In the interior of the region a (local) minimum is characterized by the derivative of the function being zero, but for a minimum on the boundary that does not need to be the case.) From this we conclude that the minimum we have found is a global minimum only when the area satises
2 2 A (y1 + y2 )
(3.87)
We do not go further in examining this point which is specic for the present example. It is interesting to note that the minimization problem we have discussed in this example has a
66
simple physical application. It is well known that due to the surface tension a soap lm will make a minimum surface area with the given boundary conditions for the lm. If we therefore attach the soap lm to two circular hoops that are positioned symmetrically about an axis, we will have created a situation like the one discussed in the example. According to the analysis we have made the lm should make a surface similar to the one shown in the gure. For physical reasons it seems also clear that if the distance between the two hoops increases, at some points the surface will tuch itself somewhere in the middle and it will collapse to two independent surfaces that cover each of the two hoops.
Summary
We have in this part of the lectures discussed some of the basic elements of analytical mechanics. The focus has been on how to dene a set of independent, generalized coordinates q that describe the physical degrees of freedom of the system, and to use these in a reformulation of the equations of motion. A main motivation for introducing the generalized coordinates is to eliminate from the description the explicit reference to constraints, and thereby to the corresponding (unknown) constraint forces. The types of motion that are consistent with the constraints at a xed time t, are referred to as virtual displacements. They correspond to changes q in the generalized coordinates with t xed. Application of Newtons second law, combined with virtual displacements of the system, allows a reformulation of the dynamics in a form which only refers to time evolution of the generalized coordinates. Two equivalent form of this dynamics are dened by Lagranges and Hamiltons equations. Lagranges equations is a set of differential equations that primarily determines the motion in conguration space, q (t). Hamiltons equations on the other hand treat the generalized coordinates q and their conjugate momenta p on equal footing and therefore determine primarily the motion in phase space, (q (t), p(t)). One of the advantages of Lagranges and Hamiltons formulations is that they specify the dynamics in a compact form through a scalar function, either the Lagrangian or the Hamiltonian. They further give an explicit scheme to follow when analyzing the physical system, where only the physical degrees of freedom participate. In addition, symmetries of the system that are represented as symmetries of the Lagrangian (and Hamiltonian) can be directly exploited to derive constants of motion and thereby effectively to reduce the number of independent variables of the system. Hamiltons principle gives a description of the dynamics that is in a sense complementary to that of Lagranges and Hamiltons equation. It is a variational principle which selects the path dened by the dynamical evolution between two xed end points q (t1 ) and q (t2 ) in conguration space. This gives a global view on the time evolution which is however equivalent to the local view given by the differential equations of Lagrange and Hamilton. The theory discussed here gives the basis for other related formulations of the dynamics of physical systems. Let me mention some of those that are not discussed in these notes. One important generalization of the theory is to the Lagrangian description of classical (and quantum) elds. In that case the continuous eld variables replace the discrete generalized coordinates of mechanical systems, and in modern eld theory this formulation is almost indispensable. Also for systems with discrete variables there are important generalizations. There is an underlying mathematical structure of Hamiltonian systems that is referred to as a symplectic structure. This can be rened and further developed by use of algebraic relations known as Poisson brackets. Furthermore, the phase space ow discussed in the lectures can be extended to a the uid like description of physical systems known as the Hamilton-Jacobi theory. The phase space description also has extensions to the description of nonlinear systems, where a richer set of physical phenomena can be found than in the linear differential equations of Lagrange and Hamilton. 67
68
The theoretical reformulations of classical mechanics mentioned above also give physics a form that lies close to the formulations of quantum mechanics. That is seen clearly in the fact that many of the central objects of the classical theory, like the Hamiltonian and the conjugate coordinate coordinates and momenta, are also central objects in the quantum description, although with a reinterpretation of these as Hilbert space variables. Other correspondences are also close, for example between the Hamilton-Jacobi theory and Schr odingers wave mechanics and between Hamiltons principle and Feynmans path integral description of quantum mechanics. This underscores the point mentioned in the introduction, that the formulations of Lagrange and Hamilton continues to hold a central position in modern theoretical physics.
Part II
Relativity
69
Introduction
At beginning of the last century one of the fundamental unsolved problems in physics was how to reconcile Maxwells equations of electromagnetism with the old principle of physics which we now refer to as Galilean relativity. On one hand Maxwells unication of the electromagnetic phenomena had been a great success, on the other hand Galileis observation that the laws of nature were the same in all inertial frames (reference frames that move with constant velocity relative to each other), was supported by experiments and observations over centuries. The problem was that Maxwells equations contained a constant with physical dimension of velocity, and the presence of such a constant was incompatible with the Galilean principle. A way out of this dilemma seemed to be to assume that the fundamental laws of nature were indeed the same in all inertial frames, but a special kind of medium was present in empty space, called the illumines aether. The electromagnetic phenomena, in particular wave propagation, should then be physical phenomena taking place in this medium, much like the propagation of waves in water. If this picture was correct Maxwells equations would not really be fundamental, they would strictly be correct only in a particular inertial frame, the rest frame of the aether. However, problems remained concerning the somewhat mysterious aether. It should ll the whole universe and it should have rather peculiar mechanical properties, but the most important problem was that there should be measurable corrections to Maxwells equation in reference frames that moved relative to the aether. Michelson and Morley unsuccessfully tried to nd such effects experimentally. The idea was that the earth could not at all times be at rest with respect to the aether, because of its orbital motion about the sun. If it at a particular time of the year was at rest with respect to the aether, it should half a year later have its maximal relative velocity. Measurements of the velocity of light at different times of the year did not show even tiny variations in the velocity. In 1905 Albert Einstein offered a solution to the problem that made the discussion about the illumines aether completely irrelevant. He insisted on the fundamental character of Maxwells equations and at the same time he upheld the idea of all inertial systems to be equivalent with respect to the fundamental laws of nature. His way of making this possible was to change the relations between coordinates and velocities as measured in different inertial frames. He introduced a new description of space and time by assuming the Lorentz transformations to give the correct transformations between inertial frames. These transformations were not new, at the mathematical level they had been identied and discussed as symmetry transformations of Maxwells equations by Larmor, Lorentz and Poincare. But the fundamental character of the transformations had not been realized. Einsteins idea was indeed revolutionary. It changed the perspective on space and time since the transformation formula showed that space and time were not independent concepts. The idea about the larger space-time emerged, where a distinction between space and time is not universal, but will change from one inertial frame to another. This idea had important implications, as Einstein showed. The length contraction and time dilatation of moving bodies are well known consequences, and also the relativistic relation between mass and energy. But the impact was deeper, since the principle of 71
72 relativity should apply to all physical laws, and all physical laws therefore should in some way reect the new relation between space and time. In these lectures we study some of the basic elements of Einsteins special theory of relativity. Our starting point is the Lorentz transformations, which dene the fundamental relations between coordinates and velocities in different inertial frames. We derive from these important kinematical relations such as length contraction and time dilatation and also the relation between relativistic mass and energy. We further discuss relativistic dynamics, where the principle of relativity is used to guide us in how to bring Newtons equations into relativistic form. Our approach will be to introduce and to make use of the natural formalism for theories where space and time are treated on the same footing. This is the four-vector formalism where vectors in three-dimensional space are replaced with vectors in four-dimensional space-time. With the use of four-vectors (and their relatives - the relativistic tensors) the physical laws can be expressed in covariant form, a form which is explicitly invariant under transitions between inertial frames. This formalism may initially appear somewhat cumbersome, but applications show that it is useful, and if one goes deeper into relativistic theory than we do in this course it becomes indispensable. In addition to working with equations we will make extensive use of Minkowski diagrams to illustrate the space time physics.
Chapter 4
The four-dimensional space-time

Space and time set the scene for the physical phenomena. To describe the phenomena we apply space and time coordinates, and these coordinates depend on our choice of reference frame. Such a reference frame we may view as a physical object which position and velocity are measured relative to, but in theoretical considerations we usually replace this object by an imagined frame with axes that dene the origin and orientation of our reference system. A specic set of reference frames are the inertial frames which we may characterize as being non-accelerated.1 We begin the description of the relativistic view of four-dimensional space-time by considering the coordinate transformation formulas between inertial frames, both in Galilean physics and in the theory of relativity.
4.1
Lorentz transformations
Let us for simplicity assume that all motion is restricted to one direction, which we take as the xdirection in a Cartesian coordinate system. The Galilean transformation between to inertial frames with relative velocity v is then given by x = x vt, y = y, z =z (t = t) (4.1)
These are the transformations used in all introductory physics (and also in every day life), and to specify that the time coordinate is the same in the two reference frames seems almost unnecessary. Assume now a small body moves with velocity u relative to the rst reference frame (called S ) and velocity u relative to the second reference frame (S ), so that u= dx , dt u = dx dt (4.2)
that gives us the standard velocity transformation formula u =uv (4.3)
The transition from one inertial system to another means simply to correct the velocities by adding or subtracting the relative velocity of the two reference systems. This situation is illustrated in Fig.4.1
Note that even if position and velocity have no absolute meaning acceleration is different. An object far from the inuence of any other object will have zero acceleration, and that denes a reference value for acceleration, both in Galilean physics and in Einsteins special relativity. However, in Einsteins general relativity this is changed, and even acceleration is no longer absolute, since gravitational effects and effects of acceleration are intermixed.
1
73
74
CHAPTER 4. THE FOUR-DIMENSIONAL SPACE-TIME
with the two sets of orthogonal coordinate axes representing the inertial frames. The transformation formula clearly shows that a theory which contains a velocity as constant parameter cannot be invariant under Galilean transformations.
y y v u
S S x x
Figure 4.1: Transition from one inertial frame S to another S , here illustrated by two coordinate systems in
relative motion along the x axis. The velocity u of a particle P and and the velocity v of the reference frame S are given relative to reference frame S . The Galilean transformations give the velocities in S by subtraction of the velocity of the reference frame itself, so that the velocity of the particle in this frame is u v and the velocity of the frame S is v . In special relativity this rule for transforming velocities is no longer valid.
The Lorentz transformations, which give the correct relativistic formula for the transition between the inertial frames S and S is v x = (x vt), y = y, z = z, t = (t 2 x) (4.4) c where we have introduced the standard abbreviation 1 = (4.5) v2 1 c2 with v as the relative velocity of the two inertial frames. These transformations are not dramatically different in form from the Galilean transformations, but they are dramatically different in interpretations and in consequences. The most prominent change in the transformation formula is that the time coordinate is no longer universal, but depends on the chosen inertial frame. It is observer dependent. Another important change is that the formula contains a constant c with the dimension of velocity. It has the physical interpretation as the speed of light. However, it is clear that when the relative velocity v is small compared to the speed of light c will there be essential no difference between the Galilean and the relativistic formulas. This is seen by making an expansion in v 2 /c2 v2 + ... (4.6) 2c2 When this is introduced in the relativistic transformation formula and only the leading terms in the expansion are kept, the transformation equations reduce to the Galilean equations, as one can readily check. =1
4.2. ROTATIONS, BOOSTS AND THE INVARIANT DISTANCE
75
Let us show that an object that moves with the velocity of light will have the same velocity in all reference systems related by the Lorentz transformations (4.4). The important point is that the transformation formula for velocity is now changed. The denitions for velocity, given by (4.2), are the same, but the Lorentz transformations between the coordinates of the two inertial frames will change the relation between u and u . For an innitesimal change in the position coordinates we have dx dt and from this follows u = dx uv = dt 1 uv c2 (4.8) = (dx vdt) = (u v )dt v uv = (dt 2 dt) = (1 2 )dt c c
(4.7)
This is the new transformation formula, which is valid when the velocity u of the object is colinear with the relative velocity v of the two inertial frames. If we now set u = c in the formula it follows directly that u = c. So there is no addition of the relative velocity of the two frames in this case, and the speed of light is indeed the same in all reference frames.
4.2
Rotations, boosts and the invariant distance
The transformations discussed above are often referred to as boosts or special Lorentz transformations. Such a transformation can be viewed as taking the rst reference frame S and changing its velocity in some direction (here the x-direction) without rotating its coordinate axes, and thereby creating the new reference frame S . The general Lorentz transformations are considered as transformations that include booth boosts and rotations. There is in fact a formal resemblance between rotations and the boosts. To see this we rst consider a rotation in the x, y -plane, which in Cartesian coordinates takes the form, x y = cos x + sin y = sin x + cos y (4.9)
where is the rotation angle. The typical feature of the rotations is that the distance between two points is left invariant by the transformations. For the transformation (4.9) this invariance is expressed by s = x + y
2 2 2
= x2 + y 2 = s2
(4.10)
with x and y representing the coordinate difference between two points and s the relative distance between the points. Let us next consider the Lorentz transformations (4.4) and introduce a new parameter by the following relations2 cosh = , sinh = (4.11)
with as the standard abbreviation for the dimensionless velocity = v/c. This is a consistent parametrization, since the two expressions satisfy the requirement of hyperbolic functions, cosh2 sinh2 = 2 (1 2 ) = 1
2 1 As a reminder the hyperbolic functions are dened by cosh = 2 (e + e ) and sinh = 1 (e e ). 2
(4.12)
76
The parameter , which is related to the relative velocity v of the two reference frames by the equation v = tanh c (4.13)
is referred to as rapidity and is sometimes a more convenient parameter to use than the velocity. It is here introduced in order to give the Lorentz transformations a form similar to that of rotations. For the special transformation (4.4) it takes the form x ct = cosh x sinh ct
= sinh x + cosh ct
(4.14)
ct
ct
Figure 4.2: Comparison between a rotation in the (x, y ) plane and a boost in the (ct, x) plane. In the rst case
the rotation will transform an orthogonal coordinate frame (blue) into a rotated orthogonal frame (green). In the second case the orthogonal frame will not appear as orthogonal after the boost transformation. However, the meaning of orthogonality is in fact changed when time is introduced as a new space-time coordinate.
We note the formal similarity with the rotations (4.9), where the time coordinate ct has taken the place of the space coordinate y and the rapidity has taken the place of the angle . But is no angle, which is shown by the fact that the trigonometric functions are replaced by hyperbolic functions. The geometric difference between the two types of transformations are demonstrated in Fig.4.2. For the Lorentz transformations the distance (in three-dimensional space) between two points is no longer invariant, but another quantity, which includes also the difference in time coordinate, takes its place. For the transformation (4.14) we nd that with a new denition of s2 the following combination of relative coordinates between two space-time points is the same in the two frames, s = x c2 t = x2 c2 t2 = s2
2 2 2
(4.15)
This follows from the properties of the hyperbolic functions. We note the important change in relative sign of the two terms, compared to that of the distance in three-dimensional space. Distance in three-dimensional space has an immediate physical meaning as a measurable quantity that is independent of our choice of coordinate system. From a mathematical point of view it is natural to consider distance as a property of space itself. It denes the geometry of three-dimensional space, which we then consider as equipped with a property referred to as a metric. The metric of three-dimensional physical space is Euclidean, which means that it is geometrically a at space. The rotations we may regard as symmetry transformations of the space, which are transformations which leave the metric invariant. The Galilean transformations is an extension of these to include also time dependent transformations that leave all distances between points unchanged.
4.3. RELATIVISTIC FOUR-VECTORS
77
To change the fundamental transformations between inertial frames from the Galilean to the Lorentz transformations, implies a change in our view of space itself. The invariant metric is no longer dened by the (Euclidean) distance between points in three dimensional space, but rather by a generalized distance that involves in also the time coordinate. The expression for the generalized distance between two points in space and time (often referred to as two events) is given by s2 = r2 c2 t2 (4.16)
with the expression (4.15) already discussed as a special case. This new metric, unlike the metric in three-dimensional space, does not have an immediate, physical interpretation. It can be expressed in terms of the three-dimensional distance |r| (and the time difference t), and under certain conditions a special reference frame can be chosen where t vanishes and the four dimensional distance is identical to the three-dimensional one. But it is important to note that the metric of four-dimensional space- time, dened by the invariant (4.16) is not a Euclidean metric. We refer to this as a Minkowski metric. The important difference between the Euclidean and Minkowski metrics is that in three-dimensional space the invariant s2 is always positive, while in the four-dimensional case that is not necessary the case. Even so it is conventional to write the invariant as a square, s2 . Depending on the relative position of the two space-time points the generalized invariant (4.16) it may be positive, zero or negative. If it is positive we refer to the separation of the two space-time points as being spacelike, if it is zero the separation is called lightlike and if it is negative the separation is timelike. Since distance in three-dimensional space is the square root of s2 , this lack of positivity in four dimensions show that the change in metric strictly speaking is not simply a change in the denition of distance. The Lorentz invariance of the line element s2 = r2 c2 t2 is directly related to the fact that the speed of light is the same in all inertial frames. To see this we note that if c denotes the speed of light in a given reference frame, two space-time points on the path of a light signal through space and time will have lightlike separation, s2 = r2 c2 t2 = 0 (4.17)
Furthermore, since s2 is invariant under Lorentz transformations, if this equation is satised in one inertial frame it will be satised in all inertial frames. This means that a signal that connects the two space-time points will travel with the same speed c in all inertial reference frames.
4.3
Relativistic four-vectors
A point in three-dimensional space can be specied by a position vector, often written as r = xi + y j + z k (4.18)
with x, y and z as the Cartesian coordinates of the vector in a particularly chosen coordinate system, and with i, j and k as the unit vectors along the orthogonal coordinate axes. These vectors dene the physical three-dimensional space as a vector space. A position vector r may be considered as being independent of any choice of coordinate system in this space, however the coordinates x, y and z do depend on such a choice. This is consistent with our physical picture of a vector r in physical, threedimensional space; it has a well-dened length and direction and can be viewed as a geometrical object that exists independent of any choice of coordinate system. The coordinates are however a
78
convenient way to characterize the vector by a set of numbers, and these will then vary from one reference frame to another. Let us write the coordinate expansion in the following way,
3
r=
k=1
xk ek
(4.19)
with {ek , k = 1, 2, 3} as a set of three orthogonal unit vectors, ek el = kl A change from one set of orthogonal vectors to another, we write as a transformation ek ek = Rkl el
l
(4.20)
(4.21)
where orthogonality of the vectors means that the coefcients Rkl satisfy the condition Rki Rli = kl
i
(4.22)
This equation gives the condition for the transformation (4.21) to be a rotation. With the vector r being independent of the transformation, the change of the unit vectors ek has to be compensated by a rotation of the coordinates xk , xk xk = Rkl xl
l
(4.23)
Due to the property (4.22) of the coefcients Rkl it is straight forward to check that the combined transformation of the coordinates and unit vectors leaves the vector r unchanged. In a similar way as three-dimensional space is viewed as a three-dimensional vector space, spacetime may be described as a four-dimensional vector space. The extension from three-dimensional space to four-dimensional space-time then leads to the extension of vectors r with Cartesian coordinates (x, y, z ) to four-dimensional vectors with coordinates (x, y, z, t), where t is the time coordinate. In order to have the same physical dimension for all four directions in space-time, we introduce, in the standard way, a time coordinate with dimension of length, x0 = ct, where c is the speed of light3 Note the convention that the coordinates of space-time are written with lifted indices, so that, x0 = ct , x1 = x , x2 = y , x3 = z , (4.24)
We shall later explain the reason for this convention. To distinguish the 4-vectors of space-time from the 3-vectors of space, we shall in these notes underline the 4-vectors. In particular, the position vector of a space time point, when decomposed in Cartesian components, we may write as x = ct + xi + y j + z k (4.25)
3 For historical reasons the time component, in the form discussed here, is taken to be the 0 component rather than the 4 component. Originally a 4 component that was imaginary was introduced for time, so that x4 = ict. The reason for that was to formally give boost transformations the form of rotations. However, this convention is not so often used anymore.
4.3. RELATIVISTIC FOUR-VECTORS
79
where we have expanded the set of three unit vectors i, j and k with a fourth vector , which points in the direction of the time axis, and by underlining the unit vectors we have indicated that they are now vectors in the extended four dimensional space-time. More often we will write the expansion in the general form
3
x=
=0
x e
(4.26)
with {e } as a orthogonal set of unit vectors in four-dimensional space-time. Note that these basis vectors are written with the indices as subscript, as opposed to the coordinates where the indices are written as superscript. This is a standard convention, which means that the coordinate independent sum (4.26) appear as a sum over pairs of equal indices, where one is an upper index and the other a lower index. We shall later discuss this convention in some detail. One important point to note is that such a set of four-dimensional unit vectors will identify uniquely an inertial reference frame. This is different from the situation with three-dimensional vectors, where a set of three orthogonal unit vectors will dene the orientation of a reference frame, but not its velocity. A four-dimensional vector can be decomposed in its time component and its three-vector part. We often write it simply as x = (x0 , r) (4.27)
Note however that this formulation is somewhat sloppy since the four vector x is to be considered as independent of any choice of reference frame, while the decomposition (4.26) refers to a specic (inertial) reference frame, since the separation in time and space depends on the choice of inertial frame. In any case, such a decomposition is often useful. Even if the four-vector formulation is attractive since it gives a compact relativistic form of physical equations, the decomposition is often needed in order to make a physical interpretation of the results. The space-time vector x, which does not refer to any specic reference frame, we often refer to as an abstract vector. A concrete representation of the vector is given by its matrix representation, which we write as 0 x x1 x= (4.28) x2 x3 As opposed to x, this matrix does depend on the choice of reference frame, and the Lorentz transformations specify the how the matrix elements change under a change of the inertial frame. In the following we shall refer to this matrix, or more generally the collection of coordinates {x , = 0, ..., 3}, simply by the symbol x. It represents the set of coordinates of a space time point in a particular inertial frame. A transition between two inertial reference frames can now be viewed as a linear transformation of unit vectors and of coordinates in much the same as way as transformation of three-dimensional unit vectors and coordinates given by (4.21) and (4.23). We write the relativistic transformations as e e = x
L e
L x
(4.29)
80
Again we notice the different positions of space-time indices, with the coefcient of the basis vector transformations written as L and the coefcients of the coordinate transformation written as L . These are not identical, but closely related, as we shall later see. Here we notice that since the spacetime vector x is coordinate independent, the expansion of this vector in the given basis, implies that the transformation coefcients have to satisfy the equation
L L = is the Kronecker delta written in four-vector notation. This equation is a direct generalization where of the condition (4.22) satised by the transformation coefcients of rotations in three dimensions. Since the transition between inertial frames is described by a Lorentz transformation, such a transformation is now identied by the set of coefcients L . It is straight forward to check that the coordinate transformation (4.4) is a special case, with coefcients given by, L0 0 = L1 1 = , L0 1 = L1 0 = , L1 1 = L2 2 = 1, while other coefcients vanish.
(4.30)
4.4
Minkowski diagrams
The vector space of four-dimensional space-time, with the relativistic metric (4.16), is referred to as Minkowski space. When discussing motion in this space it is often useful to make a graphical representation of the space, but since we cannot make a good representation of all four dimensions we usually make a restriction to the two-dimensional subspace spanned by the coordinates (x0 , x1 ) or the three-dimensional subspace spanned by (x0 , x1 , x2 ). Such a restricted representation may be sufcient when we consider motion in one or two (space) dimensions. The graphical representations of the subspaces are referred to as Minkowski diagrams. Such diagrams are especially useful in order to show the causal relations between space-time points. In Figure 4.3a a two-dimensional Minkowski diagram is shown, which is similar to the space time diagram already used in Fig.4.2, with ct and x as coordinate axes of a chosen inertial system. The coordinate axes of another inertial frame, which moves in the x-direction relative to the rst one are also shown, together with the basis vectors of the two coordinate systems. In the diagram also the lines x = ct are shown, which indicate space-time paths for light signals that pass through the reference point 0. Let us rst consider the information given by the direction of the coordinate axes in the diagram. The ct coordinate we may view as the space-time trajectory, often called the world line, of (an imagined) observer at rest at the origin of inertial frame S , and in the same way the ct axis describes the world line of an observer at rest with respect to the (moving) reference frame S . The tilted direction of the ct axis simply means that the observer at rest in S moves relative to reference frame S . However, the x axis is also tilted relative to the x axis, and that is an effect that one does not see in a similar Galilean diagram. Since the x axis describe points that are simultaneous in reference frame S , this means that the two reference frames disagree on what are simultaneous space-time events. This is one of the important predictions of relativity, that simultaneity is not universally dened, but is reference-frame dependent. Let us next consider the implications of the fact that the location of the (red) light paths in the diagram are independent of the choice of inertial frames. There is a lightlike separation between the origin and any space time point that lies between the two lines, either in the upward or the downward direction relative to 0. Space-time points that have timelike separation from 0 and appear later we
4.4. MINKOWSKI DIAGRAMS

ct ct ct
81
absolute future e0 e0 e1
O
x x
absolute xA future xB xC
O
y x
e1 particle world line absolute past absolute past
a)
b)
Figure 4.3: Two-dimensional and three-dimensional Minkowski diagrams. In both diagrams the location of
the light cones relative to the point O are shown. They indicate which space time points are causally connected with or causally disconnected from O. The rst kind of points are represented by the time-like vector xA and the lightlike vector xB , the second type by the space-like vector xC . In gure b) the world line of
a massive particle is shown. It moves with subluminal velocity, which means that the four-vector velocity is timelike. refer to as lying in the absolute future of the point 0, while points with timelike separation that appear earlier than 0 we refer to as lying in the absolute past. Absolute here means that this ordering of events is independent of the choice of inertial frame. However, for events that lie outside of the light paths, either to the right or to the left, the situation is different. These are points at spacelike separation from the origin 0. For a specic reference frame like S also these points can be characterized as being either in the past (t < 0) or in the future (t > 0), but such a characterization is now reference frame dependent. It is therefore not necessarily the same for a reference frame S that moves relative to S . In fact for any point at spacelike separation from 0 there exist some inertial frames that will place this point in the past relative to the origin 0 and other inertial frames that will place the point in the future. This relativity in the characterization of space-time points as being in the past or in the future may seem somewhat confusing, but is in reality not in conict with causality, which orders events with respect to cause and effect. This is so since two points with spacelike separation are causally disconnected in the sense that no physical inuence can propagate from one of the space-time points to the other. The speed of light sets in relativity theory an upper limit to the propagation speed of any physical signal and such a signal can therefore not propagate between points with spacelike separation. In Fig4.3b we show a three-dimensional representation of Minkowski space. The light paths to and from the origin 0 now form a double cone, consisting of a future light cone and a past light cone. Space-time points inside the light cone are causally connected to 0, in the sense that points inside the future light cone can be reached by a physical signal sent from 0 and a point within the past light cone can reach 0 with a physical signal. In the diagram three four-vectors are drawn, where xA is a timelike vector, xB is a lightlike vector and xC is a spacelike vector. In the diagram the world line of
82
a (massive) particle that pass through the origin is also drawn. Since its velocity at all times is lower than the speed of light this space-time curve is restricted to lie within the light cone. In these diagrams the light cones associated with the origin 0 have been drawn. In reality any space-time point E can be associated with a past and a future light cone. These cones order the points of space time in those that are causally connected to E and those that are causally disconnected. As mentioned above, the Minkowski diagrams are particularly well suited for showing the causal relations between space-time points. However, one should be aware of the fact that there are in other respects certain shortcomings. This has to do with the point that the Minkowski geometry of spacetime is not well represented in diagrams with Euclidean geometry. This is seen in Fig.4.3a, where the coordinate axes of reference frame S seems to have a special status, since the time and space axis have orthogonal directions. That is not the case for the coordinate axes of reference frame S , even if we know that these two inertial frames in reality are equivalent. The length scale along the two sets of coordinate axis are also not represented as equal in the diagram. So one has to be aware of this, that angles and lengths will not be correct if measured directly from the diagram.
4.5
General Lorentz transformations
So far we have focussed on the special Lorentz transformations. These are the transformations that change the velocity of the inertial frame without rotating its axes. A special case of these is the boosts in the x direction, but the velocity of a general boost can have an arbitrary direction. The special Lorentz transformations (or boosts) are therefore characterized by three parameters, namely the three components of the velocity vector v that relates the two inertial frames of the transformation. Let us denote a general transformation of this type by B (with reference to this as a boost). A general Lorentz transformation is a transformation between inertial frames that may also include a rotation of the axes of the second reference frame with respect to the rst one. Such a transformation can therefore be seen as a composite operation, rst a boost and then a rotation4 L = RB (4.31)
The Lorentz transformation L denes a linear map of the vector coordinates of the rst reference frame (S ) into the vector coordinates of the second reference frame (S ). We may write it as x = Lx where x, x and L are matrices. Written out explicitly the matrix equation is 0 x0 L 0 x 1 L1 0 = x 2 L2 0 L3 0 x3 L0 1 L1 1 L2 1 L3 1 L0 2 L1 2 L2 2 L3 2 0 x L0 3 1 L 3 x1 L2 3 x2 x3 L3 3 (4.32)
(4.33)
The decomposition of the Lorentz transformation L in (4.31) can similarly be read as a matrix product of the boost matrix B and the rotation matrix R. Both these are 4x4 matrices, but the rotation matrix only mix the space coordinates x1 , x2 and x3 , and leaves the time coordinate x0 unchanged.
It can also be dened with the operations in opposite order, L = B R , but in general B will then be different from B and R will be different from R since these operations do not commute.
4
4.5. GENERAL LORENTZ TRANSFORMATIONS
83
The general Lorentz transformations, as dened above, are homogeneous linear transformations, which imply that the origin of the two coordinate systems are mapped into each other by the transformation. However, a transformation between inertial frames can also involve a shift of the origin. This leads to the inhomogeneous Lorentz transformations, which we may write as x = Lx + a where a represents the displacement of the origin. In matrix form a is 0 a a1 a= a2 a3 (4.34)
(4.35)
where the four parameters describe the shift of the origin along the four space-time axes. The inhomogeneous Lorentz transformations depend all together on 10 parameters, 3 of these are rotation parameters, another 3 are boost parameters and nally 4 are translation parameters. In mathematical terms this set dene a 10 parameter transformation group referred to as the inhomogeneous Lorentz group or the Poincar e group. The group property of the set implies that the successive application of two transformations will create a new transformation from the same set.5 The homogeneous transformation dene a smaller subgroup, which is the 6 parameter homogeneous Lorentz group or simply the Lorentz group. The rotations form an even smaller, 3 parameter subgroup of the Lorentz group. However, one should note that the set of boosts do not form a group, since the composition of two boosts with different directions will not be a pure boost, but will also include a rotation. This is purely relativistic effect with interesting physical consequences. A particular consequence is the Thomas precession effect, where a spinning particle which follows a bended path will show precession of the spin even if no force acts on the spin. The full set of inhomogeneous Lorentz transformations dene the fundamental symmetry group of special relativity. These symmetry transformations can in fact be interpreted in two different ways. They can be interpreted as passive transformations, which is the picture we use in these notes. This means that the transformation of coordinates is explained as a change of reference frame while the physical systems that are described are not changed in position or motion. When a symmetry transformation is instead interpreted as an active transformation this means that the change of coordinates corresponds to a physical change in the location of the processes described by the coordinates, while the reference frame is left unchanged. Such an active transformation could be to change the motion of a physical body by shifting its position, by rotating it and by changing its velocity. It is of interest to note that when we work with coordinates there is no difference between these situations that is seen in the description. This is a consequence of the fact that the transformations describe symmetries of the theory. A common property of all the space-time transformations discussed above is that they leave invariant the line element between space time points, s2 = r2 c2 t2 (4.36)
5 The group property of the Lorentz transformations means that the composition of any two Lorentz transformation will dene a new Lorentz transformation and the inverse of a Lorentz transformation is also a Lorentz transformation. This group property is almost obvious, with the Lorentz transformations being dened as mappings between inertial frames.
84
and this was in fact, for a long time, regarded as the basic condition that dened the relativistic symmetry transformations. However, there exist some discrete space-time transformations that leave the line element (4.36) unchanged, but which have been shown, by experiments, not to be fundamental symmetries in the same sense. These are the space inversion and time reversal transformations dened by the transformation matrices, 1 0 0 0 1 0 0 0 0 1 0 0 , T = 0 1 0 0 P = (4.37) 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 1 Since they only change the sign of either r or t obviously s2 is left unchanged. Most physical processes are in fact invariant under these transformations, but in elementary particle physics small effects which break P and T symmetry have been detected. These are associated with the weak nuclear forces.
Chapter 5
Consequences of the Lorentz transformations

The relativistic form of the fundamental space-time symmetries, expressed by the Lorentz transformations, has consequences for all physical theories at the fundamental level. Some of these may refer to as kinematical consequences, since they are directly linked to the relativistic transformations of space and time. One of these is the length contraction effect, which is the effect that a body in motion appears as shorter in the reference frame where the body moves than in the reference frame where it is at rest. Another kinematical effect is the time dilatation effect, which is the effect that time seems to run slower for a body in motion than for a body at rest. We shall discuss these effects and some further consequences of them, in particular the famous twin paradox, which has to do with the effect that two persons that follow different space-time paths between a common point where they depart and a common point where they meet again, will perceive a difference in the time spent on the journey.
5.1
Length contraction
We consider a situation where the length of a moving body is measured. For simplicity, let the body be a rod with length L0 when measured in its inertial rest frame. It is oriented along the x-axis in this reference frame, which we refer to as S . Another inertial frame S , called the laboratory frame, is oriented with the axes parallel to those of S , and measured relative to this frame the rod is moving in the x-direction with the velocity v , as illustrated in Fig. 5.1. We shall refer to the front end of the rod as A and the rear end as B . The space-time coordinates of these points in the two reference frames are related by the Lorentz transformations xA = (xA vtA ) xB v xA ) c2 v = (xB vtB ) tB = (tB 2 xB ) c tA = (tA
(5.1)
where the time coordinates of the two end points are independently chosen. We note that for the measurement of length in the rest frame S the time coordinates of the end points are unimportant, since the space coordinates do not change with time. The length of the rod is simply the difference between the (time independent) x coordinates of the ends of the rod, L0 = xA xB 85 (5.2)
86
CHAPTER 5. CONSEQUENCES OF THE LORENTZ TRANSFORMATIONS
y y
S
B v A x x
Figure 5.1: Measurement of the length of a moving body. S is the rest frame of the moving body which has
velocity v relative to the laboratory frame S . In the rest frame the measured length has its maximum value L0 , while in the lab frame it seems length contracted with a length L < L0 .
However, in S the positions of the endpoints change with time, and therefore it is meaningless to dene the length as the difference in x coordinates unless we specify for what time the positions should be determined. The natural denition is that length should be dened as the distance measured between simultaneous events on the space-time paths of the two end points. Note that this is how length is measured also in non-relativistic physics. If distance is measured between the positions at different times, any value could be found for the length. The important point is that in non-relativistic physics simultaneity is universally dened, whereas in relativity it is reference frame dependent. Therefore we state that The length of a moving body measured in an inertial frame S is the space distance between the end points of the body measured at equal times in the same reference frame S . This means that we for the moving rod, to nd the correct expression for the length in reference frame S , should x the time coordinates of the end points so that tA = tB (rather than tA = tB ). From the Lorentz transformation formula we then derive L0 = xA xB
= [(xA xB ) v (tA tB )] = (xA xB ) = L (5.3)
This is in fact the length contraction formula, which we may also write as L= 1 L0 L0
v2 c2
(5.4) 1. The formula tells us that the
where the last inequality follows from the fact that = 1/ 1
5.2. TIME DILATATION
87
length has it maximum when measured in the rest frame of the body. When measured in an inertial frame where the body is moving it seems length contracted in the direction of motion.
ct
ct
t A= t B
x x
L B A
tA= tB
Figure 5.2: Minkowski diagram for the length measurement. The shaded area shows the space-time trajectory
of the moving rod, with A and B as the trajectories of the end points. Length measurement in reference frame S (with unmarked coordinates) should be performed for space-time points with tA = tB as indicated in the gure. This is different from measurements for points with tA = tB , which is the natural choice in the rest frame S .
In Fig.5.2 the measurement of the length between the end points of the body in reference frame S is illustrated in a Minkowski diagram.
5.2
Time dilatation
Next we consider the relativistic effect that the clock in motion seems to be slower than a clock at rest. We have to specify precisely how the comparison is done, and also here the reference dependence of simultaneity is important. Let us consider a situation similar to that of the previous section. An inertial frame S is the rest frame of a clock that is localized at the space origin (x = y = z = 0) of S . It measures the time coordinate t , and is therefore often called a coordinate clock. The clock and the reference frame is moving with velocity v along the x axis relative to a second inertial frame S (the laboratory frame). The time coordinate t of this frame we may consider as being measured by a second clock which is at rest at the space origin of S . The coordinate transformation between the two reference frames is given by the same Lorentz transformation formula (5.1) as in the discussion of the length contraction effect. The situation is illustrated in the Minkowski diagram of Fig.5.3. The time axis of reference frame
88
ct
ct
x x
Figure 5.3: The time dilatation effect illustrated in a Minkowski diagram. The orthogonal (blue) coordinate
axes dene the laboratory frame S while the tilted (green) coordinate axes dene a reference frame S that moves relative to S . The green dots on the time axis denote the events where a co-moving coordinate clock makes clicks at equal time intervals. Similarly the blue dots on the time axis of S denote the clicks of a coordinate clock in the lab frame. An observer in S nds that the clicks of the moving clock come with larger time separation than shown by his own clock. This is illustrated by the dashed blue line which shows that the click of the moving clock has larger t coordinate than the corresponding click on his own. This is the time dilatation effect, usually stated as a moving clock runs more slowly than a clock at rest. The diagram shows how this effect is symmetric with respect to the two clocks. An observer in S will note that the clicks of the clock in S will have larger t coordinate than the corresponding clicks on her own clock. The comparison is now performed with equal times in S , as shown by the dashed green line. The comparison between the two clocks is thus performed for different pairs of events on the world lines of the clocks by the two observers.
S is the world line of the moving clock, and the ticks of the clock as regular intervals are indicated in the diagram (for example with = 1min). In the same way the ticks of the clock of the laboratory frame is indicated on the time axis of S . Now we want to examine how the time scale of the moving clock is perceived in the laboratory frame, when compared with the clock at rest in this frame. Since the two clocks are not located at the same space-time points we have to make clear how this comparison should be done. The important point is that we should choose one of the frames for the comparison, and in the present case we would like to compare the two clocks in the laboratory system S . Let us then focus on two events that correspond to two subsequent clicks of the moving clock. The rst event we may take as the coincidence of the origins of the two reference frames: t = t = 0, x = x = 0. We assume this event to correspond to the rst click of both clocks. The next event corresponds to the second click of the moving clock. It has coordinates in S , (x , t ) = (0, ), while the corresponding coordinates in S we refer simply to as (x, t). The time t is then the time of the second click as registered in S , and
5.2. TIME DILATATION
89
this is the time we would like to compare with the time shown on the moving clock. The time t is readily found by using the inverse of the Lorentz transformation formula applied in (5.1), ct = (ct + This gives the time dilatation formula t= (5.6) v x ) = c c2 (5.5)
where is referred as the proper time of the moving clock, which is identical to the time of the rest frame S , while t is the coordinate time of the frame S , which is moving relative to the clock. The time dilatation formula shows that the proper time is less than the coordinate time of any inertial frame which is not identical to the rest frame of the clock. This is to be compared with the length contraction formula which says that the length of a body measured in the rest frame is larger than the length measured in any other inertial frame. It is interesting to note that even if the situation may seem asymmetric between the two reference systems S and S , that is not really the case. The coordinate clock of system S seems to be slow when viewed from reference frame S , but at the same time the coordinate clock of S seems slow when viewed from S . The explanation for this apparently paradoxical situation is again the difference in perception of simultaneity in the two reference systems. When comparing the time difference of the two clocks in the two reference systems, this is done by comparing two different sets of space time points of the clocks. In both cases the comparison is made for simultaneous events, but that means for the two reference frames to use different sets of points for the world lines of the two clocks. The situation is illustrated in the Fig. 5.3. Let us now illustrate the length contraction and time dilatation effects in a slightly different way. We introduce a set of coordinate clocks for each of the reference frames in the following way. With equal spacing L0 along the x-axis of reference system S there are placed clocks that are stationary in this system. They are synchronized, so they all show the coordinate time of S . This synchronization can be done by sending radio signals between the clocks. In the same way we introduce a set of coordinate clocks with the same spacing L0 in reference frame S . There is also a synchronization of the two sets of clocks, since the two reference frames have a common origin for their coordinate systems. This means that the clocks at position x = 0 in S and at x = 0 in S will show the same time when t = t = 0. In Fig.5.4 the situation is illustrated by viewing the two sets of clocks at time t = 0 from reference frame S . All the coordinate clocks in S show the same time t = 0 and are located with separation L0 . However the moving clocks (coordinate clocks of S ) have a different separation L0 / due to the length contraction effect and they seems to go slower due to the time dilatation effect. In addition they seems not to be synchronized when viewed from S . This is demonstrated by the Lorentz transformation formula. For space time points with t = 0, which are simultaneous in S the coordinates in S are x = x , t = v v x = 2x 2 c c (5.7)
The rst equation is simply the length contraction formula. The second equation shows that the time shown by the moving clocks depend on their positions. This is again a consequence of the reference system dependence of simultaneity. In the present case the pictured events are simultaneous in S (t = 0) but not in S .
90
y y
L0/ x x
S
L0
Figure 5.4: Moving coordinate clocks. Two sets of coordinate clocks are attached to two inertial reference
frames S and S in relative motion. The situation is here registered in reference frame S . The coordinate clocks in S show all equal time t = 0, but the coordinate clocks of S seems not to be synchronized. Due to the length contraction effect they seem more densely spaced than the clocks in S and due to the time dilatation effect they seem to be running more slowly.
5.3
Proper time
Let us assume that a body is moving with constant velocity and that S is the rest frame of the body. By the proper time of the body we mean simply the coordinate time in the rest frame. The time dilatation effect shows that this time will be different from the coordinate time of any other inertial frame that is moving relative to the body. The denition of proper time can be generalized to moving bodies in the case where the velocity is no longer constant, as we shall now discuss. Let us us then consider a more general motion where the velocity of the body is no longer constant. Therefore there does not exist an inertial reference frame which is at all times the rest frame of the body. The body we consider in the following to be sufciently small so it can be regarded as a point particle. Even if there is no single inertial rest frame for the particle valid for all points on the particles world line, there will be such a rest frame for any given point. It is an inertial frame that moves with the same velocity as the body at that particular instant. We refer to this as the instantaneous rest frame of the particle. As soon as the particle changes its velocity this inertial frame ceases to be the rest frame of the particle. The important point is that the instantaneous rest frames at different points of the world line will in general be different inertial frames. The world line of the particle we shall consider as being divided into a sequence of small line elements. For each of these the change in velocity is negligible and the instantaneous inertial rest frame can therefore be treated as the rest frame not only at a single space-time point, but for the line element. Strictly speaking this is true only for an element of innitesimal length, and that is what we shall consider. For such an innitesimal element of the particle path the time dilatation formula is valid, and we write it as d = 1 v2 dt c2 (5.8)
5.4. THE TWIN PARADOX
91
where d is the time measured in the instantaneous rest frame and t is the time measured in an inertial frame S which we use as a xed reference frame for the full journey of the particle. Since the expression (5.8) is valid for any part of the particle trajectory, we can now dene the proper time of this trajectory between two space-time points A and B as being identical to the integrated time
B
AB =
A
v (t)2 dt c2
(5.9)
The proper time is then dened as the sum (integral) of the time intervals measured in the instantaneous rest frames along the path. These do not dene a single reference frame, but rather a continuous sequence of inertial frames. The variation in velocity means that the time dilatation factor becomes a time dependent function. The proper time we may consider as the time measured on an imagined clock that is xed to the small body during its space-time journey. It should then be clear that the proper time will not depend on the choice of the reference frame S in the description of the motion. However, that is not obvious from the expression (5.9) which does seem to depend on the choice of reference frame. So, it is of interest to demonstrate more directly that proper time, as dened above, is independent of such a choice, or stated differently the proper time AB is a Lorentz invariant. We then focus again on an innitesimal element of the space time curve, and consider the corresponding Lorentz invariant line element, which we have earlier introduced. In the present case it takes the form ds2 = dr2 c2 dt2 v2 = c2 (1 2 )dt2 c 2 2 = c d
(5.10)
This shows that d 2 is proportional to the invariant ds2 and is therefore also a Lorentz invariant. The minus sign in the relation is explained by the fact that the world line of the particle has a timelike orientation. If we now compare the proper time for different world lines between the same end points A and B , the expression (5.9) indicates that the proper time may be path dependent so that the path which at average has the largest value of v 2 will have the shortest proper time. This is indeed a real physical effect, and it is the basis for the twin paradox which we shall discuss next.
5.4
The twin paradox
We consider the following situation. A pair of twins are named Anne (A) and Bjarne (B), and at a given time twin B leaves the earth on a space ship while twin A stays behind on earth. B travels at high speed far out in the universe to visit a distant space station. After a short stay he returns to the earth where he arrives several years after his departure. When he meets his twin sister A he realizes that his sister has aged more than himself. Since he is well acquainted with Einsteins theory of relativity this does not come as a surprise. It may seem paradoxical, but he knows that twin A has performed a space-time journey that is close to a travel with constant velocity, and her proper time should therefore be longer than his own, since he himself, on his journey to the distant space station, has performed a journey with large changes in velocity as seen from any inertial frames. This effect is shown by the proper time formula (5.9).
92
However, there is something else that makes this situation look like a paradox. We know that the time dilatation formula is symmetric for two inertial frames with relative velocity different from zero. If the coordinate clocks of reference frame S seem slow when compared with the clocks of reference frame S , also the clocks of S seem slow when compared to the clocks of S . We also know that the dilatation only depends on v 2 , so that the direction of the relative velocity is unimportant. To formulate situation as a paradox let us assume that the velocity of the space ship of twin B is constant and the same on the way out to the space station and on the way back, except for its direction. Then the time dilatation factor is the same on the full journey. Of course, this cannot be fully correct, since there must be a period of acceleration at the beginning and at end of the journey as well as when B is close to the space station. But we may assume these periods to be very short compared to the time spent on the rest of the journey, and therefore these short periods should only contribute with minor corrections that we may neglect. So let us now formulate this as a paradox. On the way out to the space station the relation between the rest frames of the two twins is symmetric, so the clocks of B seems to be slow measured with the clocks of A and vice versa. The time dilatation factor is constant and it is the same whether viewed from twin A or twin B . The situation is the same on the way back from the space station, with the same value for the time dilatation factor as on the way out. Based on this twin A will nd that the proper time of B is reduced with the factor relative to her proper time, and that is consistent with the time dilatation formula (5.9). But based on the symmetry between the two twins on each of the halves of the journey and the fact that the time dilatation factor is the same for the two parts, it seems that twin B could also claim that the proper time of twin A should be shorter than his. That would clearly create an inconsistency. We will resolve this apparent contradiction. First consider the situation from the point of view of twin A. Her own proper time is identical to the inertial reference frame S of the earth. (It is only in in an approximate sense an inertial frame, but since the orbital velocity of the earth is so small relative to the speed of light that is ok.) Let us denote her proper time for the whole journey by A . The space-time path of B is assumed to be symmetric with respects to its two halves, to and from the space station, and therefore when A applies the time dilatation formula to each part of the journey she obtains for the total time B = 1 1 1 A /2 + A /2 = A (5.11)
This is consistent with the time dilatation formula (5.9). Next we consider the situation from twin B s point if view. He can also apply the time dilatation - if he does it with some care. An important point to observe is that even if the speed of his space ship is the same on the way out and on the way back, the inertial rest frames on the two parts of the trip are not the same. Let us refer to these two part of the journey as I and II and the corresponding inertial frames as SI and SII . The main point is now to observe that when using the time dilatation formula he should refer to events that are simultaneous in his own reference frame. Let us apply this to the rst part of his journey, when his rest frame is SI . The time dilatation formula can be written as t A = 1 B /2 (5.12)
with tA is the time registered on the clock on earth during the time twin B is on the way to the space station. This has the same form as the time dilatation formula used by A. But note that tA = A /2 since the time on earth that is simultaneous in SI with the arrival of B at the space station is not the half time of the full journey. It is in fact an earlier time. This is illustrated in Fig. 5.5, which also shows
5.4. THE TWIN PARADOX
93
that the time on earth which is simultaneous in SII with the time of departure from the space station is later than the half time of the journey. So also for the travel back twin B may use a time dilatation formula similar to (5.11). But the two contributions to the time measured on earth do not add up to the full time of the journey. The formula that relates the proper time of B to the time registered on earth may can therefore be written as A = 2tA + tI/II = 1 B + tI/II (5.13)
This looks almost like an inverted form of the time dilatation formula (5.11), but there is a correction term tI/II . This comes from the fact that the two reference frames SI and SII do not agree on what are simultaneous events. When B suddenly changes from SI to SII as rest frames, and thereby changes the denition of simultaneous events, this is registered by B as a jump in the time coordinate of A. This jump in the time coordinate which follows from the fact that two different inertial rest frames are used on the travel of twin B demonstrates the lack of symmetry between twin A and B , and it reconciles the equations for time dilatation used by the two twins. In fact consistency between Eqs.(5.11) and (5.13) can now be used to determine the time jump tI/II , A = = which gives tI/II = (1 1 v2 ) = A A 2 c2 (5.15) 1 B + tI/II 1 A + tI/II 2
(5.14)
A more direct calculation of the time jump based on the use of the conditions for simultaneous events in the two inertial reference frames gives the same result. The conclusion is that the situation is not symmetric with respect to describing the journey for twin A and B . Both twins may use the time dilatation formula to compare the proper times of the two of them, but twin B has to be careful to add the time jump associated with the change of inertial frames. Let us also note that if we take into account that the change between the two rest frames SI and SII of twin B in reality is not innitely rapid, then the time jump tI/II will be replace by a rapid but smooth change. The space ship will have a continuous slow down in speed and re-acceleration at the space station and that will imply a smooth transition of instantaneous rest frames beginning with SI and ending with SII . This will affect the registering of simultaneous events on earth so that during the rst and second part of the journey the clocks on earth are registered as being slower than the ones on the space ship, but this is compensating by a very rapid speed up during the period of acceleration. The total effect is that, when correctly calculated, twin B should like twin A nd the proper time B to be shorter than the proper time A between the start and end point of the space-time journey. But the easiest way to compare the times registered by the twins is to use proper time formula (5.9) for the two space-time paths. This formula gives the correct proper time for any space-time trajectory, whether it is accelerated or not.
94
ct
ctII
II B A B ctI
SII
tI/II
I
{
A
xII
SI
xI
Figure 5.5: Illustration of the Twin Paradox. The Minkowski diagrams show the asymmetry between the two
twins when they use the time dilatation formula. The space-time journeys of the two twins A and B are shown by the blue lines in two Minkowski diagrams, with the world line of twin A (who remains at earth) represented as a single straight line, while the world line of B consist of two straight lines, denoted I for the ougoing part and II for the return part of the journey. (The effect of acceleration of B is neglected.) In the rst diagram the coordinate lines of the rest frame S of twin A are shown. The coordinate time of the mid-journey event of B is indicated by the green line. In the second diagram coordinate lines of the rest frames of twin B are shown. There is a discontinuity since the rest frame of the journey out (SI ) is different from the rest frame of the journey back (SII ). The point on earth which is simultaneous with the mid-journey event now splits in two, since the simultaneous events of frame SI and SII are different, now shown by the two unbroken green lines. Twin B has to include the corresponding jump in time (tI/II ) when using the time dilatation formula. The dotted green lines show the coordinate lines of the rest frames of twin B which interpolates between SI and SII in the short proper time interval when the space ship is accelerated. The red lines are included in the diagrams to show the world lines of light signals emitted at the beginning of the journey and at the mid-journey event.
Chapter 6
The four-vector formalism and covariant equations

In this chapter we discuss in a more systematic way the use of four-vectors, and in particular how to give physical equations a covariant form. In the covariant formulation all physical variables are expressed in terms of four-vectors and related objects, called (relativistic) tensors, and this formulation secures that the equations are valid in any inertial reference frame. We discuss how tensors are dened and what are their transformation properties under Lorentz transformations.
6.1
6.1.1
Notations and conventions

Einsteins summation convention
When using the four-vector notation some conventions are commonly used, and we shall make use of them also here. For example when a vector index is running over all the four values taken by the space-time coordinates, we label the index by a greek letter, while the use of a latin letter instead would normally indicate a restriction to the three values taken by the space components. For example when we write x , is allowed to take values from 0 to 3. If however we write xi the index runs instead from 1 to 3. Another convention we shall apply is Einsteins summation convention. Thus a repeated spacetime index in a product (or other expression) normally means that we should sum over the index. As an example we write in the following for the decomposition of a four-vector x on an orthogonal set of basis vectors, x = x e (6.1)
where the summation symbol is simply omitted. The repeated index tells us that we should sum over , and since it is a greek letter we know that the summation is from 0 to 3. If we at some stage should meet a case where a repeated index should not be taken as a summation index, we simply state that explicitly. In the four vector notation it is also important to correctly place the index up or down, while a similar distinction is not important four vectors in three-dimensional space. We shall soon have a closer look at this distinction. The consistent use of four-vectors (and tensors) we refer to as covariant notation, and we note as a particular rule that in the covariant notation we only sum over pairs of indices, where one is an upper index and the other a lower index. This summation is sometimes referred to as a contraction. 95
96
CHAPTER 6. THE FOUR-VECTOR FORMALISM AND COVARIANT EQUATIONS
6.1.2
Metric tensor
Physical three-dimensional space is in non-relativistic physics considered to be equipped with a Euclidean metric, dened by the invariant distance between two neighbouring points. The distance squared is in Cartesian coordinates ds2 = dx2 + dy 2 + dy 2 = dr dr (6.2)
and is invariant under rotations of the coordinate axes. As already discussed, the four-dimensional space-time of special relativity has a different metric, called Minkowski metric. It is dened by the Lorentz invariant line element ds2 = dx2 + dy 2 + dy 2 c2 dt2 dx dx (6.3)
We write this expression also as the squared norm of the vector, dx dx = dx2 , but we then have to remember that dx2 does not have to be positive. It is positive for spacelike vectors, zero for lightlike vectors and negative for timelike vectors. We may write the invariant line element in the following form, ds2 = g dx dx (6.4)
where g is referred to as the metric tensor. It can be thought of as dening a 4 4 symmetric matrix. (Note that in (6.4) Einsteins summation convention has been used.) This matrix is (in Cartesian coordinates) a diagonal matrix of the form 1 0 0 0 0 1 0 0 g = (g ) = (6.5) 0 0 1 0 0 0 0 1 From the decomposition of the vector dx = dx e and from the writing of the invariant line element as a generalized scalar product, it follows that the basis vectors satisfy a generalized orthogonality condition e e = g (6.6)
This means that the vectors are orthogonal and space vectors have a standard normalization e2 k = 1 , k = 1, 2, 3, while that time vector has the normalization e2 = 1 . The last one is negative since 0 the basis vector e0 is timelike.
6.1.3
Upper and lower indices
We have already stressed the convention that the coordinates of a four-vector x are written with upper indices, as x . However also coordinates with lower indices may be dened. The precise denition is, x = g x (6.7)
Thus a four-vector can be associated with two sets of coordinates, those with upper indices which are the standard ones (referred to as contravariant components) and those with lower indices (referred to
6.2. LORENTZ TRANSFORMATIONS IN COVARIANT FORM
97
as covariant components). The metric tensor acts as a lowering operator on the indices. This gives a simple relation x0 = x0 , x1 = x1 , x2 = x2 , x3 = x3 (6.8)
Note that the only change introduced by lowering the indices is that the sign of the 0th component is reversed. Initially it may seem cumbersome to operate with two sets of coordinates for a four-vector, which are even so closely related. However, if one is careful to place the indices correctly the relativistic equations can be simplied, and if the positions of the indices are consistently used on both sides of a relativistic equation one will gain a guarantee that it keeps the form unchanged when transforming from one reference frame to another. We note, as a special case, that the invariant line element can now be written without the metric tensor as ds2 = dx dx (6.9)
More generally, summation over a pair of four-indices, one lower and one upper will produce a Lorentz invariant quantity. The metric tensor acts as a lowering operator on the vector indices. Clearly there must be an inverse to this which acts as a raising operator. We write it as x = g x Since it is the inverse to g we have the relation
g g =
(6.10)
(6.11)
Note that the relativistic form of the Kronecker delta is written with one upper and one lower index. This is to have the indices of the two sides of the equation consistently placed. We note from the matrix form of g that the square of the matrix is identical to the identity matrix. This means that the matrix is its own inverse and therefore g and g represent the same 4 4 matrix. Nevertheless, we insist on writing this matrix with lower indices when it is used as a lowering operator of vector indices in an equation and with upper indices when it is used as a raising operator. This is to be able to place consistently all vector indices in the relativistic equations. 1
6.2
Lorentz transformations in covariant form
A Lorentz transformation, which relates the coordinates x of an inertial frame S to the coordinates x of another inertial frame S can be written in component form as x = L x where we again stress the convention for placing the indices of L, or in matrix form as x =Lx (6.13) (6.12)
1 The notation with covariant and contravariant components is even more important in the general theory of relativity where more general coordinate systems are applied. In that case the metric tensors g and g will usually no longer correspond to the same 4x4 matrix.
98
For the a boost in the x-direction the transformation is x 0 = (x0 x1 )
x 1 = (x1 x0 ) x 2 = x2 x 3 = x3 (6.14)
with = v/c and = 1/ 1 2 and v as the relative velocity of the two reference frames. If a general 4 4 matrix L should represent a Lorentz transformation, it has to satisfy a certain restriction, which follows from the requirement that the velocity of light is left unchanged by the transformation. As already noted this is related to the Lorentz invariance of the line element, which implies g dx dx = g L L dx dx = g dx dx Since this should be valid for any displacement dx , the L matrix has to satisfy the restriction g L L = g In matrix form this can be written as LT gL = g (6.17) (6.16) (6.15)
where LT represents the transposed matrix. This equation, which determines whether the 4 4 matrix L represents a Lorentz transformation, corresponds to the following condition that 3 3 rotation matrices R satisfy in three-dimensional space, RT R = 1 where 1 represents the identity matrix. (6.18)
6.3
General four-vectors
So far we have considered four-vectors as being associated with points in four-dimensional space time. However, exactly as in three dimensions vectors can be more general objects, for example associated with velocity, acceleration, vector elds etc. A general four-vector A is characterized by: it has four components A , = 0, 1, 2, 3 , the components transform as the coordinates x under Lorentz transformations, A A = L A . The Minkowski diagram is convenient to give a geometric representation of general four-vectors, in the same way as with the use of a Minkowski diagram for space-time itself. A reference frame corresponds also here to a choice of basis vectors {e , = 0, 1, 2, 3} and a vector A can be decomposed on any set of basis vectors, corresponding to different inertial reference frames, A = A e = A e (6.19)
6.3. GENERAL FOUR-VECTORS

e0 e0 A B e1 e1
99
Figure 6.1: A two-dimensional Minkowski diagram with coordinate axes corresponding to two reference
frames in relative motion. The coordinate axes of the un-primed system are perpendicular in the diagram, but not the primed axes. In reality both sets of axes dene orthonormal directions in the sense of the relativistic scalar product. The decomposition of the timelike vector A on both sets of coordinate axes are shown. The space-like vector B is orthogonal to A even if they are not perpendicular in the diagram. All the space-like vectors with tips at the hyperbolic curve (dashed green curve) have the same relativistic length, even if the Euclidean lengths in the diagram are quite different.
Thus a Lorentz transformation simply corresponds to a change of basis in Minkowski space. This is illustrated in the Minkowski diagram of Fig. 6.1, where the basis vectors e are represented as orthogonal vectors, while e are represented as non-orthogonal vectors. This difference between the two sets of basis vectors is only apparent, and follows from the fact that the Minkowski diagram gives a graphical representation in a plane with Euclidean metric, while in reality Minkowski space has a non-Euclidean geometry. The Lorentz invariant scalar product, is dened by A B = g A B = A B (6.20)
The scalar product is indenite (not positive) and separates the general four vectors, like the space time vectors dx, in three classes: space-like (A2 > 0), light-like (A2 = 0) and time-like (A2 < 0). In the two-dimensional Minkowski diagram these three classes are represented by vectors lying outside the light cone, on the light cone or inside respectively. As already noticed, orthogonality in the sense that the scalar product of two four-vectors vanishes does not mean that they appear as orthogonal in the Minkowski diagram. In the diagram 6.1 orthogonality of the two vectors A and B, dened as A B = 0, means that the two vectors have directions symmetrically about the light cone. In particular a lightlike vector, with this denition, will be orthogonal to itself. Thus, even if graphical representations in terms of the two- (or three-) dimensional Minkowski diagram is often useful, one has to remember that the geometry of Minkowski space is not correctly represented. As already discussed, angles may not be faithfully represented, and also the relativistic distances will generally not coincide with the Euclidean distances in the diagram. In particular we note that the path that appears to be the shortest one between two points with timelike separation in
100
reality corresponds to the path with largest value of the proper time between the two points. This was demonstrated in the discussion of the twin paradox.
6.4
Lorentz transformation of vector components with lower index
The index of a general four-vector can be lowered by applying the metric tensor, in the same way as for the position vector x , A = g A (6.21)
This relation leads to different transformation properties for vector components with upper indices (contravariant components) and lower indices (covariant components). We nd the following expression for the transformed covariant components A = g A = g L A = g L g A L A Note in the last line we have introduced a modied symbol for the transformation matrix L = g L g (6.23) (6.22)
where we have followed the general rule that g acts as a lowering operator and g as a raising operator. With L as the matrix elements of the 4 4 matrix L, L then are the matrix elements of the matrix = gLg 1 L = (LT )1 (6.24)
The last expression is derived from the identity (6.17), which is satised by all Lorentz transformation matrices L. Note that the covariant and contravariant components transform in inverse ways, which is consistent with the fact that the Lorentz invariant scalar product of two vectors, which can be written as a product of covariant and contravariant components of the the two vectors. Also note that the transformation coefcients L of the covariant components A are the same as the transformation coefcients of the basis vectors e , which have earlier been introduced in (4.29) and (4.30). This is consistent with a general property of the covariant formalism, namely that the position of the spacetime index of an object, as an upper or lower index, indicates uniquely the transformation property of this object under Lorentz transformations.
6.5
Tensors
The four-vector notation is useful in order to express the relativistic equations in a compact form which applies to all inertial reference frames. However, all physical quantities cannot be written as vectors, and this motivates to introduce more general objects called tensors. These are multicomponent objects
6.5. TENSORS
101
that transform in ways closely related to that of vectors. They are equipped with a set of space-time indices rather than with a single index, where the number of indices is called the rank of the tensor. The four-vector is a special case, it is a rank 1 tensor. A rank 2 tensor is written as T , = 0, 1, 2, 3; = 0, 1, 2, 3. (6.25)
It has all together 16 components. The important property of a tensor is the way its components transform under a change of reference frame. The transformation is determined by the number and position (up or down) of its space time indices. Thus there is one Lorentz transformation matrix for each index, so that the rank 2 tensor transforms as T T
= L L T
(6.26)
As an example, we may from two vectors A and B easily form a rank 2 tensor C = A B (6.27)
This composition is called the tensor product of the two vectors. Another rank 2 tensor that we will meet later in the course is the electromagnetic eld tensor F . This tensor is antisymmetric in and and is composed by the electric and magnetic eld strengths so that F 0k , k = 1, 2, 3 are the electric components and F kl , k, l = 1, 2, 3 are the magnetic components. Tensors may, like vectors, be written with upper indices or lower indices. These are related by the action of the metric tensor. For rank 2, we then have four related tensors T , T = g T , T = g T , T = g g T (6.28)
With the introduction of tensors we have a series of different, but related relativistic objects at our disposal: A B C D etc. rank 0 (scalar) rank 1 (vector) rank 2 rank 3 no vector index one vector index two vector indices three vector indices (1 component) (4 components) (16 components) (64 components)
We note that a contraction, i.e., a summation of one upper and one lower index will transform a tensor into a new tensor, with rank reduced by 2. For example A = A is a scalar, B = B is a vector etc. When the relativistic equations are expressed in terms of tensors, they are said to be in covariant form. When the equations are written in covariant form they are expressed in terms of variables with simple, standardized transformation properties. One can then easily check that the two sides of the equation transform in the same way, so that the equation is valid in any reference system. To check that a covariant equation has the correct form we note that free indices (that are not summed over) should have the same positions (up or down) on both sides of the equation, repeated indices that are summed over should appear with one in the upper position and one in the lower position.
102
As an example, the equation of motion of a charged particle in an electromagnetic eld has the following compact covariant form mx = eF x (6.29)
where the differentiation is here with respect to the Lorentz invariant proper time of the particle. A vector A we may consider as a geometrical object which is independent of any choice of reference frame, while the components A of the vector do depend on such a choice. Also for general tensors we may take a similar point of view. The tensor components T .. then represent a geometrical object T. And whereas the components of this object transforms with the change of reference frame, the (abstract) tensor T itself is independent of any choice of reference frame. However there is a difference between vectors and general tensors when it comes to making concrete representations of their geometrical form. While a vector has an immediate visual interpretation as an object with length and orientation, such a simple picture is not available for general tensors. There are however special cases where a visual picture may work. For example an antisymmetric rank two tensor can be represented as an antisymmetric product of two vectors. This (generalized) vector product of the vectors may be taken to represent a (at) surface element in four dimensions. Such a surface element is characterized by a an area and by an orientation of the surface element in four-dimensional space.
6.6
Vector and tensor elds
In a similar way as vectors in three-dimensional space often appear in the form of vector elds, vectors and tensors in four-dimensional space-time may also appear in the form of vector and tensor elds. As a particular example the electromagnetic eld is in covariant relativistic form described by the rank two tensor eld F (x). Let us list some of the tensor elds we may meet in relativistic theories: = (x) A = A (x) F = F (x) etc. scalar eld vector eld rank two tensor eld eld
The elds are here written in component form and the space time variable x here means the full set of coordinates x = (x0 , x1 , x2 , x3 ). Under a change of inertial reference frames, dened by a Lorentz transformation L, the elds transform in the following way Scalar eld Vector eld Tensor eld etc. (x) (x ) = (x) A (x) A (x ) = L A (x) F (x) F (x ) = L L F (x)
One should note that there are two changes under the transformation. The eld components transform according to the rank of the tensors, with the number of Lorentz matrices determined by their rank. But also the space-time argument changes, with x = L x . This change simply means that the untransformed as well as the transformed elds refer to the same space-time point, but this point is represented by different sets of coordinates in the two inertial reference frames connected by the Lorentz transformation. Physical elds, like the electromagnetic eld, will usually satisfy a set of eld equations, and when formulated as relativistic equations we are often interested in expressing them in covariant form. They
6.6. VECTOR AND TENSOR FIELDS
103
are typically differential equations, and we will therefore discuss in general terms how differentiation with respect to the space-time coordinates are treated in the covariant formalism. We rst examine the four gradient of a scalar eld (x), written as A ( x ) = (x) (x) x (6.30)
We have here introduced the symbol to represent the derivative with respect to x and we will use this convenient notation in the following. We have also by writing the partial derivative of as A indicated that the components of the derivative transform as covariant four-vector components, but that needs to be proven. In order to do so we note that the change of space time coordinates x x can be viewed as a change of variables for the elds. Derivatives with respect to x can then be related to derivatives with respect to x by the chain rule. For the differentiation operators we write this as x = x x x or simply as
(6.31)
x x
(6.32)
Since the Lorentz transformation can be written as x = L x we nd x = L x (6.34) (6.33)
but it is actually the derivative for the inverse transformation that we need. To invert the transformation we make use of the property of the Lorentz transformation matrix g L L = g By use of this identity we can re-write the transformation equation (6.33) as g L x = g L L x = g x (6.36) (6.35)
and by further applying the raising operator on the index and changing the name of some of the summation indices we nd the inverse transformation formula x = g L g x = L x where we have made use of the denition L = g L g . As a result we nd x = L x (6.38) (6.37)
to be compared with the transformation matrix (6.34). The relation between derivatives with respect to the original and the transformed space-time coordinates can then be written as
= L
(6.39)
104
and this shows that the partial derivatives transform in the same way as the covariant components of a vector. In particular this gives for the four gradient (x ) = L (x) (6.40)
which is identical to the transformation equation for a covariant vector eld (see(6.22). The rule for writing an equation in covariant form when it involves derivatives is therefore simple. The equation should be written in tensor form (including scalars and vectors) with each space-time derivative adding a covariant four-vector index to the expression. The equation (6.30) for the fourgradient therefore has a correct covariant form. In the same way the four-divergence of a vector eld A (x) can be written in the covariant form as (x) = A (x) (6.41)
with (x) as a scalar eld. We nally note that the transformation properties of the partial derivatives means that we can form a Lorentz invariant quadratic derivative operator = g = 2 1 c2 t (6.42)
This is called the dAlembertian and is an extension from the Laplacian in three space dimensions to an operator in four space-time dimensions. As indicated by the contraction between an upper and a lower space-time index this operator transform as a scalar under Lorentz transformation.
Chapter 7
Relativistic kinematics
We discuss in this chapter how to describe velocity and acceleration as relativistic four-vectors. The concept of proper acceleration is introduced and an example of motion with constant proper acceleration is investigated.
7.1
Four-velocity and four-acceleration
We consider the motion of a point particle through space and time. It can be described by a time dependent position vector, which we decompose in its time and space parts, dened with respect to some unspecied inertial frame, x(t) = (ct, r(t)) (7.1)
Let us introduce the particle velocity by the time derivative of the four vector, this also decomposed in its time and space parts, dx dr = (c, ) dt dt (7.2)
However, the derivative of the four-vector x(t), when differentiated with respect to time t of the chosen inertial frame, is itself not a four-vector. As a direct demonstration of this we consider the special case where a particle is moving along the x-axis with velocity u relative to a coordinate system S . Assume another inertial frame S is moving relative to this frame with velocity v , also in the direction of the x axis, so that the coordinates of the two frames are related by a special Lorentz transformation (boost) in the x direction. The time derivative of the position vector, when decomposed in the coordinates of the two frames, will have the form S: dx = (c, u, 0, 0) , dt S : dx = (c, u , 0, 0) dt (7.3)
dx when all the four space-time components are shown. The velocities are given by u = dx dt and u = dt and the relation between these is given by the relativistic transformation formula for velocities, (4.8),
u =
uv 1 uv c2
dx dt
(7.4)
Clearly the transformation of the components of form of a four-vector transformation.
between the two inertial frames does not have the
105
106
CHAPTER 7. RELATIVISTIC KINEMATICS
The reason for this is easy to understand. The position vector is differentiated with respect to x the time coordinate of a specic reference frame, and the resulting vector, d dt , will therefore not be coordinate independent. This result suggests that we need to use a Lorentz invariant time parameter in order to dene velocity as a four-vector. Such a parameter is in fact available in the form of the proper time of the moving particle. As already discussed the proper time of a particle is directly related to the invariant line element of the particle path and is therefore also a Lorentz invariant. We therefore dene the four velocity of the particle as U= or in the component form U = dx d (7.6) dx d (7.5)
with as the proper time coordinate. The Lorentz invariance of the proper time is shown explicitly by the denition of the time difference d for an innitesimal section of the particles world line, d 2 = 1 dx dx c2 (7.7)
With as a Lorentz invariant it is clear that the vector components U and x transform in the same way, which secures that U as dened above is a four vector. The denition of the proper time (7.7) furthermore shows that all the four components of U cannot be independent. This is shown explicitly by evaluating the Lorentz invariant U2 = U U = c2 (7.8)
For any motion of the particle the four velocity is thus a timelike vector with a xed (negative) norm squared. This can be seen in another way by expressing the four-velocity in terms of the (referencer frame dependent) velocity v = d dt . We have U = dx d d (ct, r) = d dt d = (ct, r) d dt = (c, v)
(7.9)
where we have decomposed the four-vector x into its time and space-parts (with respect to the undt specied inertial frame), and where we have made use of the time dilatation formula d = . In this formulation the constant value for the Lorentz innvariant U2 follows form the identity U2 = 2 (v 2 c2 ) = c2 (7.10)
We note that the presence of the factor in the expression (7.9) for U is important in order to dene this as a four-vector. It is now fairly obvious how to dene the corresponding four-acceleration, A= dU d2 x = 2 d d (7.11)
7.1. FOUR-VELOCITY AND FOUR-ACCELERATION
107
Again, since is a Lorentz invariant parameter, the transformation properties of the components of U and A will be the same. We would now like to relate the four-vector A to the usual (reference-frame dependent) accelerv ation a = d dt . This we do by decomposing the four-vector into its time and space parts in a similar way as we have done for the four-velocity U, A = ( dU 0 dU , ) dt dt (7.12)
and we examine the two parts separately. For the 0 component we have d dU 0 =c dt dt and for the three-vector part dU d dv d d = ( v) = + v = a + v dt dt dt dt dt The time derivative of the -factor is d dt = = = = = d v2 1 (1 2 ) 2 dt c v 2 3 v dv (1 2 ) 2 2 c c dt 1 3 1 dv2 2 c2 dt 1 dv 3 2 v c dt 1 3 2 v a c (7.14) (7.13)
(7.15)
This gives for the time and space components of the four-acceleration d va = 4 dt c d v a A = 2a + v = 2a + 4 2 v dt c A0 = c
(7.16)
These expressions are valid in any chosen inertial frame, with v as the (time dependent) velocity of the particle in this frame and a as the time derivative of the velocity in the same frame. If we focus on the space part of the four-vector A we note that it has one part which is proportional to the acceleration a (in the chosen inertial frame) and the proportionality factor can by interpreted as a time dilatation factor when the proper time rather than the coordinate time is used in the time derivative. However there is another term which in direction is proportional to v rather than a. This comes from the time derivative of the time dilatation factor. There are now two new Lorentz invariants that we can construct with the help of A (and U). The rst one is UA=U dU 1 dU2 = =0 dt 2 dt (7.17)
108
This result follows from the fact that U2 is a constant. The other Lorentz invariant is A 2 = A 2 A0 = 4 a 2 + 6
2
(v a)2 c2
(7.18)
where the last expression is valid in any chosen inertial reference system. We have already noted that the four-velocity U is a timelike vector. Since A is orthogonal (in the relativistic sense) to a timelike vector, it has itself to be a spacelike vector, as one can also show by a direct calculation. Since A is spacelike it means that one can by properly choosing the inertial frame transform the time component A0 to zero. As shown by the expressions (7.16) this happens when v and a are orthogonal. In particular this is the case in the instantaneous inertial rest frame of the particle, where v = 0. The acceleration measured in instantaneous inertial rest frame is referred to as the proper acceleration of the particle. Let us denote the proper acceleration as a0 . We should stress that this acceleration is for any point on the particles world line measured in the inertial reference frame where the particle is instantaneously at rest. This means that when we follow the motion of the particle, the proper acceleration refers to a (continuous) sequence of inertial frames, each of them associated with a particular point on the particle path. The proper acceleration will in general vary along the path, so that it can be regarded as a function of the proper time of the particles world line, a0 = a0 ( ). When decomposed in the instantaneous rest frame the four-acceleration then gets a particularly simple form, A( ) = (0, a0 ( )) (7.19)
This means that we can identify the Lorentz invariant (7.18) with a2 0 , and therefore we have the following relation between the proper acceleration and the acceleration measured in another inertial frame
4 2 6 a2 0 = a +
(v a)2 c2
(7.20)
This shows that the proper acceleration is larger than the acceleration measured in any other inertial frame, a0 a. Let us consider two special cases. For motion in a circular orbit we have v a = 0 and therefore a0 = 2 a (7.21)
In the rest frame the acceleration is enhanced by the factor 2 , and this we may see as a time dilatation effect, due to the double differentiation with respect to proper time rather than coordinate time. The other special case is linear acceleration where v a = v a. In this case we nd
4 2 2 a2 0 = (a +
v2 2 a ) c2 = 4 (1 + 2 2 )a2 = 6 (1 2 + 2 )a2 = 6 a2 (7.22)
so in this case the enhancement factor is even larger, a0 = 3 a (7.23)
109
7.1.1
Hyperbolic motion through space and time
We will illustrate the discussion of the previous section by considering a space travel with constant proper acceleration. Let us therefore assume that a space ship is leaving earth for a travel far out in the universe. The ship is maintaining a constant direction of velocity and the engines are providing a thrust so that the effective gravitational eld on board is kept constant and equal to the gravitational eld on the surface of the earth. This means that the acceleration relative to an inertial rest frame, that is the proper acceleration of the ship, is the same at all times of the travel, with a0 = g = 9.8m/s2 . The problem to be discussed is how this travel appears in an earth-xed frame, which we can assume to be (to a good approximation) an inertial reference frame. Since the motion of the space ship is assumed to be linear we have for the velocity and acceleration, as seen in the earth-xed reference frame, v a = va, and the relation between the (constant) proper acceleration and the acceleration measured in the earth-xed frame is a= a0 g = 3 3 (7.24)
The acceleration therefore seems to decrease with time, when measured at earth, and this we can view as a consequence of the time dilatation effect. By integrating the above equation we can nd the position of the space ship as a function of its proper time. We choose the x axis of the inertial frame in the direction of the motion. First we re-write the equation as a differential equation for = v/c, d dt d 1 g 1 g = = a= = (1 2 ) d d dt c a 2 c (7.25)
It is now convenient to substitute with the rapidity , which we have earlier introduced. It is related to by = tanh , which gives d d d 1 d =( tanh ) = 2 d d d cosh d and 1 2 = 1 tanh2 = 1 cosh2 (7.27) (7.26)
The differential equation for therefore gets the following simple form when expressed in term of the rapidity g d = d a with solution = g a (7.29) (7.28)
We have then assumed that the time coordinates are t = = 0 at the beginning of the journey, when the velocity vanishes, and therefore = = 0. The solution for the velocity is then g = tanh( ) c (7.30)
110
which for the factor gives g = cosh( ) c (7.31)
The relation between the coordinate time t and the proper time can be determined from the time dilatation formula g dt = d = cosh( )d c which by integration gives t= c g sinh( ) g c (7.33) (7.32)
In a similar way the x coordinate can be found by integrating the expression for the velocity, dx dx g = = c = c sinh( ) d dt c which gives x= c2 g cosh( ) g c (7.35) (7.34)
In the last expression we have for simplicity chosen the integration constant to be zero. Note that this means that the x coordinate is not zero at the beginning of the journey, but rather x(0) = c2 /g . To sum up, the coordinates of the space ship in the inertial frame of the earth are given by ct = c2 g sinh( ) , g c x= c2 g cosh( ) , g c y=z=0 (7.36)
when the proper time is used as the time parameter of the space ships world line. From this follows that the coordinates satisfy the equation x2 (ct)2 = c4 g2 (7.37)
In a two-dimensional Minkowski diagram, with ct and x as coordinate axes, the world line of the space ship will therefore dene a hyperbel. This is illustrated in the Fig. 7.1. To get some feeling for what this means let us consider how time and position of the space ship, as registered on earth, develops as a function of the time coordinate registered on the space ship. We rst note that the proper acceleration a0 = g denes a time constant 0 = c = 0.97 year g (7.38)
when g = 9.81m/s2 . So this time constant is very close to 1 year. This also means that the start value of the space ships x coordinate, which is also the x coordinate of the earth is x0 = c2 = 0.97 lightyear g (7.39)
111
ct Earth
Space ship
Figure 7.1: The hyperbolic space-time path of the accelerated space ship. In this Minkowski diagram of the
earth xed frame the path of the earth is the solid green line parallel to the time axis. The path of the space ship, which has constant proper acceleration, denes the part of a hyperbola, shown by the solid blue line in the diagram. The dashed blue line shows the remaining part of the hyperbola. The asymptotes of the hyperbola (red lines) correspond to motion with the speed of light.
In the table the change in position and coordinate time is shown for a sequence of increasing proper times of the space ship. t x x0 1y 1.2 y 0.5 ly 2y 3.6 y 2.8 ly 3y 10.0 y 9.1 ly 5y 74 y 73 ly 7y 548 y 547 ly 11 y 30 000 y 30 000 ly 15 y 1.6 106 y 1.6 106 ly
Table 1: Space-time position of a space ship with hyperbolic motion. The table shows a list of distances
and coordinate times for increasing proper times . For large the distance and coordinate time increase exponentially with the proper time of the space ship.
The numbers shown in the table are quite remarkable. Even if the acceleration as felt in the space ship is quite modest; it is no more than the acceleration of gravity experienced at the surface at the earth, the speed and the distance to the earth increases rapidly. Already after one year, as measured onboard the space ship, the distance to the space ship is half a light year. After a little more than 2 years the space ship will have a distance equal to the distance to our nearest star. Then the velocity really becomes large. After 11 years onboard the space ship it will pass the distance to the center of the galaxy and after 15 years the distance to the Andromeda galaxy. All this is due to the time dilatation effect, or alternatively due to the length contraction effect, since distances between heavenly objects seem to shrink when observed from the space ship. As shown by the Minkowski diagram the speed of the space ship seems to approach the speed of light, so that already after 3 years it has reached a velocity v = 0.995c. The numbers of the table also seem to indicate that space travels even into distant parts of the
112
universe may be possible with a travel time of a few years and under conditions that seem quite agreeable. However, as shown by the corresponding coordinate time on earth, and by comparison with conclusions made in the discussion of the twin paradox, it is clear that if the ship return to the earth it will experience a major jump forward in earth time as compared to the time experienced on board the space ship. There is another major obstacle to carrying out such a travel. If the time dilatation effect should cut down the time of the journey in a substantial way, the space ship has to reach velocities close to the speed of light. This creates a serious energy problem. How should it be possible to feed to the engines the large energies needed? It seems impossible to bring along all this fuel along, even with the most efcient conversion of fuel into energy. So the only possibility seems for the space ship to be recharged with energy during the travel. But the safest conclusion may seem to be that for a space ship to maintain a constant proper acceleration on the time scale of years, even at the modest value of a = g , is outside the reach of any practical setting. However, we shall include a further discussion of this energy problem in the next chapter.
7.2
Relativistic energy and momentum
The relativistic space-time symmetries introduce important changes in the description of energy and momentum as compared to that of non-relativistic physics. Also a new understanding of the energy contained in matter is introduced, as captured by the famous Einstein formula E = mc2 . In this section we examine rst the relation between energy and momentum for a single particle. We next consider consequences of conservation of these physical quantities for systems of particles. We consider a point particle of mass m. When moving, this particle will, in the non-relativistic 2 description, carry momentum p = mv and kinetic energy E = 1 2 mv . For a free particle these are both constants of motion and for a collection of particles these quantities are conserved when we sum over the contributions from all particles. Also in special relativity energy and momentum are conserved, provided we modify the denitions of these quantities. These changes in the denitions of energy and momentum are important only when the velocities approach the speed of light. For small velocities they reduce to their non-relativistic form. To nd the correct relativistic form of energy and momentum we apply the formalism of fourvectors, with the idea to rewrite the non-relativistic three-vector momentum as a relativistic fourvector. The four-vector form makes the expression independent of any particular inertial frame, and if it reproduces correctly the non-relativistic three-momentum in reference frames where v/c is small, this gives a strong indication that the correct relativistic expression has been found. That this is indeed the case has been demonstrated in many ways experimentally in relativistic processes where energy and momentum are conserved. A similar formal approach will later be used when non-relativistic equations are updated to their covariant relativistic form. The natural assumption is to replace the three-vector velocity v by the four velocity U in the denition of the momentum. The expression for the four-momentum will then be P = mU (7.40)
Written as a four-vector the expression for the momentum should be independent of any choice of reference frame. We next consider the non-relativistic limit of this four-vector. It is then convenient to separate the time component from the space component (in an arbitrarily chosen inertial frame) in the same way
7.2. RELATIVISTIC ENERGY AND MOMENTUM as we have earlier done with the four-velocity (see (7.9)), P = (mc, mv)
113
(7.41)
Since approach the value 1 for low velocities, the three-vector part has the correct non-relativistic limit p = mv mv v << c (7.42)
We therefore conclude that the correct three-vector part of the relativistic momentum is mv p = mv = 2 1 v c2
(7.43)
At this point we make a comment on the notations that we apply. When decomposing the four vector and four-acceleration, we write these with capital letters, U = ( U 0 , U) A = ( A0 , A ) (7.44)
This is because the space components of these four vectors are not identical to the three-vectors v and a. Even in the relativistic context the original denitions of velocity and acceleration are valid as the quantities measured in a specic inertial frame, and we therefore make a distinction between these and the three-vector parts of U and A. As far as the momentum is concerned the situation is different. The measured three-vector part is identical to p = mv, and the expression mv is only to be considered as the non-relativistic approximation. For this reason we do not make any distinction between P and p, and use in the following the relativistic denition for p with the old expression valid only for velocities v << c. When we make the transition from non-relativistic to relativistic theory by replacing three-vectors an additional component is introduced, the time component of the vector. It is of interest to understand the meaning of this additional component. For the four-momentum this is mc P 0 = mc = (7.45) 2 1 v c2 To see the physical interpretation we consider its non-relativistic form by making an expansion to rst order in v 2 /c2 , 1 v2 P 0 = mc + m + ... 2 c When multiplied with c this gives 1 cP 0 = mc2 + mv 2 + ... (7.47) 2 The second term is identical to the (non-relativistic) kinetic energy of the particle, while the rst term is a constant with physical dimension of energy. It is called the rest energy of the particle and is here simply a constant. We refer to the full expression as the relativistic energy of the particle, E = mc2 = mc2 1
v2 c2
(7.46)
(7.48)
114 and the expression for the rest energy is
E0 = mc2
(7.49)
Since E0 is a constant we may simply subtract it to get the correct relativistic form for the kinetic energy, T = E E0 = ( 1)mc2 When T is expanded in powers of
v2 , c2
(7.50)
the rst terms are (7.51)
1 3 v4 T = mv 2 + m 2 + ... 2 8 c
So for small velocities the expression for the kinetic energy reduces to the non-relativistic expression, but there are higher order relativistic corrections. However, even if the rest energy here only appears as an innocent looking constant, the formula indicates the presence of a relation between mass and energy, and we know that this relation has farreaching consequences. Mass can be converted to energy, and as we shall discuss that can be seen already in a study of inelastic collisions. But the true signicance is, as we all know, in the eld of nuclear physics, where large amounts of free energy are created by converting small amounts of mass, either in nuclear reactors or in nuclear bombs. The basis for this is the large conversion factor c2 which is present in the rest energy formula. This shows that a small mass of m = 1g is equivalent to a large rest energy E0 = 0.9 1014 J . To sum up, the relativistic four-momentum can be separated in a time component which is the energy of the particle divided by c, and a space component which is the relativistic momentum threevector. The expressions are P=( E , p) = ( c mc 1
v2 c2
mv 1
v2 c2
(7.52)
7.3
The relativistic energy-momentum relation
From the four-moment P we can form the following Lorentz invariant P 2 = P P = p2 A direct calculation gives P2 = 2 m2 v 2 2 m2 c2 v2 = m2 c2 (1 2 ) c = m2 c2 E2 c2 (7.53)
(7.54)
From this follows the relativistic relation between energy and momentum for a freely moving particle E 2 c2 p2 = m2 c4 (7.55)
7.3. THE RELATIVISTIC ENERGY-MOMENTUM RELATION or E= This replaces the non-relativistic relation E= 1 2 p 2m p2 c2 + m2 c4
115
(7.56)
(7.57)
The connection between the two expressions is found by making an expansion in p2 /mc2 , 1 2 p + ... (7.58) 2m which is essentially the same as the expansion (7.47). The rst term is the rest energy and the second term the non-relativistic kinetic energy. The presence of the rest energy in the energy-momentum relation has one important consequence. This is seen by considering the limit m 0. In this limit the expansion in powers of p/mc makes no sense, and that is reected in the difference the limit m 0 makes for the relativistic and nonrelativistic energy. In the relativistic case we get in this limit E = mc2 + E= p2 c2 + m2 c4 cp , p = |p| (7.59)
The limit is well dened and gives an energy which is proportional to the absolute value of the momentum. In the non-relativistic case the limit gives instead 1 2 1 p = mv 2 0 (7.60) 2m 2 where we have assumed that the velocity is nite also in this limit. Since both momentum and energy vanish in this limit, the reasonable conclusion is that the non-relativistic formalism has no place for massless particles. The conclusion is different in special relativity, where the formalism is open for the presence of massless particles. That is fortunate, since nature seems to provide such particles, with the photons being the most well-known example. Let us derive some further consequences for massless particles. We rst note that if m = 0, the relativistic expressions for E and p gives the following expression for the velocity p v = c2 (7.61) E E= which should be compared with the non-relativistic expression v = p/m. In the limit m 0 the relativistic expression gives p v= c (7.62) p which means that in absolute value the speed of the particle is identical to the speed of light. Thus a massless particle always moves with the speed of light, and this is independent of what the energy carried be the particle is. Therefore, we cannot think of a massless particle as being accelerated to the speed of light, it has simply to be born with the speed of light. This is contrasted by the property of massive particles: A particle with mass m = 0 can never reach the speed of light. This is demonstrated by the form of the relativistic energy E= mc2 1
v2 c2
vc
(7.63)
116
7.3.1
Space ship with constant proper acceleration
We return to the situation discussed in Sect. 7.1.1 where a space ship was assumed to perform a spacetime journey with constant proper acceleration far out in the universe. The acceleration would give a monotonic increase in the velocity of the space ship, which then would asymptotically approach the speed of light. In the discussion of the space-time motion we only briey commented on the point that such a journey cannot go on indenitely, since the limitation of available energy will end the journey after a nite time. Let us now consider this limitation in some detail. We assume that the total mass of the space ship at the beginning of the journey is m0 with m1 as the mass of the ship without fuel. Since we do not know what kind of engine the space ship has we only seek an upper limit to its efciency. Let us for simplicity assume that all the mass of the fuel is converted to energy according to the Einstein formula E = mc2 . This energy is used to increase the velocity and therefore the momentum of the space ship. This is done by sending the exhaust gas with maximum momentum in the opposite direction of the velocity of the space ship. The energy momentum relation (7.56) tells us that this happens if massless particles are emitted from the space ship. So we assume that photons are emitted in one direction, and the space ship due to this emission is accelerated in the opposite direction. Let us consider what happens in a short time interval d on the space ship. In this time interval an amount of mass dm is converted to energy, and the photons that carry the energy away also carry an amount of momentum dp = dm c. This gives the same amount of momentum to the ship, but in the opposite direction. In the inertial frame which is the instantaneous rest frame of the space ship at time , the space ship at a little time later, at + d , will have a velocity slightly different from zero. The velocity is dv = dm c/m and this gives for the proper acceleration dv c dm = (7.64) d m d where we have assumed that the proper acceleration is kept xed at the level of the gravitational acceleration on the surface of the earth. We note that this gives a differential equation for the change with time of the mass of the space ship a0 = g = dm mg = d c with an exponential function as solution g m( ) = m0 exp( ) c (7.66) (7.65)
We denote by T the time onboard when all fuel has been consumed, so that m(T ) = m1 . This gives g m1 = m0 exp( T ) (7.67) c If we make the assumption that 90% of the space ships weight at the beginning of the journey is fuel this gives the following (proper) time onboard the ship when it runs out of fuel c T = ln 10 2.3 years (7.68) g The speed of the space ship is then g m2 m2 1 v = tanh( T )c = 0 2 0.98c c m2 0 + m1 (7.69)
7.4. DOPPLER EFFECT WITH PHOTONS and the time dilatation factor is
117
g 1 m0 m1 + )5 (7.70) = cosh( T )c = ( c 2 m1 m0 This is indeed a large velocity and gamma factor. The coordinate time at earth and the distance to the ship is at this point c g c2 g sinh( T ) 5 years , x x0 = cosh( T ) 4 lightyears (7.71) g c g c Even if this does not bring the space ship out to distant galaxies, the distance is still very impressive, comparable to the distance to the closest star. One should however note that the assumptions we have made are rather unrealistic. In particular this is so for the assumption that all mass of the fuel is converted to energy, which should be compared to the efciency of mass conversion of about 1% for the nuclear fusion process where hydrogen is transformed into helium. A more realistic estimate would denitely limit the space travel much more than shown by the numbers above. However, the idea that a rocket engine based on emission of photons could give a constant acceleration over a long time, and thereby bring a space ship in an efcient way far outside the solar system, may not be such a bad idea. t=
7.4
Doppler effect with photons
Even if the speed of a light signal does not change when changing from one inertial reference frame to another, the frequency of the light signal will appear different in the two frames. This is the Doppler effect, which is well known for wave propagation also in non-relativistic physics. The correct relativistic Doppler shift formula can be found by considering light as a propagating wave, but another way to derive it, which is in fact simpler, is to make use of the transformation formula for relativistic four-momenta. This is the approach we take here, when we consider the transformation of fourmomentum for a massless photon between two inertial frames and use the de Broglie relations to translate this to a transformation of frequencies. Let us then consider the situation where a photon is emitted from a space-time point 0 which is the origin of an inertial reference frame S . In this frame the photon momentum is directed with angle relative to the x axis in the x, y plane, p = p(cos i + sin j) Since the photon is massless the components of the four-momentum in this frame are P = p (1, cos , sin , 0) (S frame) (7.73) (7.72)
Let us assume that the photon is absorbed by a detector in another inertial reference frame S that moves with velocity v in the x direction relative to S . The four momentum in this frame has components P = p (1, cos , sin , 0)
0 1 2 3
(S frame)
(7.74)
The components of the two reference frames are related by the Lorentz transformation p p p p = (p0 p1 ) = (p1 p0 ) = p2 = p3 (7.75)
118
which gives p p cos p sin = p(1 cos ) (7.76)
= p(cos ) = p sin
Only two of these are independent equations, as one can readily check, and these two equations can be used to solve for p and cos , p cos = (1 cos )p cos = 1 cos (7.77) (7.78)
The rst one of these gives the Doppler shift formula. To show this we make use of the de Broglie formula which gives the link between the particle and wave nature of the photon, p = E/c = h/c, with as the photon frequency. Eq. (7.77) then gives the frequency transformation formula = (1 cos ) (7.79)
This equation shows how the frequency of a light signal changes between to inertial frames in relative motion. The frame S moves with velocity c relative to S and is the angle between the photon and the relative velocity of the two frames, as measured in S . Clearly the same formula should be applicable if we interchange the two frames. This gives = (1 + cos ) (7.80)
where we have introduced a sign change for the relative velocity. The formula can be re-written as = 1 (1 + cos ) (7.81)
Consistency between (7.79) and (7.81) then gives a relation between the angle measured in the two frames, and this is the same as the equation (7.78). We conclude that the Doppler shift can be expressed either as in (7.79) or in (7.81), depending on whether the angle of the light signal refers to the inertial frame S where it is emitted or the inertial frame S where it is absorbed. We consider now some special cases. a) = 0: The light signal is emitted in the direction of motion of reference frame S . Seen from S the emitter of the signal is moving away from the receiver. The formula is = (1 ) = 1 1+ (7.82)
The light is now redshifted since the frequency in S is lower than in S . b) = : The light signal is emitted against the direction of motion of reference frame S , so that the emitter is moving towards the receiver. The formula is = (1 + ) = 1+ 1 (7.83)
7.5. CONSERVATION OF RELATIVISTIC ENERGY AND MOMENTUM

y
119
S
R
S
E x
Figure 7.2: The Doppler effect. A photon is emitted from a sender E in an inertial frame S at an angle with
respect to the x axis. The photon is received by a receiver R in another inertial frame S which moves with velocity v relative to the rst reference frame. The photon is received at a different frequency in S than the frequency of the emitted photon in the reference frame S . Also the direction of the photon appears different in the two frames, as discussed in the text.
and the light is blue shifted in reference frame S . c) = /2: The light signal is now received with direction orthogonal to the velocity of reference frame S . This gives = 1 (7.84)
Even in this case there is a Doppler shift. We may view this as due to a time dilatation effect, where time in S is seen as slow when viewed from reference frame S . The light signal is redshifted. d) = /2: The light signal is now emitted at 90o degrees in S , and formula is now = (7.85)
The time dilatation effect works the other way, and the light signal is blue shifted. In reference frame S the angle is larger than 90o , as follows from Eq.(7.78). This means that the signal is received with a velocity component against the motion of the frame, which is consistent with the blue shift.
7.5
Conservation of relativistic energy and momentum
An important property of energy and momentum is that these physical quantities are conserved, when we consider the total sum of contributions from all parts of a physical system. This is true both in non-relativitic and in relativistic physics. However, the relativistic form of the conservation laws is different from the non-relativistic one, and there are differences in physical consequences. We will examine these differences.
120
Figure 7.3: A collision process. Two particles are moving freely until they reach a collision area (yellow circular area). As a result of the collision a set of new particles emerge. Relativistic four-momentum is conserved in the process.
Let us consider this for a general collision process, as schematically shown in Fig. 7.3. In this process a set of particles are initially freely moving, but then enter a region of interaction. From this region another set of particles are emerging and these, in the nal state, are again freely moving. The collision process may be elastic, in which case the initial and nal sets of particles are identical, but it may also be inelastic, with the outgoing particles being different from the incoming. For simplicity we assume that radiation can be neglected during the collision process, which means in the relativistic description that we assume that massless particles are not emitted. In the non-relativistic case we may formulate three conservation laws which apply to the collision process. They are pi =
i f
pf p2 f
f
p2 i
i
2mi
2mf mf
+Q (7.86)
mi =
i f
where the index i refers to the incoming particles and f to the outgoing ones. The rst equations states that total momentum is conserved. The second one states that total energy is conserved. This does not mean that total kinetic energy needs to be conserved. In inelastic collisions that is not the case, and Q then measures how much energy that is transformed from kinetic to other forms of energy (internal energy) in such an inelastic collision. In the relativistic setting these conservation laws is replaced by a single four-vector equation, Pi =
i i
Pf
(7.87)
which states that the total four-momentum is preserved in the process. If the non-relativistic limit should be reached in the correct way from this equation, then Eqs.(7.86) should follow from (7.87) when the particle velocities are small compared to the speed of light. We shall check that this is the case. The space component of the four-vector equation has the same form as the non-relativistic equation for conservation of momentum. But the meaning is different because of the relativistic form of
7.5. CONSERVATION OF RELATIVISTIC ENERGY AND MOMENTUM momentum. Expressed in terms of velocities it is i m i vi =
i f
121
f m f vf
(7.88)
where the gamma factors of the particles are missing in the non-relativistic equations. However, in the non-relativistic limit, v/c 0, we have 1 and the relativistic equation reproduces the non-relativistic equation (as it should). We next consider the time component of the four-vector equation (7.87), which we may write as i mi c2 =
i f
f mf c2
(7.89)
Obviously, if the non-relativistic limit also here is taken as 1 the equation will reproduce the non-relativistic mass conservation equation. However, this raises the question how the non-relativistic equation for conservation of energy should be reproduced. The answer is that this is also contained in the time component of (7.88), but we have to keep the rst order contributions in v 2 /c2 when we make an expansion in this small quantity. This gives the following equation 1 2 mi c2 + mi vi = 2 1 2 mf c2 + mf vf 2 (7.90)
By use of the non-relativistic form of the momentum p it can be re-written as p2 i 2mi =

f
p2 f 2mf
+(
f
mf c2
mi c2 )
i
(7.91)
This is seen to have the same form as the non-relativistic energy conservation equation, but with an explicit expression for the Q term, Q=
i
mi c2
mf c2
f
(7.92)
This result is interesting and important. It shows that mass is not conserved in a strict sense in special relativity. Instead, in an inelastic collision, where Q = 0, there will be a mass difference between the initial and nal states, that is determined by the ratio Q/c2 . So mass is in such a process converted to kinetic energy or kinetic energy is converted to mass. (If massless particles are emitted, mass may partly be converted to kinetic energy and partly to radiation.) This gives a concrete, physical interpretation of the rest energy E = mc2 of a massive body. A dramatic application of this relation between mass and energy is in nuclear ssion reactions, where a fraction of the mass of an unstable nucleus is converted to kinetic energy and radiation energy. We consider two examples of such inelastic collisions. The rst is a completely inelastic collision where two bodies collide and create a single larger body. In the process heat is created and we assume that the heat energy is stored in the body as internal energy. Let us for simplicity assume the two bodies before the collision have equal mass m0 . We consider the collision process in the center-of-mass system where the two bodies before the collision have momenta of equal size but with opposite directions. The larger body that is produced in the collision has mass M and sits at rest in this reference frame. The three-vector part of the relativistic momentum
122
conservation equation simply states that the total momentum vanishes in this frame. The conservation equation for energy is 2mc2 = M c2 (7.93)
with as the (common) gamma factor of each of the two colliding particles. The equation shows that the mass of the body that is formed in the collision is larger than the sum of the mass of the two particles before the collisions, m = M 2m = 2m( 1) > 0 When expanded in powers of v 2 /c2 , we have = 1 +
1 v2 2 c2
(7.94)
+ .... This gives for v << c (7.95)
1 2( mv 2 ) = m c2 2
which shows that the kinetic energy of the colliding particles is present, after the collision, as an increase in the mass of the larger body that is formed by the collision. This result is independent of how energy is stored in the body, but in the present case it seems natural to identify Q = mc2 with the heat created by the collision. The mass energy formula in fact suggests quite generally that if a body is heated, the increase in internal energy will lead to an increase in its mass. However, the mass increase obtained by heating the body is under normal conditions extremely small. The inverse of the process considered above is a ssion process where a body is split in two parts by an explosion of some sort. In that case the total mass after the explosion is smaller than the total mass before the explosion, and the missing mass is converted to kinetic energy (and radiation) according to the mass conversion formula E = mc2 . To conclude, in special relativity the total four-momentum of an isolated system is always conserved. This conservation law reduces to the standard expressions for conservation of energy and momentum in the non-relativistic limit. However, a consequence of the relativistic formula is that the total mass is not strictly conserved. The change of mass in a physical process relates to the difference in total kinetic energy and radiation energy in the initial and nal states.
7.6
The center of mass system
Consider a composite system which is isolated from the surroundings so that no external forces act on the system. In non-relativistic physics the center of mass of the system, R, is dened by MR =
k
mk rk
(7.96)
where rk , k = 1, 2, ... denote the position vectors of the small parts of the system, with masses mk , while M = k mk is the total mass of the system. Without external forces the center of mass is non-accelerated and therefore we can nd an inertial reference frame where it is at rest. This is the center-of-mass system, which is then characterized by = P MR k = 0 mk r
k
(7.97)
In the center-of-mass system (the CM system for short) the total momentum of the physical system therefore vanishes.
7.6. THE CENTER OF MASS SYSTEM
123
In relativistic physics the center of mass is not a well dened concept. This can be seen in the following way. In the denition of the CM-position vector R it is essential that the sum (7.96) is performed at equal times for all parts for the system. This is a denition which is independent of choice of inertial frame in non-relativistic physics, since there equal time is a universal concept. However in relativistic physics that is no longer true. If we therefore dene the sum over a three-dimensional space with time coordinate t = constant in one inertial frame, that is different from dening the sum over three-dimensional space with t = constant in another inertial frame. There will in general be no simple relation between the result of to such different summations. As a result we simply give up the idea of dening, in general, the center of mass of an extended system. Even so, the center-of-mass system is both a well dened and useful concept in relativistic physics. This follows from the fact that the total four-momentum P of the system is well dened, and the condition that the space part vanishes, as in (7.97), species an inertial frame which we identify as the center-of mass-system. The condition that identies the center-of mass system therefore is P=
k
pk = 0
(7.98)
which is now written in a form that is correct both in non-relativistic and relativistic physics, but in the latter case one has to remember that for massive particles the right denition of relativistic momentum is p = mv. The total momentum P is a reference-frame independent four-vector, in spite of the fact the sum over contributions from all part of the physical system seems to depend on the choice of frame. This is in fact a consequence of conservation of the total four-momentum for an isolated physical system. For a system of non-interacting particles this is quite clear, since the momentum is conserved for each particle individually. This means that the sum of the particles four momenta will be independent of the points on the particle world lines that are chosen when performing the sum. In particular the result is the same whether they are summed at equal times in one inertial frame or another. For the same reason the sum of the four-vectors will result in a new four-vector. In the general case the same conclusion can be reached by use of momentum conservation expressed as a local conservation law. However, we will not here give a detailed derivation of this result. We conclude that the components of the total four-momentum of an isolated system transform as components of a four-vector, and since it is a time-like vector an inertial frame can always be found where the space part of the vector vanishes. This is the center-of-mass system, and it is a unique reference frame up its orientation in space.
124
Chapter 8
Relativistic dynamics
In non-relativistic physics Newtons second law is the basic dynamical equation. It is not valid in relativistic physics, unless some changes are introduced. We will here examine the question of how to correctly update the law to relativistic form. Our approach is based on the general idea of re-writing the non-relativistic equation in a covariant relativistic form. This means that we express the equation in terms of four-vectors and tensors, in such a way that it has the correct non-relativistic limit for low velocities v << c. The covariant form secures that the equation is valid in all inertial frames. Whether the equation is really correct is at the end a question to check experimentally, but at least the formal properties demanded by relativistic invariance will be satised by this approach.
8.1
Newtons second law in relativistic form

dp dt
Our starting point is the (non-relativistic) Newtons second law, which we write as F= (8.1)
with p = mv. It is is here assumed to apply to a small body (point particle) which carry momentum p and is subject to a force F. This is a three-vector equation, which in relativistic form should be generalized to a four-vector equation. As an obvious attempt to do so we write the it in the following relativistic form K= dP d (8.2)
So the non-relativistic momentum is replaced by the relativistic four-momentum and coordinate time is replaced by proper time of the particle. The time derivative of the four vector then is also a fourvector. On the right-hand side we have simply replaced the three-vector force F with a four-vector K which we refer to as the four-force. We shall examine what constraints that physics puts on this vector, but for the moment we just note that the new equation has a correct covariant form. The equation can also be written as K = mA (8.3)
with m as the (rest) mass of the particle and A as the proper acceleration. This follows since P = mU P dU and therefore d d = m d with U as the four-velocity of the particle. Note however, since the threevector part of A is generally not identical to the acceleration a in a chosen inertial frame, the threevector part of the right-hand side of (8.3) is not simply ma. 125
126
CHAPTER 8. RELATIVISTIC DYNAMICS
The next step is to relate the four-force K to the (non-relativistic) three-vector force F. To this end we decompose the four-vector in its time and space components, with reference to some unspecied inertial reference frame, K = (K 0 , K) The three-vector part of the equation (8.2) is then K= dp dp = d dt (8.5) (8.4)
with p = mv as the relativistic momentum. The factor that appears in the equation is a time dilatation effect. Let us now return to the original form (8.1) of Newtons second law, and assume that also in the context of relativistic physics the three vector force F is dened so that (8.1) is correct. However, since p should then be the relativistic momentum, then generally F = ma. This means that the nonrelativistic form of Newtons second law is valid also in relativistic theory, but only when expressed p in terms of d dt and not when expressed in terms of a. It is then clear that the correct non-relativistic equation is obtained when 1 and p therefore is changed from its relativistic to its non-relativistic form. We have now established the relation K = F and we examine further the time component of K, K0 = 1 dE dP 0 = d c dt (8.7) (8.6)
The relativistic energy-momentum relation is E 2 = p2 c2 + m2 c4 and the time derivative of this equation gives E This further gives dE p dp = c2 =vF dt E dt (8.10) dE dp = c2 p dt dt (8.9) (8.8)
where we have made use of the relativistic relation v = c2 p/E . It is interesting to note that with the relativistic generalization introduced for the three-vector force F, the expression for the power is v F, precisely as in non-relativistic physics. The four-force, when decomposed in time and space components can then be written as 1 K = ( v F, F) c (8.11)
While the space part is proportional to the three-vector force F, the time component is proportional to the power of the force, v F.
8.1. NEWTONS SECOND LAW IN RELATIVISTIC FORM
127
One should note from the above expressions that even when the three-vector force F is a velocity independent force, the four-force K will quite generally depend on the velocity of the particle. This point is also demonstrated by the fact that the four-force, which is proportional to the four-acceleration, is always orthogonal, in the relativistic sense, to the four-velocity, KU=AU=0 (8.12)
We further note that since U is a time-like vector this implies that the K is a spacelike vector. From this follows that we can always nd an inertial frame where the time component of the force vanishes. The expression (8.11) shows that this happens in the instantaneous rest frame of the particle, where the four-force reduces to the form K = (0, F) (restframe) (8.13)
8.1.1
The Lorentz force
We shall consider, as a special case, the force that acts on a charged particle in an electromagnetic eld, and derive the corresponding covariant form of the equation of motion. The non-relativistic form of the tree-vector force is F = e(E + v B) (8.14)
with e as the charge and v as the velocity of the particle, and this is in fact a valid expressions also for relativistic velocities. The corresponding expressions for the time and space components of the four-force are 1 K = e( v E, E + v B) c (8.15)
We would now like to write this in covariant form, and that requires that we introduce the electromagnetic eld tensor. This is an antisymmetric tensor where the electric eld appears as the time components and the magnetic eld as the space components in the following way, F 0k = F kl =
m
1 Ek c
k = 1, 2, 3
klm Bm
k, l = 1, 2, 3
(8.16)
where klm is the three-dimensional Levi-Civita symbol and the Cartesian components are here numbered 1,2,3. In matrix form the eld tensor is 1 1 1 0 c E1 c E2 c E3 1 E1 0 B3 B2 c F = (F ) = (8.17) 1 E2 B3 0 B1 c 1 B2 B1 0 c E2 Constructed in this way F transforms indeed as a relativistic tensor under Lorentz transformations. We consider rst how the time component of the four-force can be expressed in terms of the electromagnetic eld tensor. The expression is K 0 = e v E = eF 0 U c (8.18)
128
with U as the four-velocity of the particle. The space part we re-write in a similar way, K k = e (Ek +
lm
vl Bm ) +
l
= e (cF We next make use of the following identities, U 0 = U0 = c , This gives K k = e(F k0 U0 +
l
0k
F kl vl )
(8.19)
U l = Ul = vl ,
F 0k = F k0
(8.20)
F kl Ul ) = eF k U
(8.21)
We can now combine the expressions for K 0 and K k into a single equation for the components of the four-force, K = eF U (8.22)
With the above expression for the four-force the equation of motion of the charged particle can be re-written in covariant form. The general relativistic form of Newtons second law, K = mA = m here gives the equation mx = eF x (8.24) d2 x d 2 (8.23)
where the time derivative marked by the dot here means differentiation with respect to the proper time of the particle. We nally note that this covariant equation is equivalent to the non-covariant equation of motion, dp = e(E + v B) dt (8.25)
which has the same form in the non-relativistic limit. However, in the relativistic case the right-hand side of the equation cannot simply be replaced by ma, due to the presence of the gamma factor in the expression for the momentum, p = mv.
8.1.2
Example: Relativistic motion of a charged particle in a constant magnetic eld
We rst briey examine the non-relativistic motion of the charged particle in the constant magnetic. This has in Sect.3.3.1 been done by use of Hamiltons equations. Here we derive the motion in a more direct way from Newtons second law. The equation of motion is ma = ev B (8.26)
with e as the charge of the particle. There is no force along the direction of the eld, and the motion in this direction is therefore a constant drift. For simplicity we assume initial conditions with no velocity
8.1. NEWTONS SECOND LAW IN RELATIVISTIC FORM
129
in this direction, and the motion is therefore restricted to a plane orthogonal to B. The velocity is also restricted to this plane, and the equation of motion shows that the force is orthogonal to the velocity, 2 so that a v = 0. Consequently the kinetic energy is conserved, T = 1 2 mv is constant. There is another constant of motion which can be derived from the equation of motion. The equation can be written as d (mv er B) = 0 dt which gives the following constant of motion, mv er B er0 B (8.28) (8.27)
with r0 as a constant vector. The form we have chosen for the constant vector, on the right-hand side of the equation, is consistent with the fact that this vector is restricted to the plane orthogonal to B when v has no component along the magnetic eld. We rewrite the equation as mv e(r r0 ) B = 0 (8.29)
and this shows that r0 can be absorbed in a shift of the origin in the plane of motion. We assume in the following that to have been done, so the motion satises the equation mv = er B (8.30)
This equation shows that the velocity is orthogonal to the position vector, r v = 0, so that r2 is a constant of motion. From the arguments given above we conclude that the particle moves in a circle with constant velocity. With r0 = 0 the center of the orbit is at the origin of the coordinate system, but without this restriction the center can be placed anywhere in the plane. The circular frequency is = v eB = r m (8.31)
2 2 which is the cyclotron frequency. The kinetic energy is T = 1 2 m r which shows that the radius increases with energy so that T is proportional with r2 . The expressions given above are valid for non-relativistic velocities, v << c, which means that 1 T << 2 mc2 . That restricts the kinetic energy to be much smaller than the rest energy of the particle. In the following we will lift this restriction and study how the motion changes when we use the correct relativistic equation of motion. The relativistic equation of motion can be written as
dp = ev B dt
(8.32)
2 with p = mv. The power of the force vanishes, dE dt = F v = 0 , which means that mc is constant. As a consequence the velocity is constant in the relativistic description as well as in the non-relativistic approximation. There is also a constant of motion corresponding to (8.28), but now written as
mv er B er0 B
(8.33)
130
We may also in this case chose to shift the origin in order to have r0 = 0. The equation is then mv = er B (8.34)
which is a relativistic generalization of (8.30). This shows that v r = 0 and therefore r is a constant, just as in the non-relativistic case. We conclude that in the relativistic case, just as in the non-relativistic case the charged particle moves with constant speed in a circle. There is however a difference between the two cases as far as the circular frequency is concerned. Eq.(8.34) gives mv = erB from which follows = eB 0 m (8.36) (8.35)
where we have introduced the symbol 0 = eB/m for the non-relativistic cyclotron frequency. This shows that in the relativistic case the circular frequency decreases with the speed, and therefore with the energy of the particle. We can also nd the frequency as a function of the radius of the orbit if we write the equation as = 0 Solving this for we nd = 0 1+
2 r2 0 c2
2 r2 c2
(8.37)
(8.38)
The corresponding expression for the relativistic energy of the particle is E= 1+

2 r2 0 mc2 c2
(8.39)
As shown by this expression the energy is a quadratic function of r for non-relativistic velocities but this changes to a linear dependence in the relativistic regime. The decrease in circular frequency with velocity or energy of the circulating charge is important in a type of particle accelerator called cyclotrons. In these accelerators charged particles are circulating in a strong magnetic eld and energy is fed to the particles by applying an electric eld which oscillates with the circular frequency of the particles. In the early cyclotrons where the particles moved nonrelativistically the frequency of the eld was kept xed. Later on accelerators were built to accelerate beams of particles to relativistic speeds. In these accelerators, called synchro-cyclotrons, the frequency was synchronized with the decreasing circular frequency of the accelerated particles. In isochronous cyclotrons a different approach is taken to compensate for the relativistic effect. These accelerators work with constant electric eld frequency, but the strength of the magnetic eld is increased with time. As shown by Eq.(8.36) the circular frequency of the circulating particles can be kept xed by compensating for the increase of by a similar increase in the value of B .
8.2. THE LAGRANGIAN FOR A RELATIVISTIC PARTICLE
131
8.2
The Lagrangian for a relativistic particle
In the Lagrangian formulation of Newtons mechanics introduced earlier in the course, the time coordinate plays a different role than the generalized coordinates of the system. Time is there a parameter for the path of the system through conguration space. For a particle moving through three dimensional space this means that the space coordinates and the time coordinate appear in different ways in the formalism. This difference seems to create a problem when extending the formalism to relativistic theory, where time and space coordinates are mixed by the Lorentz transformations. We consider the question of how to introduce a relativistic particle Lagrangian, and in order to do so we make an attempt to follow the same approach as in the previous section, where relativistic generalizations of physical formulas were introduced by re-writing the equations in covariant form. We apply this rst to a freely moving particle of mass m. Instead of considering the Lagrangian directly we start with the action, which is the integral of the Lagrangian for an arbitrarily chosen time dependent path in conguration space. For a free particle it has the form S= T dt = 1 2 dt mr 2 (8.40)
with T as the kinetic energy of the particle. As a rst step in a relativistic modication of the integral we note the following correspondence T dt Edt = cP 0 dt (8.41)
where the right-hand side has the left-hand side as a non-relativistic limit, except for the presence of the rest energy E0 = mc2 . This immediately suggests that one could add a term to make the expression Lorentz invariant T dt P dx P dx = P dr cP 0 dt = (P v cP 0 )dt 1 1 (mv 2 mc2 mv 2 ...)dt = ( mv 2 mc2 ...)dt 2 2 (8.42)
However, we have to check the non-relativistic limit when the term P dr has been added. We have (8.43)
where ... denotes higher order terms in v 2 /c2 which are omitted in the non-relativistic approximation. We note that the term we have added has in fact changed the sign of the non-relativstic kinetic energy. To compensate for this we change the sign in the correspondence (8.42). The contribution from the rest energy is of no importance, since a constant contribution to the Lagrangian is irrelevant for Lagranges equations. As a result we have found the following expression for the action integral which is Lorentz invariant and has the correct non-relativistic limit, S= P dx = m U dx = m U U d = mc2 d (8.44)
We have here applied the identities P = mU , dx = U d, U U = c2 , with as the proper time of the particle. It may initially seem very natural that in the Lorentz invariant form the proper time should be used as a time parameter rather than the coordinate time. However, that creates a problem which is
132
seen from the expression we have found. With as a time parameter it looks like the Lagrangian is a constant, and that does not make sense. The reason for this problem can be understood in the following way. If the equation of motion should be derived from Hamiltons principle, we should consider changes in the action under variations in the space-time path with xed end points. That leads to Lagranges equations, in the way we have discussed, provided the time parameter has xed values at the end points. That is not the case with the proper time. As we have discussed, and demonstrated in the twin paradox, the difference in proper time between two space-time points is path dependent, so it will change when the path is changed even if the end points are xed. The conclusion we draw is that the expression we have found for the action may be good, but the proper time is not a good parameter to use if we want to derive the equation of motion from the action integral. To circumvent the problem we may simply introduce a new, unspecied parameter , which is assumed to take xed values at the end points of any given space-time path when we perform variations in the path with the end points xed. For this new parameter we nd the following expression cd = dx dx = dx dx d d d (8.45)
where one should note that the minus sign under the square rote is simply to compensate for the fact that dx is a timelike vector for the path of the particle. The expression given by the equation is clearly invariant under arbitrary changes in the parameter, . The action can then be written as S = mc2
1 0
g x x d
(8.46)
dx d .
where the parameter values at the end points are called 0 and 1 and where we here dene x = The corresponding Lagrangian is then L = mc g x x and Lagranges equations have the standard form d d L x L x = 0, = 0, 1, 2, 3
(8.47)
(8.48)
It should be straight forward to check whether the equations found in this way are the correct L equations of motion for a free particle. We rst note that all the coordinates x are cyclic, x = 0, so that we have the following set of constants of motion L x = mc k x x x where k satises the condition k k m2 c2 . This gives k = mc n (8.50) with n as a timelike unit vector, n n = 1. If we now introduce the four velocity, U = x d d , Eq.(8.49) implies, U = c n (8.51) (8.49)
133
which shows that the four velocity is a constant timelike vector with relativistic norm squared, U U = c2 . This is clearly the correct expression for a free particle which moves with constant velocity. Next we consider the Lagrangian of a charged particle in an electromagnetic eld. This can be obtained from the free eld Lagrangian by simply adding a contribution from the electromagnetic potentials. The Lagrangian has the following Lorentz invariant form L = mc g x x + eA x (8.52)
where the scalar potential and vector potential A are identied as the components of the fourpotential A = ( 1 c , A). It is again a straight forward exercise to check that the corresponding Lagranges equations give the correct relativistic equations of motion in the relativistic form (8.24). Let us nally point out that even if covariant expressions have been used in order to construct the Lagrangians with correct relativistic form, there is no problem to re-express them with the coordinate time as parameter, for any chosen inertial frame. After all the parameter can be freely chosen and in particular chosen to coincide with such a coordinate time. The main point is that the action we have found does not depend on the choice of parameter. In particular this means that if the coordinate time is chosen, the action of a free particle should be written as S = mc d = Ldt (8.53)
dt By use of the time dilatation formula d = this gives the following expression for the Lagrangian, when the coordinate t is chosen as parameter
L=
mc2 = mc2
v2 c2
(8.54)
We note that this expression, even if there is some resemblance, is not identical to the energy E = mc2 . In this regard it is different from the non-relativistic Lagrangian. For a charged particle in an electromagnetic eld the corresponding Lagrangian is L = mc2 1 v2 e + ev A c2 (8.55)
The expressions (8.54) and (8.55) are not Lorentz invariant, but nevertheless correct expressions for the relativistic Lagrangians, as our derivation has shown.
134
Summary
This part of the lectures has focussed on some of the basic elements of the special theory of relativity. The starting point has been the fundamental space-time symmetries expressed in the form of Lorentz transformations. They dene the transition between Cartesian coordinates of different inertial frames, and the basic difference between these transformations and the Galilean symmetry transformations of non-relativistic physics is the mixing of space and time coordinates in the relativistic case. An important consequence of this is that distance in non-relativistic theory, in the form of the length of the three-vector r, is replaced by another invariant which also includes the time difference, s2 = r2 c2 t2 . Invariance of this quantity is directly related to the basic property of Lorentz transformation, that the speed of light does not depend on the choice of reference frame. The change from Galilean to Lorentz transformations as the fundamental symmetry transformations has many important consequences for relativistic kinematics and dynamics, as rst demonstrated by Albert Einstein. We have here derived the kinematical effects of length contraction and time dilatation and have stressed the important point that measurements should be performed at simultaneity of the observer who makes the measurement. For the time dilatation effect this understanding is applied to the twin paradox, which is resolved by taking into account the change in denition of simultaneous events that is performed by one of the twins during his space-time journey. An introduction to the formalism of four-vectors and tensors has been given, and we have discussed how to apply this formalism when dening covariant relativistic equations. The formalism has been used at several places to derive the relativistic expressions that correspond to known nonrelativistic quantities. The idea is to seek covariant expressions, which secures that the expressions are valid in any inertial frame, and to impose the condition that the expressions have the correct nonrelativistic limit. This formal approach indeed produces correctly the relativistic extensions of the non-relativistic expressions for the physical quantities and equations. In particular we have introduced the four-vector description of velocity and acceleration, and we have discussed the meaning of proper acceleration. As an example we have studied the so-called hyperbolic motion of a space ship with constant proper acceleration and effects of the time dilatation for time registered on the space ship and on earth. The denition of the conserved four-momentum has been shown to have important consequences. The energy and three-vector momentum are there composed into a four component object, and Lorentz invariance imposes a particular form to the relation between energy and momentum for a moving object. This involves in particular a conversion formula between mass and energy, which is Einsteins famous equation E = mc2 . By considering the case of inelastic collisions between particles we demonstrate that this relation is not only a curious coincidence, but it shows that mass can be converted to energy in real physical processes, with a large conversion factor between mass and energy. This points to the well-known and dramatic effect of releasing huge amounts of energy in nuclear processes. In the chapter on Relativistic Dynamics we have examined how to update Newtons second law to a relativistic equation and to give meaning to the four-vector force. As a particular application we have 135
137
examined how to give the equation of motion for a charged particle in an electromagnetic eld the correct covariant form. Finally we have discussed how to bring relativistic equations into Lagrangian form and have shown how to resolve the problem which appears when using proper time as the time parameter. The approach has been illustrated by deriving Lagrangians for a free particle and for a charged particle in an electromagnetic eld, both in the covariant and the non-covariant forms.
138
Part III
Electrodynamics
139
Introduction
Each of the three parts of this course is associated with one or two scientists that have played a particularly important role in developing the theory, and who have in this way put their nger prints in a lasting way on the development of the science of physics. In the rst part on analytical mechanics the key gures were Lagrange and Hamilton and in the second part on relativity the central person was Einstein. In this third part of the course, on electrodynamics, the physicist that played the most decisive part in developing the theory was James Clark Maxwell (1831 - 1879). Maxwell collected and modied the equations that are now known as Maxwells equations, and in this way built the foundation for the classical theory of electromagnetism. On the basis of these equations it was shown convincingly that light is an electromagnetic wave phenomena, and that the many other electric and magnetic phenomena can be understood as different realizations of the underlying fundamental theory of electromagnetism. The intention of this part of the lectures is to analyze the fundamentals of electrodynamics on the basis of Maxwells equations. We begin by discussing the non-relativistic form of the equations and then show how to bring them into relativistic, covariant form. The use of electromagnetic potentials is important in this discussion and the following applications. The idea is to examine solutions of Maxwells equations under different types of conditions. This include solutions of the free wave equations, solutions with stationary sources and solutions to the equations with time dependent charge and current distributions. The expansion in terms of multipoles is important in this discussion, and we put emphasis on the study of the radiation phenomena. The discussion here is restricted to some of the fundamental (and elementary) aspects of Maxwell theory. They are important aspects of the theory, but most of the further interesting (but more demanding) ones are left out. Thus we do not consider effects of special boundary conditions and we do not (in this rst form of he lecture notes) include a discussion of electromagnetism in polarizable media.
141
142
Chapter 9
Maxwells equations
In this chapter we establish the fundamental electromagnetic equations. Historically they were developed by studying the different forms of electric and magnetic phenomena and rst formulated as independent laws. These phenomena included the creation of electric elds by charges (Gauss law) and by time dependent magnetic elds (Faradays law of induction), and the creation of magnetic elds by electric currents (Amp` eres law). We will rst recall the form of each of these individual laws and next follow the important step of Maxwell by collecting these in a set of coupled equations for the electromagnetic phenomena. Maxwells equations, which gain their most attractive form when written in relativistic, covariant form, is the starting point for the further discussion, where we examine different types of solutions to these equations.
9.1
Charge conservation
The electric and magnetic elds are produced by charges, either at rest or in motion. These charges satisfy the important law of charge conservation. This is a law that seems to be strictly satised in nature. The carriers of electric charge, at the microscopic level these are the elementary particles, may be created and may disappear, but in these processes the total charge is always preserved. With Q as the total electric charge within a given volume, charge conservation may simply be expressed as dQ =0 dt (9.1)
However, this equation is correct only when there is no charge passing through the boundary surface of the selected volume. A more general expression is therefore dQ = I dt (9.2)
where I is the current through the boundary surface. This equation for the integrated charge and current can be reformulated in terms of the local charge density and current density j, dened by Q(t) =
V
(r, t)d3 r ,
I (t) =
S
j(r, t) dS
(9.3)
In these expressions V is the (arbitrarily) chosen volume with S as the corresponding boundary 143
144
CHAPTER 9. MAXWELLS EQUATIONS
dS I S Q V
Figure 9.1: Charge conservation: The change in the charge Q within a volume V is caused by a current I through the boundary surface S . surface, d3 r is the three dimensional volume element and dS is the surface element, with direction orthogonal to the closed surface. Charge conservation then gets the form d dt (r, t) d3 r +
V S
j(r, t) dS = 0
(9.4)
The last term can be re-written as a volume integral by use of Gauss theorem, and this gives the following integral form of the equation ( j(r, t) + (r, t)) d3 r = 0 t (9.5)
Since charge conservation in the form (9.5) is valid at any time t and for an arbitrarily small volume centered at any chosen point r, it can be reformulated as the following local condition on the charge and current densities +j=0 t (9.6)
This form for the condition of charge conservation, as a continuity equation, we will later apply repeatedly. When expressed in terms of densities we have a view of charge as something continuously distributed in space. However, we know that at the microscopic level charge has a granular structure, since it is carried by small (pointlike) particles. We may take the view that the continuum description is based on a macroscopic approximation where the local charge is averaged over a volume that is small on a macroscopic scale but sufciently large on the microscopic scale to smoothen the granular distribution of charge. In most cases this will be sufcient for our purpose. However, the description of charged particles can also be included by use of Diracs delta function. For a system of pointlike particles the charge density and current density then take the form (r, t) =
i
ei (r ri (t)) ei vi (t) (r ri (t)) (9.7)
j(r, t) =
i
In these expressions the label i identies a particle in the system, with charge ei , time dependent position ri (t) and velocity vi (t).
9.2. GAUSS LAW
145
In general, when the motion of the charged particles can be described by a (smooth) velocity eld v(r, t), we have the following relation between current and charge densities j(r, t) = v(r, t)(r, t) (9.8)
Note, however, for currents in a conductor there are two independent contributions, from the electrons and from the ions, and these move with different velocities, ve and va , so that the total current has the form j(r, t) = ve (r, t)e (r, t) + va (r, t)a (r, t) (9.9)
For the usual situation, with total charge neutrality and with the ions sitting at rest, the expressions for total charge and current densities are (r, t) = 0 , j(r, t) = ve (r, t)e (r, t) (9.10)
9.2
Gauss law
This law expresses how electric charge acts as a source for the electric eld. As all of the electromagnetic equations it can be given an integral or differential form. In integral form it relates the ux of the electric eld through any given closed surface S to the total charge Q within the surface, Q E dS = (9.11)
S 0
In this equation
is the permittivity of vacuum, with the value

0
= 8.85 1012 C2 /Nm2
(9.12)
Eq.(9.11) can be rewritten in terms of volume integrals as E dV = dV

V V 0
(9.13)
where on the left hand side Gauss theorem has been used to rewrite the surface integral as a volume integral. Since this equality should be satised for any chosen volume V , the integrands should be equal, and that gives Gauss law in differential form E= (9.14)
0
Gauss law is the fundamental equation of electrostatics, where the basic problem is to determine the electric eld from a given charge distribution, with specied boundary conditions satised by the eld. In its simplest form the problem is to determine the eld from an isolated point charge, in which case Gauss law in integral form can easily be solved under the assumption of rotational symmetry. Thus with the charge located at r = 0 and with the electric eld of the form E = E r/r, Gauss law gives q 4r2 E = (9.15)
0
with q as the charge. This gives the expression for the Coulomb eld of a stationary point charge q r E(r) = (9.16) 4 0 r 2 r Due to the fact that Gauss law gives a linear differential equation for the electric eld, the solution for a point charge can be extended to the full solution for a charge distribution. We shall return to the discussion of stationary solutions of Maxwells equations later on.
146
9.3
Amp` eres law

1 d c2 dt
This law expresses how electric currents produce a magnetic eld. The integral form is B ds = 0 I + E dS (9.17)
and it shows that the line integral of the magnetic eld around any closed curve C gets two contributions, one from the total electric current I passing through C and the other from the displacement current, which is dened by the time derivative of the electric ux through a surface S with C as boundary. In this equation the vacuum permeability has been introduced. The value of this constant is given by 0 = 107 N/A2 4 (9.18)
To re-phrase this in differential form, the left-hand-side is rewritten as a surface integral by use of
I C S B dl
Figure 9.2: Amp` eres law: The circulation of the magnetic eld B around a closed curve C is determined by
the current and the time derivative of the electric ux through the loop.
Stokes theorem and the current is expressed as a surface integral of the current density ( B) dS = 0 j dS + 1 d c2 dt E dS (9.19)
Since this should be satised for an arbitrarily chosen surface S we conclude there is a pointwise equality which gives Amperes law in differential form B 1 E = 0 j c2 t (9.20)
Amperes law shows that an electric current gives rise to a magnetic eld that circulates the current, but it also shows that a changing electric eld produces a magnetic eld. The origin of the time derivative of the electric eld in the equation may not be so obvious, but this term, which was introduced by Maxwell, is important for the set of equations to be consistent and to have solutions in the form of propagating waves. Thus the propagation of the wave is based on the properties of the elds that time variations in E will produce a magnetic eld B (Amperes law) and, at the next step, the time variations in B will re-produce the E eld (Faradays law). Another interesting point to notice is that without the contribution from the time derivative of E, the equation (9.20) would be in conict with the conservation of electric charge. This is seen by
9.4. GAUSS LAW FOR THE MAGNETIC FIELD AND FARADAYS LAW OF INDUCTION147
taking the divergence of the equation, which would without the electric term give rise to the equation j = 0 for the current. However, by comparison with the continuity equation for the charge current one sees that this is correct only if the charge density is not changing with time. The form of the electric term is in fact precisely what is needed to reproduce the continuity equation, provided there is a specic connection between the constants 0 and 0 . To demonstrate this we take the divergence of Eq.(9.20) and apply Gauss law, 0 j = 1 1 E= 2 c2 t c 0 t 1 c2 (9.21)
This is precisely the continuity equation when

0 0
(9.22)
which is indeed a correct relation. This shows that conservation of electric charge is not a condition that should be viewed as being independent of the electromagnetic equations. It can be derived from the laws of Gauss and Ampere, and can therefore be seen as a consistency requirement for these two electromagnetic equations.
9.4
Gauss law for the magnetic eld and Faradays law of induction
An important property of the magnetic eld is that there exist no isolated magnetic pole. This means that the total magnetic ux through any closed surface S vanishes,
B
Figure 9.3: Gauss law for magnetic elds: The total magnetic ux through any closed surface S is zero. Therefore magnetic ux lines have no end points. B dS = 0
(9.23)
and in differential form this gives B=0 (9.24)
It has a similar form as Gauss law for the electric eld, but in this case there is no counterpart to the electric charge density. Expressed in terms of eld lines, this means that magnetic eld lines are always closed, whereas the electric eld lines may be open, with end points on the electric charges.
148
The Faraday induction law states that the integral of the electric eld around a closed curve C is determined by the time derivative of the magnetic ux through a surface S with C as boundary E ds = d dt B dS (9.25)
There is an obvious similarity between this equation and Amperes law, with the electric eld interchanged with the magnetic eld. By use of the same method as in the discussion of Amperes law, we rewrite the equation in differential form, E+ B=0 t (9.26)
The main difference is that there is no counterpart to the electric current in this equation. We also note from this equation the electric eld will in general not be a conservative eld. Faradays law of induction describes the important phenomenon of induction of an electric eld by a variable magnetic eld. This effect is the basis for electromagnetic generators, where mechanical work is transformed into electric energy.
9.5
Maxwells equations in vacuum
Maxwells equations consists of the four coupled equations for the electromagnetic eld that we have discussed separately, 1. 2. 3. 4. E= B
0
1 E = 0 j c2 t
B=0 E+ B=0 t (9.27)
These equation show how electromagnetic elds are produced by electric charges and currents. They should be supplemented with the continuity equation for charge +j=0 t (9.28)
which however, as we have seen, does not appear as an independent equation, but rather as a consistency condition for Maxwells equation. They should also be supplemented with the equation of how the electromagnetic elds act back on the charges, here in the form of the equation of motion for a charged particle dp = e(E + v B) dt (9.29)
with p as the (mechanical) momentum of the particle. Together these equations form a closed set that describe the complete dynamics of a the physical system of electromagnetic elds and charged particles.
9.5. MAXWELLS EQUATIONS IN VACUUM
149
Maxwells equations have several interesting symmetry properties. One of these is the symmetry under Lorentz transformations. This symmetry of the equations was found even before special relativity was formulated as a theory. There is an obvious conict between the equations and the old, Galilean principle of relativity, since the they involve a constant with the dimension of velocity, namely the velocity of light c. This is problematic for the old principle of relativity, since the transformation from one inertial frame of reference to another will change all velocities. This conict was resolved only when Einstein formulated his daring proposal that time and space have to be viewed together as a unity, the four-dimensional space-time, and that a change of reference frame would transform not only the space coordinates but also time. Maxwells equations dene in fact a fully relativistic theory, developed before Einstein formulated the theory of relativity. This is seen most clearly when the equations are formulated in the language of four-vectors and tensors. Another symmetry that is clearly seen in Maxwells equations is the symmetry under an interchange of electric and magnetic elds. In fact, without the source terms, in the form of the charge or current densities, equations 1. and 2. are changed to equations 3. and 4. (and vice versa) by the following change in the elds, E cB and cB E1 . Even with sources there are symmetries between the electric and magnetic equations, and this can be exploited when solving problems in electrostatics and magnetostatics. There have been speculations in the past whether the symmetry in Maxwells equations between E and B should be extended to the general form of the equations, by including source terms also for the equations 3. and 4. The lack of sources for these equations in (9.27) can be understood as reecting the lack of magnetic monopoles in nature, since magnetic poles seem always to appear in pairs of opposite sign. However, the existence of magnetic monopoles in the form of magnetic charges carried by new types of elementary particles can not be fully excluded. To take this possibility into account in Maxwells equations would mean to include source terms also in equations 3. and 4., in the form of magnetic charge and current densities. In that case there would be two different types of sources for the electromagnetic eld, electric charges and currents, and magnetic charges and currents. There have been performed several experimental searches for elementary particles with magnetic charge, but so far with negative results. We shall here proceed in the usual way, by assuming that no magnetic charges and currents exist, and therefore by keeping Maxwells equations in the standard form (9.27).
9.5.1
Electromagnetic potentials
When the possibility of magnetic charges have been excluded and equations 3. and 4. therefore are homogeneous, the electric and magnetic elds can be expressed in terms of the electromagnetic potentials. These are referred to as the scalar potential and the vector potential A, and they are dened by E = A , t B=A (9.30)
These expressions depend on the fact that Maxwells equations 3. and 4. are source free and in fact make these two equations satised as identities, as one can readily check. The use of electromagnetic potentials therefore effectively reduce the set of eld equations to 1. and 2. In addition to reducing the number of eld equations, the use of potentials is helpful when solving the eld equations.
1 This transformation, which is referred to as a duality transformation, is a special case of eld rotations of the form E cos E + sin cB and B cos B sin E/c. Without the source terms and j, Maxwells equations are invariant under general transformations of this form.
150
CHAPTER 9. MAXWELLS EQUATIONS Expressed in terms of the potentials Maxwells equations get the form 1. 2. A= t 0 1 2 1 2 A + ( A) 2 2 2 A = 0 j c t c t 2 +
(9.31)
These equations can be simplied further by imposing certain gauge conditions on the potentials. Gauge transformations are transformations of the potentials that leave the electromagnetic elds unchanged. They have the form t A A = A + =
(9.32)
where = (r, t) is an arbitrary (well behaved) differentiable function of the space and time coordinates. It is straight forward to check that such a transformation will not change E or B, E E = A =E+ = E t t t (9.33)
B B = A = B + = B
The usual understanding is that gauge transformations do not correspond to any physical operation, since they leave E or B unchanged, but only reects a certain freedom in the choice of electromagnetic potentials which represent a given electromagnetic eld conguration. This freedom can be exploited by making specic gauge choices in the form of conditions that the potentials should satisfy. Two commonly used gauge conditions are the following 1) A = 0
Coulomb gauge Lorentz gauge
(9.34) (9.35)
2) A
= 0
The Lorentz gauge condition has a covariant form when expressed in terms of the 4-vector potential with components A . It is dened by A = ( 1 c , A) so that the scalar potential is (up to a factor 1/c) the time component and the vector potential is the space component of the 4-potential. This gauge condition is often used when it is important to keep the relativistic form of the equations. The Coulomb gauge condition, on the other hand is often used when charged particles, such as electrons in atoms, move with non-relativistic velocities and the relativistic form of the equations is not so important. Also other types of gauge conditions can be imposed in order to simplify the electromagnetic equations, but it is important that the constraints they impose on the potentials should not correspond to any constraint on the electromagnetic elds E and B.
9.5.2
Coulomb gauge
For the Coulomb and Lorentz gauge conditions one can show explicitly that these only affect the choice of potentials, but do not constrain the electromagnetic elds E and B in any way. Let us consider how this can be demonstrated for the Coulomb gauge. Assume A is an arbitrary vector potential, that does not satisfy the Coulomb gauge condition. We will change this to a vector potential A that does satisfy the condition A = 0, and since the two potentials should be equivalent in the
9.5. MAXWELLS EQUATIONS IN VACUUM
151
sense that they represent the same magnetic eld B, they should be related by a gauge transformation, A = A + . The Coulomb gauge condition then implies that the function should satisfy the equation 2 = A (9.36)
We recognize this as having the same form as Gauss law in the static case, where the electric eld is determined by the electrostatic potential, E = with no contribution from the A eld. Expressed in terms of the potential the electrostatic equation is 2 = / 0 , and this equation we know, for an arbitrary charge distribution, to have a well dened solution as the superposition of the Coulomb potentials from all the parts of the distribution. The solution of (9.36) should then have the same form as the solution of the Coulomb problem for a charge distribution, with / 0 replaced by A. (We shall later discuss the electrostatic case explicitly.) This shows that, for any electromagnetic eld conguration, one can make a gauge transformation of the vector potential to a form that satises the Coulomb gauge condition. With the Coulomb gauge condition satised, A = 0, Maxwells equations take the form 1. 2. 2 =
0
1 2 2 A 2 2 A = 0 j c t
(9.37)
where in the second equation we have introduced the transverse current density, dened by j = j
0
(9.38)
It is called transverse since it is divergence free, j = 0. This follows by applying equation 1. and the continuity equation for charge, jT 2 t = j+ t = 0 = j
0
(9.39)
Eq.(9.38) can therefore be re-interpreted as a standard (Helmholtz) decomposition of the vector eld, in a divergence-free (transverse) and a curl-free (longitudinal) component, j = j + j (9.40)
and Eq. 2. then shows that only the divergence-free component contributes to the equation. One should also note that Eq. 1. is non-dynamical in the sense that it involves no time derivative. It can thus be solved like the electrostatic equation, to give the potential expressed in terms of the charge distribution . That is the case even if is time dependent. This means that dynamical evolution of the electromagnetic eld, in the Coulomb gauge, is described by the vector potential A alone, while the scalar potential is uniquely dened by the charge distribution at any given time.
152
9.6
Maxwells equations in covariant form
The covariant form of Maxwells equations is based on the introduction of the electromagnetic eld tensor. It is an antisymmetric, relativistic tensor F , constructed by the E and B elds in the following way F 0k = F kl = 1 Ek , k = 1, 2, 3 c k, l = 1, 2, 3 klm Bm ,
(9.41)
with summation over the repeated index m, and with klm representing the three dimensional LevyCivita tensor. This tensor is antisymmetric in any pair of indices and is consequently different from 0 only when all indices klm are different. This set of indices then dene a permutation of the set 123, and with the denition 123 = 1 the value of klm for a permutations of 123 is determined by the antisymmetry property of the tensor. Written as a 4 4 matrix the eld tensor takes the form 0 E1 /c E2 /c E3 /c E1 /c 0 B3 B2 (F ) = (9.42) E2 /c B3 0 B1 E3 /c B2 B1 0 The reason for the electric and magnetic elds to be arranged into a common object F is that the two elds are mixed under Lorentz transformations. Such a mixing is implicit both in Maxwells equations and in the equation of motion for a charged particle (9.29). In the latter case this is obvious since a reference frame can be chosen where the particle is instantaneously at rest. In such a reference frame there is no contribution to the force from the magnetic eld, and therefore the effect of the electric eld in this frame must be equivalent to both the electric and magnetic elds in another frame where the particle is moving. It is an interesting fact that the mixing of the E and B elds is correctly expressed by combining them in a linear way into the antisymmetric, rank two tensor F . We continue to show that the set of Maxwells equations, in the form (9.27), gets a simple compact form when expressed in terms of the electromagnetic eld tensor. The two rst equations can be rewritten as 1. 2. 0k 1 F = xk c 0 kl F + 0 F k 0 = 0 j k x xl
(9.43)
and these two equations can now be merged into a single covariant equation F = 0 j (9.44)
where the abbreviation = x , introduced in Part II of the lecture notes, has been used. In the equation we have also introduced the 4-vector current density j , which is composed by the charge and current densities in the following way
(j ) = (c, j)
(9.45)
so that the original 3-vector j is extended to a 4-vector j by taking c as the time component of the current.
9.6. MAXWELLS EQUATIONS IN COVARIANT FORM
153
The electromagnetic equation (9.44) is expressed in covariant form, since the objects in the equation are all labelled by 4-vector indices, and all indices are placed in a consistent way. Consistent means that any free index (which is not summed over) like in (9.44), is placed either upstairs or downstairs in all places where it is used, and any contracted index (repeated index that is summed over) appears repeated in pairs with one upper and one lower index. These rules for covariance implies that the terms on the two sides of the equation transforms in the same way under Lorentz transformations (hence the name covariance), and therefore shows explicitly the relativistic invariance of the equations. Even if the focus on correct use of the positions of the vector indices can initially seem somewhat cumbersome there is therefore an obvious gain. When working with covariant equations the relativistic form is at all steps preserved and the correct position of the indices can be used as a form of book keeping to avoid errors when working with the equations. The continuity equation for charge can also be written in covariant form when the four-vector current is introduced. The covariant form is j = 0 (9.46)
as we can readily verify by separating the time derivative from the space derivative and using the fact that the time component of the 4-current is the charge density (up to a factor c). We have already noticed that charge conservation is needed if Maxwells equations should be consistent. This is seen very clearly from the covariant equation (9.44), where the continuity equation of the current follows from the antisymmetry of the electromagnetic tensor. We continue with bringing the source free equations 3. and 4. into covariant form. We rst note that the two equations can be expressed in terms of the electromagnetic eld tensor as 3. 4.
klm
lm F =0 xk 0m 1 F + klm 2 xl
(9.47)
klm
lm F =0 x0
(9.48)
lm . We introduce now the four where we in these equations have used Ek = cF 0k and Bk = 1 2 klm F dimensional Levi-Civita tensor which is fully antisymmetric in interchange of the indices, and which further satises 0klm = klm . As one can readily check the two equations now can be merged into a single, compact equation
F = 0
(9.49)
To simplify the covariant form of Maxwells equations even further we introduce the dual eld tensor = 1 F F (9.50) 2 If we check what this denition means for the components of the dual tensor, we nd that it is derived from the eld tensor by a duality transformation 1 1 E B , B E c c Written as a 4 4 matrix the dual eld tensor is therefore 0 B1 B2 B3 0 E3 /c E2 /c ) = B1 (F B2 E3 /c 0 E1 /c B3 E2 /c E1 /c 0 : F F (9.51)
(9.52)
154
The original four Maxwell equation can now be written as two compact, covariant equations F F = 0 j = 0 (9.53)
In this form the symmetry of the equations under duality transformation, which interchanges F and , is seen very clearly and also the difference, that the magnetic current of the second equation F is missing.
9.7
The electromagnetic 4-potential
The lack of a magnetic current in Maxwells equations makes the symmetry between the electric and magnetic elds not fully complete. However, for the same reason the eld tensor can be expressed in terms of the electromagnetic 4-potential A in the following way F = A A (9.54)
As discussed at an earlier stage, the 4-potential is composed of the non-relativistic potentials in such a way that the time component is A0 = /c with as the original scalar potential and the space part of A is identical to the vector potential A. When the eld tensor is expressed in terms of the 4-potential the second of the two covariant Maxwell equations is satised as an identity, as one can verify by in terms of A . This means that Maxwells equations are reduced to one 4-vector expressing F equation, which is A A = 0 j (9.55)
As a last step to simplify the equation we again make use of the freedom to change the potential by a gauge transformation. In covariant form such a transformation is A A = A + (9.56)
where is an unspecied differentiable function of the space time coordinates. In the covariant formulation it is straight forward to check that such a transformation of the 4-potential will not change the eld tensor. The freedom to change the potential in this way can be used to bring it to the form where the covariant Lorentz gauge condition is satised, A = 0 When this condition is satised we nd Maxwells equation reduced to its simplest form A = 0 j One sometimes use the symbol
2
(9.57)
(9.58)
for the differential operator,

2
= 2
1 c2 t
(9.59)
It is called the dAlembertian operator and is an extension of the three dimensional Laplacian 2 to four dimensions.
9.8. LORENTZ TRANSFORMATIONS OF THE ELECTROMAGNETIC FIELD
155
9.8
Lorentz transformations of the electromagnetic eld
When the electric and magnetic elds are collected in the electromagnetic eld tensor, this means that the correct transformation of E and B under Lorentz transformations have been implicitly assumed. This is of course not simply postulated, it is based on the assumption that Maxwells equations (as well as the equation of motion of charged particles) have the same form in all inertial reference frames, and this is in turn a well established fact based on experimental tests. We will here take the tensor properties of F as the starting point, and show from this how Lorentz transformations mix the electric and magnetic components. We consider rst eld transformations under a simple boost in the x direction. It is given by the Lorentz transformation matrix (L ) = 0 0 0 0 0 0 1 0 0 0 0 1
(9.60)
which means that the only non-vanishing matrix elements are L0 0 = L1 1 = L1 1 = L2 2 = 1 L0 1 = L1 0 = (9.61)
The tensor properties of the electromagnetic eld tensor implies that the transformed eld is related to the original eld by the equation F
= L L F
(9.62)
We extract from this formula the transformation equations for the components of the electric and magnetic elds, in the case where the matrix elements of the Lorentz transformation are given by (9.61). The x component of the electric eld is, E1 = cF
01
= cL0 0 L1 1 F 01 + cL0 1 L1 0 F 10 = 2 (1 2 )E1 = E1 = c(L0 0 L1 1 L0 1 L1 0 )F 01 (9.63)
which shows that the component in the direction of the boost is unchanged. In the orthogonal directions the components of the transformed eld are E2 = cF
02
= cL0 0 L2 2 F 02 + cL0 1 L2 2 F 12 = (E2 vB3 ) = E2 cB3 (9.64)
156 and E3 = cF
03
= cL0 0 L3 3 F 03 + cL0 1 L3 3 F 13 = E3 + cB2 = (E3 + vB2 ) (9.65) These expressions can be written in a form which is independent of the choice of coordinate axes by introducing the parallel and transverse components of the electric eld E = E1 , The transformation formulas then are E =E , E = (E + v B) (9.67) E = E2 j + E3 k (9.66)
and shows that the component of the eld in the direction of the boost velocity v is unchanged, while the transverse components (orthogonal to v) are a mixtures of the of the original transverse components of the electric and magnetic elds. The expressions for the transformed magnetic eld are similar, B1 = F
23
= L2 2 L3 3 F 23 = F 23 = B1 B2 = F
13
(9.68)
= L1 1 L3 3 F 13 L1 0 L3 3 F 03 1 = B2 + E3 c v = (B2 + 2 E3 ) c
12
(9.69)
B3 = F
= L1 1 L2 2 F 12 + L1 0 L2 2 F 02 1 = B3 E2 c v = (B3 2 E2 ) (9.70) c We write these in a coordinate independent way as v B = B , B = (B 2 E) (9.71) c The transformation formulas for E and B have almost the same form and they are related by the duality transformation already discussed, 1 1 E B, B E c c The symmetry under this transformation, which transforms between the two eld tensors F and , could in fact have been used to derive the transformation formula for the B eld directly from F the transformation formula for the E eld.
9.8. LORENTZ TRANSFORMATIONS OF THE ELECTROMAGNETIC FIELD
157
9.8.1
Example
As a simple example we assume that in the reference frame S there is no electric eld, and a constant magnetic eld B = B0 k directed along the z axis. The moving frame S has a velocity v in the x direction. We split the elds in parallel and transverse components, E = E = 0 , B = 0 , B = B0 k (9.72)
For the parallel components of the transformed elds we nd E = E = 0, and for the transverse components E = (E + v B) = vB0 i k = vB0 j v B = (B 2 E) = B = B0 k c Collecting these terms we nd that the elds in the reference frame S are E = vB0 j , B = B0 k (9.75) B =B =0 (9.73)
(9.74)
Also in this reference frame the magnetic eld points in the z direction, but it is stronger than in S due to the factor which is larger than 1. In addition there is an electric eld in the direction orthogonal to both the velocity of the transformation and to the magnetic eld.
9.8.2
Lorentz invariants
From the electromagnetic tensor F we can construct several Lorentz invariant quantities. These are certain combinations of the electric and magnetic eld strengths that take the same value in all inertial frames. For a general tensor T the trace T is such an invariant, but in the present case the trace vanishes since F is antisymmetric. This means that there is no invariant that is linear in the components of E and B. However, there are two quadratic expressions that are Lorentz invariants. These are I1 = I2 = 1 1 F F = B2 2 E2 2 c 1 1 F F = E B 4 c
(9.76)
It is easy to check that for the example just discussed we get the same expression for the two invariants, whether we evaluate them in reference frame S or S ,
2 I1 = B0 ,
I2 = 0
(9.77)
we note in particular that even if the E and B elds get mixed by the Lorentz transformation, the fact that E dominates B (E2 > c2 B2 ) or B dominates E (E2 < c2 B2 ) can be stated without reference to any particular inertial frame.
158
9.9
Example: The eld from a linear electric current
In the following we consider the situation where a constant current is running in a straight conducting wire, as shown in the Fig. 9.4. In the reference frame S that is stationary with respect to the conductor the current takes the value I and the conductor is electrically neutral. The magnetic eld will circulate the current and outside the conductor the eld strength B is determined by Amp` eres law B ds = 0 I (9.78)
Assuming the conductor to be rotationally symmetric this determines the eld to be B= 0 I e 2r (9.79)
with r as the distance from the centre of the conductor and e as a unit vector circulating the current. Due to charge neutrality the electric eld orthogonal to the current vanishes, but there is an electric eld inside the conductor that drives the current. It is given by je = E0 (9.80)
with as the conductivity. We assume E to have a constant value inside the conductor, with the same value also outside, close to the conductor. The eld strength B refers to the reference frame S where the ions of the conducting material are at rest. In this frame the electrons are moving with an average velocity ve . We have I = Aje = Ave e (9.81)
where e and je are the average charge and current densities of the electrons and A is the cross section area of the conductor. (For simplicity we assume the current density to be constant over the cross section.) We will now introduce a second inertial reference frame S that moves with the average velocity of the electrons. In this frame the ions move with the velocity ve , while the electrons are (on average) at rest. The elds in the reference frame S are given by the eld transformation formulas. To use these we need rst to split the elds in a parallel component (along the conductor) and a normal component (orthogonal to the conductor). For the elds in S these components are E = E0 , E = 0 , B = 0, B = 0 I e 2r (9.82)
The transformation formulas, with velocity ve for reference frame S along the conducting wire, give E = E = E0 , B = B = 0, E = (E ve B) = ve 0 I er 2r ve 0 I B = (B + 2 E) = e c 2r
(9.83)
The magnetic eld is also in this reference frame circulating the current, but now it is stronger, enhanced by the factor . The electric eld we note to have, in addition to the parallel component E0 , a normal component that is radially directed, out from the conductor. This normal component may seem
9.9. EXAMPLE: THE FIELD FROM A LINEAR ELECTRIC CURRENT
159
B I E
a)
ve I L b)
Figure 9.4: Electromagnetic elds of a linear current. In gure a) the directions of the electric and magnetic
elds are indicated as seen in the rest frame S of the conductor. In gure b) are indicated two volumes of equal length in S , where one of them is stationary with respect to the ions (blue dots) and the other moves with the electrons (red dots). In S the charges neutralize each other, while in the rest frame S of the electrons that is not so due to the length contraction effect. The non-vanishing charge density of the conductor in S explains the presence of a radially directed E eld in this frame, which follows from the Lorentz transformation of the elds from S to S .
somewhat unexpected, since it indicates that the conducting wire in reference frame S is not charge neutral. A charge density is needed along the wire in order to create a radially directed electric eld. We will check that these results are consistent with Maxwells equations by evaluating the charge and current densities in the transformed reference frame. In reference frame S the charge and current densities of the electrons are e and je = ve e , while the charge and current densities of the ions are i = e (due to charge neutrality) and ji = 0. To nd the corresponding quantities in reference frame S we use the fact that charge and current densities together form a 4-vector j = (c, j). We use the standard transformation formula for 4-vectors to give the charge and current densities in S , e = (e je i ji
2 ve ve 1 j ) = (1 )e = e e 2 2 c c = (je ve e ) = 0 ve = (i 2 ji ) = e c = (ji vi i ) = ve e
(9.84)
This gives for the total charge and current densities in S , j = e + i = 1 1 e e = e e = 2 e = je + ji = ve e
(9.85)
The charge density in S is indeed different from zero and the current density is modied by a factor . We check now that the expressions given above for the transformed the charge and current
160
densities are consistent with what we have found for the transformed elds. We note that the enhancement of current in S , by the factor , is consistent with the corresponding enhancement of the transformed magnetic eld. To the check the consistency of the transformation of the charge density and the electric eld we consider Gauss law in S . We denote by Q the charge in a piece of the conducting wire of length L, so that Q = LA. According to Gauss law this charge should create a radially directed electric eld given by 2r L Er = which gives Er = A 0 I e A je A = 2 = = ve 2r 0 2r 0 2rc 0 2r (9.87) Q
0
(9.86)
where in the last step we have used the relation 0 0 = 1/c2 . The expression for the radial component of the electric eld found in this way is indeed consistent with the result found by applying the transformation formula for the electromagnetic eld. Although it may initially seem strange that the conducting wire is charge neutral in one reference frame, but not in the other, it is a clear consequence of the description of charge and current as components of the same 4-vector current. Lorentz transformations will mix the time and space components of of the 4-current. As a nal point, we shall examine how the results we have obtained can also be understood as a consequence of length contraction. Let us then consider an imaginary container that includes a part of the conducting wire and moves with the speed ve of the electrons along the wire. The length measured in S is L and the number of electrons within the container is N . The total electron charge within the container is therefore Qe = N e = ALe (9.88)
with e as the electron charge. Let us next consider another container of the same length L in S , but which is at rest. At a given instant the two imagined containers will overlap, and due to charge neutrality the number of electrons within the rst container is the same as the number of ions within the second container (if we assume each ion contributes with one conduction electron). For the total ion charge within the second container we therefore have Qi = Qe = N e = ALe (9.89)
We now regard the situation as it appears in reference frame S . The lengths of the containers appear with a different lengths from those in S , due to the length contraction effect, but the number of particles in each container is unchanged. If we rst consider the length of the electron container, we nd that it is Le = L. The container is longer in S than in S , since S is the rest frame of the electrons. As far as the other container is concerned the situation is opposite, since S is the rest frame of the ions. Therefore the length of this in S is Li = L/ . Thus the two containers still contain the same amount of charge as in S , but since the length of the containers are different, the charge densities have changed. We have e = Qe 1 Qe 1 = = e Le A LA (9.90)
9.9. EXAMPLE: THE FIELD FROM A LINEAR ELECTRIC CURRENT and i = Qi Qe = = e Le A LA
161
(9.91)
The total charge density in S is therefore = e + i = 1 e e = 2 e (9.92)
This gives a result for the charge density in reference frame S that agrees is with what we have already found by use of the Lorentz transformation formulas.
162
Chapter 10
Dynamics of the electromagnetic eld

Maxwells equations show that the electromagnetic eld has it own dynamics. It can propagate as waves through empty space and it can carry energy and momentum. This was realized by Maxwell, and since the propagation velocity is identical to the speed of light, that convinced him that light is such an electromagnetic wave phenomenon. In this chapter we rst discuss the wave solutions of Maxwells equations with particular focus on polarization of electromagnetic waves and next examine how Maxwells equation determine the energy and momentum densities of the electromagnetic eld.
10.1
Electromagnetic waves
Waves are solutions of Maxwells equations in the source free case, with j = 0. We study this situation in the Lorentz gauge, with A = 0. The eld equation then is A = 0 where the differential operator = 2 1 2 c2 t2 (10.2) (10.1)
has the form of a wave operator in three dimensional space. The wave equation (10.1), which is a linear differential equation, has a complete set of normal modes as solutions. In the open space, without any physical boundaries, a natural choice for such a set of normal modes are the monochromatic plane waves, of the form A (x) = A (0) eik x
(10.3)
with A (0) as the amplitude of eld component . The four vector k with components k , decomposes in a time component that is proportional to the frequency of the wave and a space component that is the wave number, k = ( , k) c (10.4)
The plane wave solution (10.3) is a complex solution of Maxwells equation. Such a complex form is often convenient to use since it makes the expressions more compact. However, but should keep in mind that the physical eld is real, and should be identied with the real (or imaginary) part of the solution. 163
164
CHAPTER 10. DYNAMICS OF THE ELECTROMAGNETIC FIELD
To check that plane waves of the form (10.3) are solutions of the wave equation (10.1) is straight forward. We nd A = k k A which shows that the function (10.3) is a solution provided k k = 0 (10.6) (10.5)
This means that the 4-vector k is a light like vector (sometimes also called a null vector), and this gives rise to the well-known linear relation between frequency and wave number for electromagnetic waves, = c|k| (10.7)
The Lorentz gauge condition further demands the two 4-vectors k and A to be orthogonal in the relativistic sense k A = 0 (10.8)
One should note that the Lorentz gauge condition does not x uniquely the 4-potential for a given electromagnetic eld. This is readily seen by assuming A to be a general potential which satises no particular gauge condition. By a gauge transformation A A = A + (10.9)
it can be brought to a form which does satisfy the Lorentz gauge condition A = 0, provided satises the equation = A (10.10)
However, is not uniquely determined by the equation, since to any particular solution of this differential equation one can add a general solution of the homogeneous (wave) equation = 0. In the present case one can use this freedom to set the time component of the potential to zero, A0 = 0, and the remaining vector part will then satisfy the Coulomb gauge condition A = 0. When written in non-covariant form the wave equation for A is (2 1 2 ) A(r, t) = 0 c2 t2 (10.11)
This shows that the three components of the vector potential satisfy three identical, uncoupled wave equations, but the three components are coupled by the Coulomb gauge condition. In the A0 = 0 gauge the non-covariant form of the plane wave solution is A(r, t) = A0 ei(krt) where the amplitude A0 is a complex vector that should be a orthogonal to k, k A0 = 0 in order to satisfy the Coulomb gauge condition. (10.13) (10.12)
10.2. POLARIZATION
165
The general solution to the electromagnetic wave equation (10.11) can now be written as a superposition of plane waves A(r, t) = d3 k A(k) ei(krt) , k A(k) = 0 (10.14)
where each Fourier component A(k) has to satisfy the transversality condition. We have in this discussion of electromagnetic waves assumed that they propagate in the open innite space. The plane waves then dene a complete set of normal modes of the eld. If the situation instead correspond to wave propagation inside some given boundaries, for example inside a wave guide the normal modes are not the innite plane waves but solutions that are adjusted to the given boundary conditions. To nd the normal modes of the electromagnetic eld is then more demanding, but the general solution is again a (general) superposition of these modes.
10.2
Polarization
The plane wave solution (10.12) for the electromagnetic potential gives related expressions for the electric and magnetic elds. For the electric eld we nd E(r, t) = and for the magnetic eld B(r, t) = A = ik A0 ei(krt) = ik A(r, t) We note that both these elds satisfy the transversality condition kE=kB=0 and are related by 1 B = n E, c E = cn B (10.18) (10.17) (10.16) A = i A0 ei(krt) = i A(r, t) t (10.15)
with n = k/k as the unit vector in the direction of propagation of the plane wave. Thus the triplet (k, E, B) form a right handed, orthogonal set of vectors. We further note that for a monochromatic plane wave the two electromagnetic Lorentz invariants previously discussed both vanish E2 c2 B2 = 0 , EB=0 (10.19)
The monochromatic wave is, as we see, specied on one hand by the wave number k, which gives the direction of propagation and the frequency of the wave, and on the other hand by the orientation of the electric eld vector E in the plane orthogonal to k. The degree of freedom specied by the direction of E we identify as the freedom of polarization of the electromagnetic wave. We shall take a closer look at the description of different types of polarization. As follows from Eq.(10.18) it is sufcient to focus on the electric eld E, since the magnetic eld B is uniquely determined by E. Written in complex form the electric eld strength of the plane wave has the form E(r, t) = E0 ei(krt) (10.20)
166
B E
Figure 10.1: Field vectors of a plane wave. The wave number k together with the two eld vectors E and B
form a righthanded set of orthogonal vectors.
where the amplitude E0 is in general a complex vector. We consider the real part of the eld (10.20) as the physical eld. When decomposed on two arbitrarily chosen orthogonal real unit vectors e1 and e2 in the plane orthogonal to k, the real eld gets the general form E(r, t) = E10 e1 cos(k r t + 1 ) + E20 e2 cos(k r t + 2 ) (10.21)
where the two amplitudes E10 and E20 as well as the two phases 1 and 2 may be different. When the two components are in phase, 1 = 2 the wave is linearly polarized, but in the general case the wave is elliptically polarized with circular polarization as a special case. We will discuss these different types of polarization in some detail, but let us rst re-introduce the complex notation in the following way. We write the complex amplitude of the electric eld as E0 = E0 where E0 is now a real (positive) amplitude and (10.21) if we make the following identications E0 =
1 1 1
(10.22)
is a complex unit vector. The real part has the form
2 + E2 E10 20
= cos ei1 e1 + sin ei2 e2
(10.23)
with cos = E10 , E0 sin = E20 E0 (10.24)
In complex notation the general monochromatic plane wave then has the form E(r, t) = E0 ei(krt)
1
(10.25)
This expression is equivalent to (10.21) in the sense that the latter is the real part of the former, but the complex eld (10.25) is usually more convenient to work with due to its more compact form. The corresponding magnetic eld strength is B(r, t) = B0 ei(krt) with B0 = E0 /c and
2 2
(10.26)
=n
= sin ei2 e1 + cos ei1 e2
(10.27)
10.2. POLARIZATION
167
The two unit vectors malization relations
and
i
are referred to as polarization vectors. They satisfy the orthonor-
= ij ,
n=0
(n = k/k )
(10.28)
and the set of vectors (Re 1 , Re 2 , n) (or equivalently the set (E, B, k)) form a right handed set of orthogonal vectors. The different types of polarization can be analyzed by considering the orbit described by the real vector E(r, t) in the two-dimensional plane when the time coordinate t changes for a xed point r in physical space. We consider rst some special cases. Linear polarization
e2
E
e1
Figure 10.2: Linear polarization. The electric eld vector E oscillates in a xed direction orthogonal to the
wave number k, while the magnetic eld vector B oscillates in the direction orthogonal to E. The vector k is in this gure directed out of the plane, towards the reader.
This corresponds to the case where the two orthogonal components of the real electric eld oscillates in phase, which means 2 = 1 (or 2 = 1 + ). The E eld then oscillates along a xed axis orthogonal to k and the B eld oscillates in the direction orthogonal to both k and E. The axis of oscillation of E together with the axis dened by k span a two dimensional plane, which is the polarization plane of the electromagnetic eld. The realvalued electric eld then has the form E(r, t) = E0 cos(k r t + )[cos e1 + sin e2 ] (10.29)
The eld oscillates along a xed line which is rotated by an angle relative to the chosen unit vector e1 in the plane orthogonal to k. Circular polarization In this case the two orthogonal components of the E eld are 900 out of phase, so that 2 = 1 + / 2 or 2 = 1 /2, while the amplitudes of these components are equal, so that cos = sin = 1/ 2. Up to an over all phase factor the complex polarization vector then has the form
1
1 = (e1 ie2 ) 2
(10.30)
168
For the electric eld this gives E0 E(r, t) = [cos(k r t + )e1 sin(k r t + )e2 ] 2 (10.31)
where the sign determines whether the eld vector rotates in the positive or negative direction when t is increasing.
e2
e2
e1
e1
Figure 10.3: Circular polarization. The electric eld vector now rotates in the plane orthogonal to k, and the
magnetic eld B also rotates, but 90o out of phase with E. The direction of k is also here out of the plane, towards the reader. Two cases are shown, corresponding to right handed and lefthanded circular polarization.
Elliptic polarization Next we consider the case where the two orthogonal components are still 90o out of phase, but where the absolute value of the two components now are different. The complex polarization vector 1 now has the form
1
= cos e1 + i sin e2
(10.32)
and the (real) electric eld is E(r, t) = E0 [cos cos(k r t + )e1 + sin sin(k r t + )e2 ] E1 (r, t) e1 + E2 (r, t) e2 The expression shows that the two components for the eld satisfy the ellipse equation
2 2 E1 E2 + =1 a2 b2
(10.33)
(10.34)
when we dene a = E0 cos = E10 and b = E0 sin = E20 . This means that when we consider the eld for a xed point r in space, the time dependent vector E will trace out an ellipse in the plane orthogonal to the direction of propagation of the wave, n. The symmetry axes of the ellipse are in this case along the directions of the real unit vectors e1 and e2 and the half axes of the ellipse are given by a and b. This is a case of elliptic polarization.
10.2. POLARIZATION
169
The general case The case of elliptic polarization discussed above seems not to be the most general one, since we have xed the relative phase of the two orthogonal components of the polarization vector to be /2. However, the most general case in fact corresponds to elliptic polarization, with the only modication that the ellipse is rotated relative to the axes dened by the two chosen real unit vectors e1 and e2 . To demonstrate this we start with the general expression
1
= cos ei1 e1 + sin ei2 e2
(10.35)
which for the real electric eld corresponds to the general expression (10.21). We now write the complex phases in the form 1 = + 1 , 2 = + 2 (10.36)
where we note that is a free variable, where a change in this variable can be compensated for by a change in 1 and 2 . We then make use of the formula for the cosine of a sum, cos(k r t + n ) = cos(k r t + ) cos 1 sin(k r t + ) sin 2 to re-write the electric eld as E(r, t) = (E10 cos 1 e1 + E20 cos 2 e2 ) cos(k r t + ) Next we dene two new unit vectors E10 e1 = E10 cos 1 e1 + E20 cos 2 e2 E20 e1 = E10 sin 1 e1 + E20 sin 2 e2 (10.39) (10.37)
(E10 sin 1 e1 + E20 sin 2 e2 ) sin(k r t + )
(10.38)
where E10 and E20 are xed by the normalization conditions of the vectors. The two new vectors e1 and e2 should also be orthogonal, and that gives the following condition,
2 2 E10 cos 1 sin 1 + E20 cos 2 sin 2 = 0
(10.40)
This equation we can regard as an equation to determine when 1 and 2 are xed, thereby exploiting the freedom in the choice of this variable. The electric eld, when expressed in terms of the new real unit vectors has the form E(r, t) = E10 e1 cos(k r t + ) + E20 e2 sin(k r t + ) (10.41)
and by comparing with (10.33) we see that it has the same form already discussed, where the two orthogonal components of the eld is 90o out of phase. Thus the polarization is elliptic, but the symmetry axes are rotated relative to the original unit vectors e1 and e2 . The rotation angle is determined by equations (10.39) and (10.40). Fig. 10.4 shows a case of elliptic polarization where the electric eld vector and the magnetic eld vector trace out two orthogonal ellipses under the time evolution. A physical example of medium which can change the eccentricity of the polarization ellipse of light is a birefringent crystal. When passing through such a crystal a beam of light will be split into
170
e2
e1
Figure 10.4: Elliptic polarization. The time dependent electric eld vector E now describes an ellipse in the
plane orthogonal to k, while the magnetic eld vector B describes a rotated ellipse. Also here the direction of k is out of the plane, towards the reader.
two components which pass at different speed through the crystal. These two components, which have orthogonal polarization with respect to the optical axis of the crystal, are called the ordinary and extraordinary ray. Assume a beam passes trough the crystal in a direction orthogonal to the optical axis, initially with linear polarization with equal amplitude for the ordinary and extraordinary component (which means polarization at 450 degree relative to these two directions). Due to the difference in speed through the crystal there will be a relative phase difference introduced between the two components so that the light that emerges from the crystal will in general be elliptically polarized. The eccentricity will depend on the speed of the two components inside the crystal and on the crystal width. An optical device with the property of changing the polarization in this way is called a wave plate. A half-wave plate will change the relative phase of the two components by /2 (900 ) so that linearly polarized light that enters the crystal with polarization at angle 450 relative to the optical axis will leave the crystal also with linear polarization, but now with the direction of polarization orthogonal to that of the incoming light. Finally, one should note that all the effects of polarization that we have discussed in this section can be viewed as being consequences of superposition. In all cases the monochromatic plane wave can be viewed as a superposition of two linearly polarized plane waves, with polarization along two arbitrarily chosen orthogonal directions. The different types of polarization are then produced by varying the relative amplitudes and the relative phases of these two partial waves.
10.3
Electromagnetic energy and momentum
Maxwells equations describe how moving charges give rise to electromagnetic elds and the Lorentz force describe how the elds act back on the charges. Since the eld acts with forces on charged particles, this implies that energy and momentum is transferred between the eld and the particles, and consequently the electromagnetic eld has to be a carrier of energy and momentum. The precise form of the energy and momentum density of the eld is determined by Maxwells equation and the Lorentz force, under the assumption of conservation of energy and conservation of momentum. To demonstrate this we consider a single pointlike particle that is affected by the eld. The charge
10.3. ELECTROMAGNETIC ENERGY AND MOMENTUM and momentum density in this case can be expressed as (r, t) = q (r r(t))
171
j(r, t) = q v(t) (r r(t))
(10.42)
with q as the charge of the particle, and with r(t) as the time dependent position vector of the particle and v(t) as the velocity. The Lorentz force which acts on the particle is F = q (E + v B) (10.43)
This force will in general change the energy of the particle, and the time derivative of the energy is d Epart = F v = q v E dt (10.44)
Energy conservation means that the total energy of both eld and particle is left unchanged. With Ef ield as the eld energy within a nite (but large) volume V with boundary surface , energy conservation takes the form d (Ef ield + Epart ) = dt S dA (10.45)
with S as the energy current density of the eld and dA as the area element on the surface. The right hand side of the equation is the energy loss in V due to the energy current through the boundary surface, for example due to radiation. The time derivative of the eld energy can now be written d Ef ield = q v E dt S dA = j E dV S dA (10.46)
where the expression for the time derivative of the particle energy has been re-written by use of expression (10.42) for the current density. In the last form the equation is in fact valid for arbitrary charge congurations within the volume V . By use of Amperes law the current density can be replaced by the electric and and magnetic elds in the following way j= 1 B 0
0
E t
(10.47)
and this gives for the volume integral in (10.46) j E dV = 1 E ( B) 0

0E
E dV t
(10.48)
We further modify the integrand of the rst term by using eld identities and Faradays law of induction, E ( B) = (B E) + B ( E) = (E B) B B t (10.49)
172 This gives j E dV
= =
0E V
E B 1 1 + B (E B) dV t 0 t 0
0E 2
d dt
1 2
1 1 2 B dV 0 0
(E B) dA
(10.50)
where in the last step a part of the volume integral has been rewritten as a surface integral by use of Gauss theorem. Writing the eld energy as a volume integral, Ef ield = V u dV , with u as the energy density of the eld, and separating the volume and surface integrals, we get the following form for Eq. (10.46) d dt 1 1 (u ( 0 E2 + B2 ))dV = 2 0 V (S 1 (E B)) dA 0 (10.51)
Since this equation should be satised for an arbitrarily chosen volume and for general eld congurations, we conclude that the integrands of the volume and surface integrals should vanish separately. This determines the energy density as a function of the eld strength, u= and the current density S= 1 EB 0 (10.53) 1 2
0E 2
1 2 B 0
(10.52)
These are the standard expressions for the energy density and the energy current density of the electromagnetic eld, and the derivation shows that the eld equations combined with energy conservation leads to these expressions. The vector S is also called Poyntings vector. The expression for the momentum density of the electromagnetic eld can be derived in the same way. We start with the expression for the time derivative of the particle momentum, d Ppart = F = q (E + v B) dt and for energy conservation d (Pf ield + Ppart ) = 0 dt (10.55) (10.54)
In this case we assume for simplicity the momentum density to be integrated over the innite space in order to avoid the surface contribution. We follow the same approach as for the eld energy, by applying Maxwells equations to replace the charge and current densities with eld variables. By further manipulating the expression, and in
10.3. ELECTROMAGNETIC ENERGY AND MOMENTUM particular assuming surface integrals to vanish, we get d Pf ield = dt = = = = d dt E + j B dV 1 E 1 ( B 2 ) B dV 0 c t E 1 ) B dV 0 ( E) E + ( B 0 0 t B E 0 E+ 0 B dV t t
0 E (
173
E) +
0E
B dV
(10.56)
This gives the following expression for the eld momentum density is then g=
0E
(10.57)
and we note that, up to a factor 1/c2 = 0 0 , it is identical to the energy current density S. In the relativistic formulation the energy density and the momentum density are combined in the symmetric energy-momentum tensor, 1 T = (F F + g F F ) 4 (10.58)
The energy density corresponds here to the component T 00 and Poyntings vector to (c times) the components T 0i , i = 1, 2, 3.
10.3.1
Energy and momentum density of a monochromatic plane wave
We consider a plane wave with the electric and magnetic eld vectors related by 1 B = n E, c E = cn B (10.59)
where n = k/k is a unit vector in the direction of propagation of the wave. This gives B2 = E2 /c2 and therefore the energy density of the eld is 1 1 u = ( 0 E2 + B2 ) = 2 0
0E 2
(10.60)
with equal contributions from the electric and magnetic elds. Poyntings vector, which determines the energy current and momentum densities of the plane wave, is S= 1 EB= 0
0 cE 2
n = u cn
(10.61)
It is directed along the direction of propagation of the wave, and the last expression in (10.61) is consistent with the interpretation that the eld energy is transported in the direction of the propagating wave with the speed of light.
174
10.3.2
Field energy and potential energy
Let us consider two static charges q1 and q2 at relative position r = r1 r2 . The Coulomb energy of the system is U (r) = q1 q2 4 0 r (10.62)
and the usual picture is that the energy is considered as a potential energy of the two charges. However, in the preceding discussion we have found an expression for the local energy density of the electromagnetic eld, which should also apply to this static situation. This raises the question of how the potential energy of the charges is related to the electromagnetic eld energy. An important point to notice is that we should not consider the two energies as something we should add in order to obtain the total energy of the system of charges and elds. Instead the integrated eld energy is identical to the total electromagnetic energy of the charges and elds and the potential energy can be extracted as the part of this energy that depends on the position of the static charges. We demonstrate this by calculating the integrated eld energy of the two charges. The integrated eld energy is E= 1 2
0
E(r )2 d3 r
(10.63)
where the electrostatic eld E is the superposition of the Coulomb eld from the two charges, E(r ) = E1 (r ) + E2 (r ) = q1 r r1 q2 r r2 + 4 0 |r r1 | 4 0 |r r2 | (10.64)
The eld energy then has a natural separation into three parts E = E1 + E2 + E12 (10.65)
where the rst two parts are the contributions from the Coulomb energies of each of the two charges disregarding the presence of the other, E1 = E2 = 1 2 1 2
0
E1 (r )2 d3 r = E2 (r )2 d3 r =
q1 2 32 2 q2 2 32 2
d3 r q1 2 = 4 r 8 0 d3 r q1 2 = r4 8 0
0 0
dr r2 dr r2 (10.66)
We note that these two terms are independent of the positions of the particles. They are referred to as the self energies of the particles and these energies are in a sense always bound to the particles in the Coulomb eld surrounding each of them. Except for the different charge factors the self energy of the two charges are the same, but we note that for point particles the integrated self energy diverges in the limit r 0. This is a separate point to discuss and we shall return to this question briey. The third contribution to the eld energy comes from the superposition of the Coulomb elds of the two particles, E12 (r) =
0
E1 (r ) E2 (r )d3 r
(10.67)
As indicated in the equation it depends on the distance between the two charges. To calculate this term it is convenient to introduce the Coulomb potential of one of the particles E1 = 1 . We
10.3. ELECTROMAGNETIC ENERGY AND MOMENTUM
175
restrict the volume integral to a nite volume V and extract, by partial integration, a surface term as an integral over the boundary surface S of the volume V , E12 = = =
0 V 0 V 0 S
1 E2 d3 r (1 E2 ) d3 r + 1 E2 dS +
0 V 0 V
1 E2 d3 r (10.68)
1 (r ) q2 (r r2 ) d3 r
In the rst term Gauss theorem has been applied to re-write the volume integral of the divergence as a surface integral, and in the second term Gauss law for the electromagnetic eld has been applied to re-write the divergence of the electric eld as a charge density. Since we consider point charges this density is proportional to a Dirac delta function. Let us now assume the volume tends to innity. We note that the surface integral tends to zero, since far from the charges the product 1 E2 falls off with distance as 1/r 3 . We are then left with the volume integral, which is easy to evaluate due to the presence of the delta function, E12 (r) = q2 1 (r2 ) = q2 q1 q1 q2 = = U (r) 4 0 |r1 r2 | 4 0 r (10.69)
This shows that the Coulomb potential can be identied as the part of the total eld energy that depends on the distance between the charges and is due to the overlap of the electric elds of the two charges. This demonstrates that the potential energy of the charges in the electromagnetic eld is a part of the total electromagnetic eld energy, rather than something that should be added to the eld energy. We return now to the question of how to understand the expression for the self energy terms. For an isolated point charge q located at the origin the energy of the Coulomb eld is E= 1 2
0
E2 d3 r =
q2 8 0
dr r2
(10.70)
and this energy is obviously innite due to the divergence of the integral as r 0. A reasonable assumption is that there is nothing wrong with the expression for the eld energy, but that the idealization of treating the charge as being located at a mathematical point is the origin of the problem. Thus as soon as we assume that the charge has a nite size a, with this as an effective cutoff of the integral, the energy becomes nite, Ea = q2 8 0
a
q2 dr = r2 8 0 a
(10.71)
This, at least formally, solves the problem with the innite energy. However, to make a consistent picture of physical particles like electrons as small charged bodies is not so simple. That is a problem that exists not only in the classical theory; also in the quantum description of particles and elds there are innities associated with the electromagnetic self energies that have to be taken care of by the theory. A standard way to treat the self energy problem is based on the fact that the self energy is bound to each individual charge and therefore is not important for the interactions between the particles. One may therefore avoid the problem of a precise theory of point like particles by simply assuming the
176
energy carried by the eld to be nite and assuming that the only physical effect of this energy is to change the mass of the charged particle. This change is given by Einsteins relation mc2 = Ea = q2 8 0 a (10.72)
The physical mass of the particle can then be written as a sum m = mb + m (10.73)
where mb is the so called bare mass, which is the (imagined) mass of the particle without the Coulomb eld. When the physical mass enters the equations of motion that means that the mass renormalization effect of the self energy has been included and all other effects of the self energy can be neglected. Finally, let us use this interpretation of the self energy to give an estimate the value of the length parameter a for an electron. We know that m me , with me as the physical electron mass and with equality meaning that all the electron mass is due to the electromagnetic energy of its Coulomb eld. In this limit we get e2 = me c2 8 0 a a= e2 8 0 me c2 (10.74)
With a more explicit model of the electron as a charged spherical shell of radius re a similar calculation of the electromagnetic energy gives the same result as the one obtained by a simple cutoff in the integral, except for a factor 2, re = e2 4 0 me c2 (10.75)
This value is called the classical electron radius. It numerical value is re = 2.818 1015 m (10.76)
which shows that it is indeed a very small radius, comparable to the radius of an atomic nucleus.
Chapter 11
Maxwells equations with stationary sources

We return to the original form of Maxwells equations in the Lorentz gauge, and assume the 4-current j , and therefore both the charge and current density to be independent of time, = (r) , j = j(r) (11.2) A = 0 j (11.1)
Note that this is the case only in a preferred inertial frame (which we may refer to as the laboratory frame). When exploiting the time independence, the covariant form of the equation is therefore not important. With time independent sources we may also assume the electromagnetic potential A , as a solution of (11.1), to be time independent. This means that the Lorentz gauge condition again reduces to the Coulomb gauge condition, A = 0 and Maxwells equation has a natural decomposition in two independent equations 2 = (11.3)
0
where the scalar potential = A0 /c determines the electric eld and vector potential A determines the magnetic eld. Since there is no coupling between the equations for the E and B elds, the two cases can be studied separately. Equation (11.3) is then the basic equation in electrostatics, where static charges give rise to a time independent electric eld, while equation (11.4) is the basic equation in magnetostatics where stationary currents give rise to a time independent magnetic eld. As differential equations they are of the same type, known as the Poisson equation, and even if there are some differences, the methods of nding the electrostatic and magnetostatic elds, with given sources, are much the same. We examine now the two cases separately.
2 A = 0 j
(11.4)
11.1
The electrostatic equation
Since the electrostatic equation (11.3) is a linear differential equation, the solution can be seen as a linear superposition of contributions from pointlike parts of the charge distributions. For a single 177
178
CHAPTER 11. MAXWELLS EQUATIONS WITH STATIONARY SOURCES
point charge located at the origin, the charge density is (r) = q (r), with q as the electric charge. In this case the electric eld is most easily determined by use of the integral form of Gauss law, and by exploiting the rotational invariance of the eld, E(r) = E (r) r/r. For a spherical surface centered at the origin, Gauss law then takes the form, 4r2 E (r) = which determines the electric eld as E(r) = q 4
0
q
0
(11.5)
r2
r r
(11.6)
which is the standard form of the Coulomb eld. The corresponding Coulomb potential is also easily found to be (r) = q 4 0 r (11.7)
dq=(r) dV r-r r r
Figure 11.1: The electrostatic potential. The potential at a point r is determined as a linear superposition of
contributions from small pieces dq of the charge, located at points r in the charge distribution.
For a charge distribution (r) which is no longer pointlike, the potential can be written directly as a sum (or integral) over the Coulomb potential from all parts of the distribution, (r) = 1 4 0 (r ) 3 d r |r r | (11.8)
The corresponding electric eld strength is E(r) = = 1 4 0 (r ) (r r )d3 r |r r |3 (11.9)
In reality the above solution is a particular solution of the differential equation. A general solution will therefore be of the form (r) = c (r) + 0 (r) (11.10)
11.1. THE ELECTROSTATIC EQUATION
179
where c denotes the solution given above and 0 is a general solution of the source free Laplace equation 2 0 = 0 (11.11)
The solution (11.8) written above implicitly assumes certain boundary conditions that are natural in the open, innite space, namely that the potential falls to zero at innity. When we consider the electric eld in a nite region V of space, with given boundary conditions on the on the boundary surface S , the contribution from 0 will generally be important. This contribution to the potential will correct the contribution from the integrated Coulomb potential so that the total potential satises the boundary conditions. As a particular situation we may consider the electric eld produced by a charge within a cavity of an electric conductor. Since the boundary surface of the conductor is an equipotential surface, the function 0 is determined as a solution of the Laplace equation (11.11) with the following boundary condition on the surface of the conductor 0 (r) = c (r) = 1 4 0 (r ) 3 d r |r r | rS (11.12)
More generally we often make a distinction between two types of boundary conditions where either the potential is specied on the boundary (Dirichlet condition) or the electric eld E = is specied (Neuman condition). The problem of determining the potential which satises the correct boundary conditions is generally a non-trivial problem and several methods has been developed to approach the problem for different types of boundary condition. This problem we shall not discuss further. Instead we shall assume that the simple form (11.8) of the potential in the open, innite space to be valid. The integral expression for (r) in a sense solves the electrostatic problem, even if the integral has to be evaluated for a specied charge distribution in order to determine the electrostatic potential. However, when far from the charges the integral can be simplied by use of a multipole expansion, and we will next study how such an expansion can be performed to give useful approximations to the electrostatic potential.
11.1.1
Multipole expansion
This expansion is based on the assumption that the point r where we are interested in determining the potential and the electric eld is at some distance from the charge distribution. To be more precise let us assume that the charge distribution has a nite extension of linear dimension a, as illustrated in Fig. 11.2. We assume the point where to nd the potential lies at a distance form the charges which is much larger than a. With the origin chosen to lie close to the charges we may write this assumption as r >> r a (11.13)
where r is the variable in the integration over the charge distribution. In the integral formula for the potential we may introduce the small vector r /r, and make a Taylor expansion in powers of the vector.
180
r r
Figure 11.2: Multipole expansion. The expansion is based on the assumption that the distance to the point
where the electric eld is determined (measured) is large compared to the linear size a of the charge distribution. The ratio r /r, between the distance r to a part of the charge and the distance r to where the potential is evaluated is used as expansion parameter in the multipole expansion.
The inverse distance between the points r and r, when expressed in terms of is 1 |r r | = (r2 + r 2 2r r ) 2 = = 1 1+ r r r
2
1
r r 2 r r
1 2
1 2
r 1 1 2 + 2 r r 1 f ( ) r
(11.14)
We make a Taylor expansion of the function f ( ) introduced at the last step, f ( ) = f (0) +
i
f 1 (0) + 2
ij
2f + ... i j (11.15)
= 1+
r + r
ij
1 xi xj (3 2 ij )i j + ... 2 r
and re-introduce the integration variable r , in the corresponding expansion of the inverse distance 1 1 rr 1 = (1 + 2 + |r r | r r 2 1 4 0 r 3 (r r )2 r 2 2 r4 r + ...) (11.16)
For the electrostatic potential this gives the following expansion (r) = (r ) 1 + rr 1 + 2 r 2 3 (r r )2 r 2 2 r4 r + ... d3 r (11.17)
0 (r) + 1 (r) + 2 (r) + ... with n as the nth term of the expansion of the potential in powers of .
11.1. THE ELECTROSTATIC EQUATION We consider the rst terms in the expansion, beginning with the monopole term, 0 (r) = 1 4 0 r (r ) d3 r = q 4 0 r
181
(11.18)
where q = (r ) d3 r as the total charge of the charge distribution. This shows that the lowest order term of the expansion gives a potential which is the same if the total charge was collected in the origin of the coordinate system. This rst term will give a good approximation to the true potential if the point r is sufciently far away and the origin is chosen sufciently close to the (center of the) charge distribution. The second term of the expansion is the dipole term, 1 (r) = 1 4 0 r 3 (r )r r d3 r = rp 4 0 r3 (11.19)
where we have introduced the electric dipole moment, p= (r) r d3 r (11.20)
This term gives a correction to the monopole term, and we note that for large r it falls off like 1/r2 while the monopole term falls off like 1/r, so the monopole term will always dominate the dipole term for sufciently large r (unless q = 0). We include one more term of the expansion in our discussion. That is the electric quadrupole term, 2 (r) = 1 8 0 r3 (r )(3(n r )2 r 2 ) d3 r = Qn 8 0 r 3 (11.21)
with n = r/r as the unit vector in direct of the point r and Qn as the quadrupole moment about the axis n. It can be written as Qn = ij Qij ni nj , with Qij = (r )(3xi xj r2 ij ) d3 r (11.22)
as the quadrupole moment tensor. The electric eld can now be expanded in the same way, E = E0 + E1 + E2 + ... (11.23)
with En = n for the n th term in the expansion. We give the explicit expressions for the rst two terms. The monopole term is E0 = 0 = q r 4 0 r3 (11.24)
which is the Coulomb eld of a point charge q located in the origin. The next term is E1 = 1 = ( rp 1 )= (3n(n p) p) , 3 4 0 r 4 0 r 3 (11.25)
with n=r/r as before. This eld is called the electric dipole eld.
182
It should be clear from the above construction that the higher the multipole index n is, the faster the corresponding potentials and electric elds fall off with distance. Thus for large r the nth multipole term of the potential falls off like r(n+1) , while the corresponding term of in the expansion of the electric eld eld falls off like r(n+2) . When considering the electric eld far from the charges often it is sufcient to consider only the rst terms of the multipole expansion. In particular that is the case when we are interested in electromagnetic radiation far from the radiation emitter, as we shall soon consider. In that case the eld is determined by the time derivatives of the multipole momenta. Since the total charge is conserved there will be no contribution from the monopole term, but for large r the main contribution will be from the electric dipole term, unless this term is absent.
11.1.2
Elementary multipoles
Elementary multipole elds can be produced by point charges in the following way. An elementary monopole eld is simply the Coulomb eld of a point charge located at the origin. This Coulomb eld has no higher monopole components. A dipole eld is produced by two point charges of opposite sign,q , located symmetrically about the origin, at positions d/2. The dipole moment of the charge conguration is p = q d. This eld has no monopole component, and in the limit where d 0 with qd xed all higher multipole components vanish and the electric eld is a pure dipole eld. Such an electric dipole eld is illustrated in Fig. 11.3.
Figure 11.3: Electric dipole potential. Two electric point charges of opposite sign, but equal magnitude , are
place at shifted positions. Equipotential lines are shown for a plane which includes the two charges. In the gure red corresponds to positive potential values and blue to negative values. The potential diverges towards the point charges.
In a similar way a pure quadrupole eld can be produced by two dipoles of opposite signs that have positions with a relative shift l. For this charge conguration only the quadrupole component of the electric eld survives in the limit l 0 with p l xed. Such an elementary quadrupole eld is shown in Fig. 11.4.
11.2. MAGNETOSTATICS
183
Figure 11.4: Electric quadrupole potential. In this case four point charges with pairwise opposite signs produce
the potentials. The charges form two shifted dipoles of opposite orientation. Equipotential lines are shown for a plane which includes all the four charges. In the gure red corresponds to positive potential values and blue to negative values. The potential diverges towards each of the four point charges.
11.2
Magnetostatics
A = 0 j
When studying magnetic elds from stationary currents the basic equation is (11.26)
with j = j(r) a s a time independent current density. We note that the equation has the same form as the equation for a static electric potential, and the Coulomb eld solution can immediately be translated to the following solution of the magnetic equation A(r) = The corresponding magnetic eld is B(r) = A(r) = 0 4 ( 1 ) j(r ) d3 r |r r | (11.28) 0 4 j(r ) 3 d r |r r | (11.27)
The gradient in the integrand can easily be calculated by changing temporarily the position of the origin so that r = 0. We have 1 r = d dr 1 r r = r r3 (11.29)
Shifting the origin back to the correct position gives 1 rr = |r r | |r r |3 0 4 j(r ) (r r ) 3 d r |r r |3 (11.30)
and gives therefore the following expression for the magnetic eld B(r) = (11.31)
184

P r
r-r
dr I
Figure 11.5: Magnetic eld from a current loop. The magnetic eld at a given point P is a superposition of contributions from each part of the current loop. The eld expressed as a line integral around the loop is derived from the general volume integral of the current density by rst integrating over the cross-section of the conductor, with area A. The above expression gives the magnetic eld from a general stationary current distribution. However, another form is often more useful, and that corresponds to the situation where the magnetic eld is produced by a current in a thin conducting cable. When the cross section can be regarded as vanishingly small the volume integral of the current density can be replaced by the line integral of the current along the curve dened by the thin cable. To nd this expression we use the following replacement in the integral, as illustrated in Fig. 11.5, j d3 r j A dr = j A dr = Idr (11.32)
Here A is the cross section area of the cable and I is the current running in the cable. This gives the following line integral representation of the magnetic eld B(r) = 0 I 4 (r r ) dr |r r |3 (11.33)
with C as the curve that the current follows. This expression for the magnetic eld produced by a stationary current is known as the Biot-Savart law.
11.2.1
Multipole expansion for the magnetic eld
For positions r far from the current a similar multipole expansion can be given for the magnetic as for the electric eld. We then expand the integrand of (11.31) in powers of r /r, and for the vector potential that gives A(r) = = 0 4 0 4r j(r ) 3 d r |r r | j(r ) 1 +
rr 1 + 2 r 2
(r r )2 r 2 2 r4 r
+ ... d3 r (11.34)
A0 (r) + A1 (r) + A2 (r) + ...
11.2. MAGNETOSTATICS
185
We derive now the explicit expression for the rst two terms of the expansion. The monopole term is A0 (r) = 0 4r j(r ) d3 r (11.35)
As we shall see this term in fact vanishes identically. We examine one of the vector components (the x component) of the integral jx (r) d3 r = = x j(r) d3 r (x j(r)) d3 r x j(r) d3 r (11.36)
and rst note that by use Gauss theorem the rst term can be re-written as a surface integral (x j(r)) d3 r = x j(r) dS (11.37)
when the integral is restricted to a nite volume V with boundary surface S . This shows that when V is expanded so that S is outside all the relevant currents then the integral vanishes. We are left with the contribution from the second term in the last line of (11.36). This is rewritten by use of the continuity equation for charge j+ to give jx (r) d3 r = x 3 d d r= t dt x d3 r = d px dt (11.39) =0 t (11.38)
where px is the x component of the electric dipole moment p. Since similar expression are valid for the other vector components we conclude that the following identity is valid j(r, t) d3 r = p (11.40)
and it is valid not only for stationary, but also for time dependent currents. In the case we consider here, with stationary currents, the time derivative of the dipole moment vanishes and therefore also the integral over the current. This gives A0 = j d3 r = 0 (11.41)
The vanishing of the monopole term seems reasonable from or previous discussion of the lack of magnetic monopoles in Maxwells equations. The next term to consider is the magnetic dipole term, A 1 (r) = 0 4r3 (r r ) j(r )d3 r (11.42)
Also here we have to make use of some identities in order to re-write the integral. We rst consider the following identity, (r j) r = j(r r ) r (j r) (11.43)
186
and examine further the volume integral of the last term, r (j(r ) r) d3 r = ek xl
k
xk jl (r ) d3 r
(11.44)
where ek , k = 1, 2, 3 are the Cartesian unit vectors. We manipulate the last integral, and leave for simplicity out the prime of the variables xk jl (r) d3 r = = xk (xl j(r)) d3 r (xk xl j(r)) d3 r xl (xk j(r)) d3 r xk xl j(r) d3 r (11.45) By the same argument as used before, the rst term, which can be rewritten as a surface integral, vanishes when the boundary of the volume is outside the region with currents. For the second term we use xk j(r) = ek j = jk , with ek as the unit vector in the direction of the xk coordinate axis, and in the last term we apply the continuity equation, j = t . This gives xk jl (r) d3 r = xl jk (r) d3 r d dt xk xl (r) d3 r (11.46)
The last term is the time derivative of a part of the electric quadrupole moment. However, we now consider a situation with time independent sources and therefore the contribution from this term also vanishes. We are therefore left with the identity xk jl (r) d3 r = xl jk (r) d3 r (11.47)
which shows that the two indices k and l can be interchanged when combined with a change of sign. When the symmetry under interchange of indices is introduced in the original integral expression, we get the following identity r (j(r ) r) d3 r = and together with Eq.(11.43) this implies (r j) r d3 r = 2 (r r ) j(r ) d3 r (11.49) r (j(r ) r ) d3 r (11.48)
For the vector potential this nally gives the following expression A1 (r) = = 0 [ (r j(r )) d3 r ] r 8r3 0 m r 4 r3 (11.50)
where m is the magnetic dipole moment of the current distribution, dened as m= 1 2 (r j) d3 r (11.51)
11.2. MAGNETOSTATICS The corresponding magnetic dipole eld is B1 (r) = A1 (r) mr 0 ( 3 ) = 4 r = 0 (3n(n m) m) 4r3
187
(11.52)
with n = r/r as before. We note that the form of the magnetic dipole eld is precisely the same as that of the electric dipole eld, with the electric dipole moment p replaced by the magnetic dipole moment m.
11.2.2
Force on charge and current distributions
The electric and magnetic multipole moments appear in various ways in electromagnetic theory. One of these is when we consider electromagnetic radiation, and we shall discuss that in the next section, another one is when we consider the electromagnetic force on a body with a non-vanishing charge or current distribution. We consider the last situation here. Let us rst consider a body with a given charge density (r) that is subject to an electric eld E(r) that varies slowly over the charge distribution. Assume we choose the origin at a central point of the body and make an expansion of the eld around this point, E(r) = E(0) + r E(0) + ... The total force that acts on the body is then Fe = E dV dV + [ r dV ] E(0) + ... (11.54) (11.53)
= E(0)
= q E + (p )E + ...
with q as the total charge of the body and p as the electric dipole moment. We note in particular the expression for the dipole force acting on the body. The multipole moments also appears in the torque acting on the body Me = = r E dV (r) r (E(0) + r E(0) + ...) dV (11.55)
= p E + ....
In the expressions above one should note that E is the external eld acting on the charge distribution. The internal eld from one part of the charge distribution to another part does not contribute, since internal forces do not contribute to the total force or torque.
188
We may describe in a similar way the magnetic force and torque acting on a current distribution. The force is Fm = = j B(r)dV j(r) (B(0) + (r )B(0) + ...)dV (11.56)
For a stationary current the rst term gives no contribution since and the magnetic force is therefore Fm = (m )B + ... Similarly the torque is Mm = m B + ...
j(r)dV = 0 as previously discussed
(11.57)
(11.58)
In both these expressions we have only included the lowest non-vanishing multipole contributions which are the magnetic dipole terms. We note that these terms have precisely the same form as the corresponding terms for the electric force and torque, with the electric moments and elds replaced by the magnetic moments and elds.
Chapter 12
Electromagnetic radiation
We consider now the full problem of solving Maxwells equations with time dependent sources. The eld equation, in the Lorentz gauge, is as before A = 0 j , A = 0 (12.1)
where the current may now depend both on space and time coordinates, j = j (r, t). The solution involves a retardation effect, since the eld at some distance from the source will respond to changes in the source at a delayed time, in accordance with the fact that the speed of wave propagation is nite. We shall, as the next step, examine how this retardation effect gives rise to the phenomenon of electromagnetic radiation.
12.1
Solutions to the time dependent equation
We note that also in this general case, with time dependent sources, the equations for each vector component of A can be solved separately, and the equations are all of the same form. The Lorentz gauge condition A = 0 is automatically taken care of by the continuity equation j = 0. In non-covariant form the differential equation to be solved is 1 2 ) f (r, t) = s(r, t) (12.2) c2 t2 where f represents one of the components of the potential and s represents the corresponding component of the current density. When discussing solutions of this equation we consider the source term s(r, t) as a known function while f (r, t) is the unknown function, to be determined as a solution of the differential equation. To proceed we introduce the Fourier transformation of the equation with respect to time. For the function f (r, t) this transformation is (2
f (r, t) =
(r, )eit d , f
(r, ) = 1 f 2
f (r, t)eit dt
(12.3)
and the same type of transformation formulas are valid for the source function s(r, t). In the Fourier transformed version time t is then replaced by the frequency variable , while the space coordinate r is left unchanged. Applied to the differential equation (12.2) the transformation gives the following equation for the Fourier transformed elds, (2 + 2 )f (r, ) = s (r, ) c2 189 (12.4)
190
CHAPTER 12. ELECTROMAGNETIC RADIATION
This differential equation, which only includes derivatives with respect to the space coordinates, shows 2 makes it a clear resemblance to the electrostatic equation. However, the presence of the constant c2 different. The differential equation (12.4) is known as Helmholtz equation. Even if there is a difference, we may take some inspiration from the Coulomb problem. As we have earlier discussed, the usual way to nd the solution of the electrostatic problem is rst to nd the electrostatic potential of a point charge, and to use this to nd a general solution by integrating over the actual charge distribution. For a point charge q the charge distribution is (r) = q (r), and the electrostatic equation is 2 = and the solution is the Coulomb potential = q 4 0 r (12.6)
0
q
0
(r)
(12.5)
This shows that we (formally) have the following expression 2 1 r = 4 (r) (12.7)
We will show that a similar relation is valid when a constant is added to 2 , in the following way (2 + 2 ) eir r = 4 (r) (12.8)
which gives the solution of the Helmholtz equation for a point source. To this end we evaluate the action of the Laplacian on the function introduced above, 2 eir r = 1 2 ir 1 e + 2 eir + eir 2 r r 2 1 = eir + eir 2 r r 1 r (12.9)
which gives (2 + 2 )( eir ) = 4eir (r) = 4 (r) r (12.10)
In the last step we have used the fact that the delta function vanishes unless r = 0, and in this point the exponential function is equal to 1. We immediately re-write the above equation in a form directly related to the problem we would like to solve 2 ei c r ( + 2 )( ) = 4 (r) c r
2
(12.11)
Note that in this expression we have explicitly made use of the fact that is only up to a sign determined by 2 . Our interpretation of this equation is now the following. Assume we modify the electrostatic equation by adding the term proportional to 2 /c2 . This change in the eld equation will modify the potential set up by a point charge so it is no longer a Coulomb potential. Actually the
12.1. SOLUTIONS TO THE TIME DEPENDENT EQUATION
191
modication is not unique, the Coulomb potential can be modied either by the factor exp(+i c r ) or exp(i r ) . However, as we shall soon see, there is a reason for choosing one these as the physical c solution. When the potential is found for a point charge we can nd the potential for a charge distribution by integrating over the distribution, in the same way as done for the Coulomb problem. Thus the potential is seen as a superposition of the potential set up by each small part of the charge distribution. This gives with s (r, )/4 as the source term the following solutions 1 f (r, ) = 4 ei c |rr | s (r , )d3 r |r r |
(12.12)
where the distance r from the pont charge now is replaced by the distance |r r | form the point of integration. The corresponding time dependent solution of the original equation is found as the Fourier integral
f (r, t) =
it f d (r, )e
1 4
ei(t
|rr | ) c
s (r , )d
1 d3 r |r r |
(12.13)
|rr | c ),
We recognize the integral in the brackets as the Fourier integral of the function s(r , t this gives for f the following expression 1 f (r, t) = 4
r| s(r , t |r c ) 3 d r |r r |
and
(12.14)
The solutions we have found are similar in form to the Coulomb potential, since the potential f is determined as the integral of the source term divided by the distance between the source and the point of the potential. But there is one important difference which has to do with the time dependence. The potential at a given time t is determined by the source at another time t = t |r r |/c. One of these is earlier than t and the other is later than t. The solution f is called the retarded solution, since t < t, and the effect that the source has on the eld therefore is delayed in time. Similarly f+ is called the advanced solution since t+ > t and the effect that the source has on the potential is advanced in time. For this reason we usually consider the retarded solution f as the physical one. Note however that Maxwells equations accept both these solutions, since they are invariant under time reversal, t t. We should understand the two types of solutions as corresponding to different types of boundary conditions. Usually we specify initial conditions, with the solution of Maxwells equation given as the retarded potential. But it is also possible to specify nal conditions with the solution given as the advanced potential. It is of interest to note that the two space time points (r, t) and (r , t ) can be connected by a light signal, since we have (r r )2 c2 (t t )2 = 0 (12.15)
as we can readily check. Thus (r , t ) lies on the past light cone relative to (r, t), while (r , t+ ) lies on the future light cone.
192

ct
(ct+, r) (ct, r) y
A
x
(ct- , r)
Figure 12.1: Advanced and retarded space time points. Given a point A with coordinates (ct, r) a point B
with retarded time coordinate t = t |r r |/c is located on the past light cone of the point A. This means that a light signal emitted from B can reach the point A. Similarly a point C with advanced time coordinate t+ = t + |r r |/c is located on the future light cone of the point A. A light signal emitted from A is then able to reach the point C .
12.1.1
The retarded potential
We now translate the results we have found to expressions for the electromagnetic potentials. In the following we shall consider only the retarded solutions, which we regard as the physical ones. For the scalar and vector potentials the expressions are 1 4 0 0 4 (r , t ) 3 d r |r r | j(r , t ) 3 d r |r r | (12.16)
(r, t) = A(r, t) =
with t = t |r r |/c referred to as the retarded time. It is interesting to note that the potentials we have found have precisely the same form as the potentials previously found with static sources. The only effect of the time dependence sits in the retardation effect, the effect that there is a time delay between the change in the charge and current distributions and the effects measured in the potentials. This means that the volume integrals in the expressions for and A are not integrals over space at constant t. Instead they are integrals over the three-dimansional past light cone of the point r. Even if the effect of time evolution of the source terms looks simple (and innocent) when we consider the potentials, that is not so when we consider the electromagnetic elds E and B. This is because the retarded time t depends on r and r . When the elds are expressed through derivatives of the potentials, this dependence of r gives rise to new terms in the expressions for E and B. These terms have an immediate physical interpretation. They describe radiation from the time dependent sources.
12.2. ELECTROMAGNETIC POTENTIAL OF A POINT CHARGE
193
ct A (ct, r)
(ct- , r)
Figure 12.2: Retarded point on the world line of a point charge. Given a point A with coordinates (ct, r) there
is only one point B which is both located on the world line and on the past light cone of A. This means that there is only one point in the volume integral of the retarded electromagnetic potentials that gives a contribution at the point A.
12.2
Electromagnetic potential of a point charge
In this case the charge and current densities are expressed as (r, t) = q (r r(t))
j(r, t) = q v(t) (r r(t))
(12.17)
with q as the charge, r(t) as the time dependent position of the charge and v(t) as the velocity. The presence of the delta function means that in the expressions we have derived for the potentials produced by charges and currents, the integral over the densities will get contributions only from a single point. However, there is a complication due to the retardation effect. We consider rst the scalar potential, (r, t) = q 4 (r r(t )) 3 d r |r r | (12.18)
One should note that the retarded time t = t |r r |/c is a function of the integration variable r and this we have to take into account when integrating over the delta function. It is convenient to introduce the argument of the delta function as a new integration variable, r = r r(t ) (12.19)
where we note that the vector r in the denition of t is a constant under the integration. The change of variable introduce a change in the integration measure given by d3 r = Jd3 r (12.20)
where J is the Jacobian of the transformation, which is the determinant of the matrix with elements Jkl = xk xl (12.21)
194
We nd for this matrix element the following expression Jkl = kl dxk t dt xl 1 r 2 + r 2 2r r = kl + vk (t ) c xl xl xl 1 = kl vk (t ) c |r r |
(12.22)
To simplify expressions we introduce (t) = v(t)/c and n = (r r )/|r r |. The matrix element of the Jacobian can then be written as Jkl = kl k nl (12.23)
When calculating the corresponding determinant it is useful temporarily to chose the x axis in the direction of n, which gives n1 = 1, n2 = n3 = 0. The result is simply 1 1 which we re-write in a coordinate independent way as J =1n The integral in the expression for the potential can now be evaluated, (r, t) = q 4 (r ) 1 d3 r |r r | 1 n (12.25) (12.24)
In this integral the effect of the delta function is simply to put r = 0, which is equivalent to r = r(t ), and the potential can therefore be written as (r, t) = q 4 0 |r r(t )|(1 (t ) n(t ) (12.26)
To simplify this expression we introduce the relative vector R(t) = r r(t) and use the label ret to indicate that expression should be evaluated at time t = t . (r, t) = q 4 0 (R R)ret 0 q 4 v RR (12.27)
The vector potential can be found in precisely the same way, and we simply give the result A(r, t) = (12.28)
ret
The expressions we have found for the potentials of a moving point charge are called the LienardWiechert potentials. We note that these expressions are valid with no restriction on the motion of the charge; it may be at rest, move with constant speed or be accelerated. Therefore the potentials implicitly contain all effects of charge in motion, in particular radiation from an accelerated charge. There is a clear similarity between the expressions found here and that of the Coulomb potential of a stationary point charge. This we see most clearly if we choose as inertial frame the rest frame of the moving charge at the retarded time t . In this frame the potential are (r, t) = q , 4 0 Rret A(r, t) = 0 (12.29)
12.3. GENERAL CHARGE AND CURRENT DISTRIBUTION: THE FIELDS FAR AWAY
195
This is simply the Coulomb potential with the distance to the charge determined by its position at the retarded time. This gives us a simple picture of how the potential in the surrounding space-time is created by the moving charge. Each point along its trajectory determines the potential on the future light cone from the chosen point as a Coulomb potential in the rest frame of the charge. If the charge is accelerated this rest frame changes along the path and this means that the potential is not that of a Coulomb potential in a xed inertial frame.
12.3
General charge and current distribution: The elds far away
We consider now the potentials of a general time-dependent charge and current distribution, but restrict the discussion to points r that are far away from the distribution. In that case the same approximation technique as used in the multipole expansions of static distributions can be used. With r denoting the position at which the potential is evaluated and r as the integration variable over the charge distribution. We again assume the origin to be chosen close to the charges so that r /r is a small quantity which we can use as an expansion parameter. The distance to the charge distribution is as before given by |r r | = r rr + ... r (12.30)
The expression for the retarded time can be expanded in a similar way, t = t r rr + + ... c rc (12.31)
We include now only the terms to order r in these expansions. When considering the scalar potential we need to make an expansion of the charge density r r r r (r , t ) = (r , t ) + (r , t ) + ... c rc t c r r = (r , tr ) + (r , tr ) + ... rc t
(12.32)
where we have introduced tr = t r/c, which is the retarded time, not for a general point r of the charge distribution, but rather of the origin r = 0. This we assume to be a central point of the distribution. From the above expressions we nd (r , t ) 1 r r = (r , tr ) + 2 (r , tr ) + ... |r r | r r c t (12.33)
where we in this expansion keep only terms that falls off with distance as 1/r or slower. This expression is now inserted in the integral expression for the potential, which gives (r, t) = = = 1 (r , t ) 3 d r 4 0 |r r | 1 r 1 (r , t )d3 r + 4 0 r c 4 0 r2 c ret q rp + + ... 4 0 r 4 0 r2 c
(r r )
r (r , t )d3 r + ... t c (12.34)
196
with q as the total charge and p as the electric dipole moment. In the expression for the potential it is the time derivative of the dipole moment at the retarded time tr = t r/c that enters. The dipole term is only the rst term of a multipole expansion of the potential, with the quadrupole term as the next. One should note that when only terms that fall off like 1/r for large r are included, the static multipoles do not contribute, but the time derivative of these do. In fact there are contributions from all higher multipoles, but for the 1/r terms the number of derivatives increases with the degree of the multipole, so that the second derivative of the quadrupole term contributes etc. We continue now to analyze the vector potential in the same way. The general expression is A(r, t) = 0 4 j(r , t ) 3 d r |r r | (12.35)
where a Taylor expansion is introduced for the current density in the same way as done above for the charge density. We nd A(r, t) = 0 4r r 0 j(r , t )d3 r + c 4cr2 (r r )[ j r (r , t )]d3 r + ... t c (12.36)
The rst term can be expressed in terms of the electric dipole moment, since have (t) j(r, t)d3 r = p which is an identity that we have earlier demonstrated (see Eq.(10.49)). To re-write the second term another identity, given by Eq.(11.46), will be needed xk jl (r) d3 r = This implies (r r )j(r , t)d3 r = 1 [ 2 r j(r , t)d3 r ] r + 1d 2 dt r (r r )(r )d3 r (12.39) xl jk (r) d3 r d dt xk xl (r) d3 r (12.38) (12.37)
1 = m r + rD n 2
where we have introduced the magnetic dipole moment m and an electric quadrupole vector Dn dened by m = Dn = 1 2 r j(r , t)d3 r r (r n)(r , t)d3 r (12.40)
with n = r/r as the unit vector in direction of r. By use of these expressions we are able to write vector potential as A(r, t) = 0 1 1 + m n+ D (p n + ...)ret , 4r c 2c n= r r (12.41)
where the subscript ret now means that the vectors should be taken at retarded time tr = t r/c. We note that both the electric and the magnetic dipole momenta, as well as the electric quadrupole
12.4. RADIATION FIELDS
197
moment contributes to the potential. (In the case of the scalar potential we did not include all these terms.) Let us next consider the magnetic eld that corresponds to the vector potential that we have found, B(r, t) = A(r, t) 1 0 1 + m n+ D r (p = n + ...)ret 4r2 c 2c 0 d 1 1 + m n+ D + (tr ) (p n + ...)ret 4r dt c 2c + ...
(12.42)
where the last term comes from the r dependence of the retarded time tr . The terms that come from derivatives of n have not been written out explicitly. We have 1 r tr = (t ) = n c c which gives for the magnetic eld B(r, t) = A(r, t) 0 1 + m n+ = (p 2 4r c 0 1 n+ + ( p+ m 4rc c + ... (12.43)
1 Dn + ...)ret n 2c 1 ... Dn + ...)ret n 2c (12.44)
One should note the difference in r dependence of the different terms. The ones that are obtained by differentiation through the retarded time variable tr fall off like 1/r, but the others, where the differentiation acts directly on the r dependent functions, fall off like 1/r2 for large r. One should also remember that in the above expression many terms have already been suppressed since they fall off rapidly with distance. This is in particular so for the static terms of the multipole expansion which we have examined earlier. A similar expression as for the magnetic eld (12.44) is found for the electric eld, but we do not write it out explicitly. In the following we shall also restrict the discussion to the radiation eld, which is the part of the eld which dominates far from the sources.
12.4
Radiation elds
In the following we shall assume to be sufciently far away from the charge and current distribution so that only terms that fall off with distance as 1/r give substantial contributions. This region is called the radiation zone. The rst term in the expression (12.44) can then be neglected and we get as expression for the magnetic component of the radiation eld, Brad (r, t) = 0 1 1 ... n) n + Dn n + ...)ret ( p n + (m 4rc c 2c (12.45)
To nd the corresponding expression for the electric eld we may write the E eld in terms of the potentials and follow the same procedure as for B. However, we may make a short cut in the following way. In the radiation zone the elds can in the neighborhood of a point r far from the sources be
198
regarded as a plane wave which propagate in the direction n. (It is not necessarily a monochromatic plane wave since the Fourier transform of the time dependent multipole momenta may contain more than one frequency.) But as previously shown, for electromagnetic plane waves we have a simple connection between E and B that is not dependent of the frequency of the wave, E = cn B. In the present case the electric eld therefore takes the form
Erad (r, t) =
0 1 1 ... n + (Dn n) n + ...)ret (( p n) n m 4r c 2c
(12.46)
The radiation elds given above are the parts of the total elds that fall off with distance as 1/r. They dominate in the radiation zone, far from the charges and currents. To be more precise there are two conditions that should be satised to be in this zone. The rst one is r >> a with a as a typical linear size of the charge and current distribution. Our derivation so far has been based on this to be satised. If that is not the case there will be elds with a faster fall off with distance (which we have omitted in the expansions) that would compete with the radiation elds in strength. The other is r >> , with as a typical wave length of the radiation. If that is not satised there are contributions to the elds where a smaller number of time derivatives of the multipole momenta could compensate for a higher power in 1/r. In particular this condition is necessary for the second term in (12.44) to dominate over the rst term. If furthermore we have the following condition satised , >> a, then the rst terms of the multipole expansions of the radiation eld, (12.45) and (12.46), would dominate over the later ones, so that the electric dipole contribution would be more important than the electric quadrupole contribution etc. The electric and magnetic dipole contributions may seem to be giving comparable contributions to the radiation, but under normal conditions that is not the case. The reason for this is that the magnetic moment depends on the charge currents and therefore on the velocity of the charges (usually electrons) this implies that the magnetic dipole term would be damped by a factor v/c relative to that of the electric dipole term, with v as the (average) velocity of the charges. So usually electric dipole radiation would be dominating the radiation for example from an an , and it is for tenna. This contribution to the radiation is described by the term which depends on p short referred to as the E 1 radiation term. However, under certain conditions this type of radiation may be suppressed so that magnetic dipole radiation would be dominating. This is the term depending , with the short hand notation M 1. Similarly electric quadrupole radiation, referred to as E 2, on m may also under certain conditions be important, etc. It is interesting to note that the radiation elds appear as a direct consequence of the retardation effect in the solutions of elds from time dependent charge and current distributions. This is clearly seen in our derivation of the magnetic eld B from the leading part of the vector potential A. This part of the potential falls off with distance as 1/r and differentiation of this factor leads to a 1/r2 dependence. That is seen in the rst term of (12.44). However, the retarded time also depends on r, and when the differentiation is done through this time dependence that gives the second term in (12.44) with a 1/r dependence. This is the magnetic component of the radiation eld.
199
12.4.1
Electric dipole radiation
When the electric dipole terms dominate the radiation, the expressions for the radiation elds simplify to Erad (r, t) = Brad (r, t) = Poyntings vector for this eld is S(r, t) = = = = 1 Erad Brad 0 c 2 B n 0 rad 0 ( pret n)2 n 16 2 r2 c 0 ( p2 sin2 )ret n 16 2 r2 c ( pret n) n 4 0 rc2 0 ret n p 4rc 1
(12.47)
(12.48)
and n and the subscript where the angle introduced in the last step is the angle between the vectors p ret is a reminder that all variables at the source should be taken at the retarded time tr = t r/c. Since S(r, t) gives the energy current density of the electromagnetic eld, the above expression shows that the radiation is, as one should expect, directed in the radial direction n away from the source of the radiation. The total power radiated is given as the integral of S over all angles, P = = = 0 2 p dd sin3 16 2 c ret 1 0 2 ret p du(1 u2 ) 8c 1 0 2 p 6c ret
(12.49)
For radiation from a linear antenna the direction of the electric dipole moment is xed by the direction of the antenna and only the amplitude oscillates in time. The angular distribution of the radiation then has the simple form S(r, t) = 0 p 2 ret sin2 n 16 2 r2 c (12.50)
where only the amplitude determined by p 2 ret is time dependent, while the direction of the dipole, given by the angle is constant. The angular distribution of the radiated energy is illustrated in Fig. 12.3, and we note in particular that maximum of the radiation is in the direction perpendicular to the direction of the antenna. Let us further assume the time variation of the electric dipole moment of the antenna to have a simple harmonic form, p(t) = p0 cos t (12.51)
200
y S
Figure 12.3: Angular distribution of radiated energy in electric dipole radiation. The electric dipole moment p
here oscillates along the x axis. The magnitude of the Poynting vector S, which gives the angular distribution of the radiated power is indicated in the gure.
with oscillation period T = 2/ . The expression for the time averaged radiated power from the antenna is then 1 T P (t)dt T 0 T 4 0 p 2 0 cos2 t dt = 6c 0 4 0 p 2 0 = (12.52) 12c We note in particular that, for xed p0 , the radiated power increases rapidly with the frequency of the oscillating dipole moment. = P
12.4.2
Example: Electric dipole radiation from a linear antenna
Let us assume a linear antenna of length L is directed along the x axis as illustrated in the gure. Let us further assume an oscillating current is induced in the antenna, of the form x I (x, t) = I0 cos( ) cos t (12.53) L The x dependence of the current shows that it has its maximum at the midpoint of the antenna and that it vanishes, as it should, at the endpoints. Charge conservation now gives a connection between the space variation in current and the time variation in the charge density, which has the form I + =0 (12.54) x t where is the linear charge density, i.e., the charge per unit length along the antenna. The equation is the one dimensional form of the continuity equation for the charge, which we earlier have formulated as an equation in three space dimensions.
201
I(x)
(x)
x -L/2 L/2
Figure 12.4: Oscillating current and charge in a linear antenna. The gure shows the current I (x) and charge
density (x) along the antenna, where the current here has a cosine form and the charge density a sinus form as functions of x. They both oscillate in time, with a phase shift of /2, so that the charge density vanishes when the current has its maximum and vice versa.
The electric dipole moment can in this case be expressed as a one dimensional integral along the antenna, p(t) =
L 2
(x, t) x dx
(12.55)
L 2
with direction p = p i along the x axis. For the time derivative we nd p = = = = =

L 2
L 2
x dx t
L 2
L 2
L 2
I x dx x ( xI ) dx + x
L 2
I dx
L 2
L 2
L 2
I dx (12.56)
L 2
2L I0 cos t The corresponding expression for the oscillating dipole moment is p(t) = p0 sin t
(12.57)
with p0 = 2L/ . The double time derivative of the dipole moment, which is needed for the radiation formula, is 2L p (t) = p0 2 sin t = I0 sin t (12.58) The formula for the radiated power now gives 0 2 P (t) = p 6c ret = 0 2 4 2 p sin tr , 6c 0 tr = t r/c (12.59)
202
which, when expressed in terms of the current amplitude, is P (t) = For the time average of the power this gives = P
2 0 L2 2 I0 3 3 c 2 2 0 L2 2 I0 sin2 tr 3 3 c
(12.60)
(12.61)
since the average value of sin2 t is 1/2. We note that the radiated power, for xed current I0 , increases quadratically with the oscillation frequency of the dipole moment. Let us nally consider the polarization of the radiation as it is measured by a receiver. As we already know both E and B are orthogonal to the direction of wave propagation which is given by n, the unit vector pointing from the antenna to the receiver. Since the dipole moment oscillates in strength but not in direction, the general expressions for the electric and magnetic elds produced in electric dipole radiation, (12.47), shows that B will be oscillating along the xed line i n and E along the xed line (i n) n. This means that the radiation eld will for any direction n of propagation be linearly polarized. The polarization plane, which is the plane dened by the direction of wave propagation and the direction of the oscillating E eld is identical to the plane spanned by the direction of the antenna and the direction from the antenna to the receiver. This is so since E oscillates in this plane in the direction perpendicular to n. The magnetic eld will then oscillate along the line orthogonal to the polarization plane.
12.5
Larmors radiation formula
As a last point we shall consider radiation from an accelerated point charge. The elds produced by a moving point charge has previously been given in the form of the Lienard-Wiechert potentials (see (12.26) and (12.28)). We consider the non-relativistic form of these potentials, which correspond to = v/c 0. The corresponding radiation elds are q0 (a n) n ]ret 4R q0 a n ]ret B(r, t) = [ 4Rc E(r, t) = [
(12.62)
In these expressions we have R(t) = r r(t) , n = R , a(t) = r(t) R (12.63)
with r as the position where the elds are evaluated and r(t) is the time dependent position vector of the moving charge. Note that in Eq.(12.62) the retarded time is measured relative to the position of the moving charge, 1 t = t |r r(t )| c (12.64)
We note that the elds given above have the same form as the electric dipole radiation elds = q a. previously found. Thus for a point charge the dipole moment is p(t) = q r(t) and therefore p There is one difference, since R and n depend on the time dependent position of the charge. However, when sufciently far from the charge this time dependence is less important.
12.5. LARMORS RADIATION FORMULA Poyntings vector for the elds is of the same form as for electric dipole radiation S(r, t) = q 2 0 [a2 sin2 n]ret 16 2 r2 c
203
(12.65)
where is the angle between the direction of the acceleration a and the direction vector n. The formula for the integrated radiated power is P (t) = 0 q 2 2 a 6c ret (12.66)
This is called the Larmor radiation formula. The expressions given above gives a simple picture of the radiation process. At a given time along the space-time trajectory of the charge, it will radiate energy at a rate proportionally to the square of the acceleration of the charge. The energy emitted in a time interval dt will then propagate as an expanding spherical shell radially outwards from the charge. The time delay when the shell moves outwards is the origin of the retardation effect. When the charge moves, the center of these shells of energy will continuously change, so that when viewed from a xed point in space the radiation is at any time directed away from the position of the charge at the retarded time. A qualitative way to understand why the radiation formula has the same form as for electric dipole radiation is based on the fact that the point charge has no spacial extension. Therefore we can disregard the position dependence of the retarded time when integrating over the charge and current distributions. In the derivation of the potentials in Sect. 2.5, this position dependence was the origin of the higher multipole contributions. However, in the present case these additional terms are avoided only when the dipole eld include the effect of motion of the charge rather than referring to a xed origin.
204
Summary
This part of the lectures, on electrodynamics, has been focussed on how Maxwells equations form the basis for our understanding of the variety of electromagnetic phenomena. Beginning from the four equations that constitute the set of Maxwells equations, we have rst seen how these can be compactied into two covariant eld equations which involve the electromagnetic eld tensor rather than the electric and magnetic eld separately. This covariant form is attractive, not only because of its compactness and elegance, but also because of the relativistic invariance of electromagnetic theory is made explicit in the covariant equations. Relativistic invariance and symmetry under Lorentz transformations are important properties of the Maxwell theory. In fact this symmetry was realized as an interesting, although apparently somewhat formal, property of Maxwells equations by people like Henri Poincar e even before the theory of relativity was introduced. But the true importance of these symmetries were understood only after Albert Einstein lifted the Lorentz transformation from being merely an interesting set of symmetries of Maxwells equations to be the fundamental symmetry of all kinds of natural phenomena. When applied to the electromagnetic theory the relativistic invariance predicts the specic way in which the E and B eds are mixed when changing from one inertial reference frame to another. As we have seen, the covariant description of the eld in terms of the electromagnetic eld tensor gives a direct information of how this mixing takes place. The problem addressed in these notes is how to solve Maxwells equation under different conditions. As a rst step it is then of interest to simplify the equations by introducing the electromagnetic potentials. These are not uniquely determined by the E and B elds, and we may therefore impose certain gauge conditions on the potentials to simplify the equations. Both the non-covariant Coulomb gauge and the covariant Lorentz gauge conditions are of interest to use, which one depends on under what conditions we will solve the equations. In these notes we have looked at three different situation. The rst one when the sources of the elds, i.e., the charge and current distributions, vanish. The second one is when the sources (in a given inertial frame) are time independent, and nally when we have the general situation with space and time dependent distributions of charge and current. In all these cases we assume that there are no non-trivial boundary conditions, so we look for solutions in the open innite space, where all elds are nite or tend to zero at innity. In the rst case, where the charge and current densities vanish, Maxwell equations have solutions in the form of freely propagating waves. These are the electromagnetic waves that span a wide variety of phenomena, depending on the frequency of the waves, from the energetic radiation, through X rays, light to microwaves and radiowaves. In our somewhat brief discussion of electromagnetic waves we have focussed on the property of polarization which characterizes all these different types of wave phenomena. The special cases of linear and circular polarization can be understood as depending on the phase and amplitude relations between two orthogonal components of the radiation, and that is also so for the general type of elliptic polarization. When the charge and current distributions are time independent the equations for the electric and 205
206
magnetic elds decouple completely and they can be examined separately in the form of electrostatic and magnetostatic equations. It is easy to nd solutions of these equations by applying the linearity of the equations. Thus the general solutions of the electrostatic problem can be found as a linear superposition of the Coulomb potentials of all the small parts of the charge distributions. The magnetostatic equations are of the same form and solutions can be found by the same method. In both cases the general solutions for the (scalar or vector) potentials can be written as integrals over the charge or current distributions. Even if explicit solutions can be found for the eld equations with general stationary sources, it is often of interest to make simplications for the resulting integral in the form of approximations that are valid for points not to close to the charges and currents. This have been done in the notes in the form of the multipole expansion. This expansion is based on the assumption that the distance from the source to the point where the potential should be determined is much larger than the extension of the source itself. For the electrostatic eld the leading term is the Coulomb potential, the next is the electric dipole potential, then the electric quadrupole potential etc. For the magnetostatic potential there is a similar expansion, but here the leading term is the magnetic dipole potential. There is in fact a simple symmetry between the electrostatic and magnetostatic expansions, so the term for term the elds E and B for dipole, quadrupole etc. are of precisely the same form. The method used with stationary sources can with some modications be used also to solve Maxwells equations with general time and space dependent sources. As a rst step in nding the general solution we have introduced the Fourier transform in time and thereby brought the equations into a form similar to the static cases. The type of differential equation we then meet is not identical to that of the electrostatic case, but the same general method can be used. This means that we rst look for solution of the problem with a point source, which is now a modied Coulomb potential. We next extend this to the general case by making a linear superposition over contributions from all pointlike parts of the charge and current densities. Finally the inverse Fourier transformation gives the solution in the form of an integral over the time and space dependent charge and current distributions. As we have seen the solution is strikingly similar to the corresponding solutions for the electrostatic and magnetostatic potentials. The main difference with time dependent sources is the retardation effect. Thus the integral over the charge and current distributions is not taken at a xed time, but the integral is instead over the past light cone relative to the point where the potential should be determined. The retardation effect one should clearly expect from the theory of relativity, where the inuence of a source on a the eld at a distant point is delayed by the limit of propagation set by the speed of light. Even if it in this sense the effect may look innocent, it contains the important physical effect of radiation from a time dependent source. To see this explicitly we have made a multipole expansion similar to the one applied to the static cases. Far from the sources, in the radiation zone, the elds that fall off with distance as 1/r will dominate. These are the radiation elds, and in the derivation of the electromagnetic elds from the potentials they appear as a consequence of the position dependence of the retarded time. We have found the expressions for the rst few terms of the multipole expansion of the radiation elds, where normally the electric dipole contribution is the most important one, but where under certain conditions also magnetic dipole and electric quadrupole contributions may be signicant. Obviously there are a lot of interesting further developments of the theory that are not covered in these lecture notes. That is true for all the three parts of the notes, where the motivation has been to focus on some of the most important and simplest parts of the classical theory of mechanics and electrodynamics. One of the main objectives have been to show that the analytic approach applied in this part of physic gives the theory an attractive and elegant form, but also to show that these methods are important in solving the fundamental equations and revealing the underlying structure of
12.5. LARMORS RADIATION FORMULA the physical phenomena.
207

Classical Mechanics and Electrodynamics

Uploaded by

Copyright:

Available Formats

Classical Mechanics and Electrodynamics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Classical Mechanics and Electrodynamics

Uploaded by

Copyright:

Available Formats

What are the three main parts covered in the notes?

What are the three main parts covered in the notes?

What method is used to solve Maxwell's equations with general time-dependent sources?

What method is used to solve Maxwell's equations with general time-dependent sources?

Classical Mechanics and Electrodynamics

Lecture notes FYS 3120

9.6 9.7 9.8

CHAPTER 1. GENERALIZED COORDINATES

1.1. PHYSICAL CONSTRAINTS AND INDEPENDENT VARIABLES

CHAPTER 1. GENERALIZED COORDINATES

1.1. PHYSICAL CONSTRAINTS AND INDEPENDENT VARIABLES

Time dependent constraint

CHAPTER 1. GENERALIZED COORDINATES

This corresponds to a time-dependent constraint equation y = (x vt) tan (1.19)

The conguration space

1.2. THE CONFIGURATION SPACE

Such a higher-dimensional surface is often referred to as a hypersurface.

CHAPTER 1. GENERALIZED COORDINATES

1.4. APPLIED FORCES AND CONSTRAINT FORCES

Applied forces and constraint forces

CHAPTER 1. GENERALIZED COORDINATES

Static equilibrium and the principle of virtual work

1.5. STATIC EQUILIBRIUM AND THE PRINCIPLE OF VIRTUAL WORK

CHAPTER 1. GENERALIZED COORDINATES

CHAPTER 2. LAGRANGES EQUATIONS

2.1. DALEMBERTS PRINCIPLE AND LAGRANGES EQUATIONS This further gives mk

can be interchanged. This gives k mk r k r qj

and the dynamical equation can therefore be written as d dt T q j T V = , qj qj j = 1, 2, ..., d (2.17)

CHAPTER 2. LAGRANGES EQUATIONS

2.1. DALEMBERTS PRINCIPLE AND LAGRANGES EQUATIONS

CHAPTER 2. LAGRANGES EQUATIONS

Figure 2.1: Atwoods machine with two weights.

2.1. DALEMBERTS PRINCIPLE AND LAGRANGES EQUATIONS

and for the Lagrange equation this gives d dt L =0 y I (m1 + m2 + 2 ) y + (m2 m1 )g = 0 R m1 m2 y = g I m1 + m2 + R 2 L y

32 pendulum bob are

CHAPTER 2. LAGRANGES EQUATIONS

Symmetries and constants of motion

CHAPTER 2. LAGRANGES EQUATIONS

We consider a Lagrangian of the general form L = L(q, q, t) (2.45)

The conjugate momentum is also referred to as generalized momentum or canonical momentum.

CHAPTER 2. LAGRANGES EQUATIONS

which is easily solved for q 1 , q 1 = 1 (k g11 g1i q i )

Example: Point particle moving on the surface of a sphere

CHAPTER 2. LAGRANGES EQUATIONS

Symmetries of the Lagrangian

40 Invariance of the Lagrangian then implies (

CHAPTER 2. LAGRANGES EQUATIONS

which we may re-write as L d L d L qk ( )qk + ( qk ) = 0 . qk dt q k dt q k (2.83)

With qk as an innitesimal change of the coordinates, it can be written as qk = Jk (2.86)

Example: Particle in rotationally invariant potential

2.2. SYMMETRIES AND CONSTANTS OF MOTION

Time invariance and energy conservation

CHAPTER 2. LAGRANGES EQUATIONS

This shows that the following quantity H=

2.2. SYMMETRIES AND CONSTANTS OF MOTION

hi (q, t)q i + f (q, t)

CHAPTER 2. LAGRANGES EQUATIONS

Generalizing the formalism

A change of the the Lagrangian L(q, q, t) L (q, q, t) (2.108)

2.4. PARTICLE IN AN ELECTROMAGNETIC FIELD