You are on page 1of 181

Theoretical Mechanics

Lecture in WS 2016/17 at the KFU Graz

Axel Maas



1 Introduction 1

2 Newtonian mechanics 3
2.1 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Newton’s laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.2 The first law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.3 The second law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.4 The third law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Point particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 The potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.6 Harmonic oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.7 Central potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.7.1 Angular momentum . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.7.2 Effective potential . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7.3 Planetary motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.8 Mass distributions and center of mass . . . . . . . . . . . . . . . . . . . . . 23
2.8.1 General properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.8.2 Two-particle systems . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.8.3 Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.8.4 Continuous distribution . . . . . . . . . . . . . . . . . . . . . . . . 29
2.8.5 Moment of inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9 Perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.10 Galileo group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.11 Rotating bodies revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.11.1 Pseudo forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.11.2 Tensor of inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35


. . . 41 3. . . . . . . . . . . . . . . . . . . 50 3. . .8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Virial’s theorem . . . . .1 Reformulating d’Alembert’s principle . . . . . . . . . . . . . . . . . 47 3. . . .2 The Euler-Lagrange equations . . . . . . . . . . . . . . . . . .1 The equivalence principle . . . . . .3 Equivalence to Lagrange’s function of the first kind . . 61 4. . . . . . . . . . . . . . . . . 79 5 Hamiltonian mechanics 81 5. . . . . 76 4. . . . . . . .3 Legendre transformations . . . . . . . . . . . . . . . . . . . . 83 5. . . . . . . . . .4 Equivalence to d’Alembert’s principle . . . .4 Equivalence to Newton’s law . . .2 Measurements of time and distance . . . . . . . . . . . . . . . . . . 94 . . .8. . . . . . . . . . . . . . . . .2 Minkowski space . . . . . . . . . . 57 4. . . . . . . . . . . . . . . . . . . 39 3.1. . . . . . . .6 Generalized momenta and cyclic coordinates . .1 Energy conservation . . . . . . . .ii Contents 3 Special relativity 38 3. . . . . . . . . .4. . . . . . . . 75 4. . . . . .4. . . . . . . . . . . . . . . . 39 3. .3 Angular momentum conservation . . . . . . .1 Integral formulation . . . . . . 88 5. . . . . . . 77 4. . . . . .1. . . . . . . . . 65 4. . . . . . . . . .1. 59 4. . . . . . .5 Invariances of Lagrange’s equation of the second kind . . . . . .3 Examples . . . . . . . . . . . . .2 Momentum conservation . . . 69 4. . . . . . . . . . . . . . . .5 Beyond conservative forces . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Generalized coordinates . . . . . . . . . . . . .4 Noether’s theorem . . . . . . . . . . . . .8. . . . . . . . . . .7 Lagrange’s equation of the first kind . . .1 Constraints . . .4. . . . . . . . . . .2 Variational formulation . . . . . . . 81 5. 89 5. . 87 5. . . . . . . . . . . . . . . . . . . . . . . 78 4. . .3 Relativistic kinematics . . . . . 90 5. . . . . . . . . . . 51 4 Lagrangian mechanics 54 4. . . 67 4.4 Relativistic version of Newton’s laws . . .4 Euler-Lagrange equations of the second kind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . 70 4.1. .1 The speed of light and Minkowski space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 5. . . . . . 92 5. .4 Hamilton’s equations . . . . . . . . . . . . . .3 The principle of d’Alembert . . . . . . . . . . 61 4. . . . . . . . . . . . . . . . . . 63 4.2 Phase space and state space . . . . . . . . . . . . . .8. . . . . . . . . . .1. . . . .1. . .8. . . . . . . . 74 4. . . . . . . . . . . . . . . . . . . . . . . . . . .8 Conservation laws . . . . . . . . . .1 Hamilton’s principle . . . . . . . . . . . . . . 67 4. . . . . . . . . . . . . . . . .1. 54 4. . . . . . . . . . . . . . . . . . . . . . . .

Contents iii

5.5 Action principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.5.1 Modified Hamilton’s principle . . . . . . . . . . . . . . . . . . . . . 98
5.5.2 Maupertius’ action principle . . . . . . . . . . . . . . . . . . . . . . 99
5.5.3 Fermat’s principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.5.4 Jacobi’s principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.6 Covariant Hamiltonian mechanics . . . . . . . . . . . . . . . . . . . . . . . 104
5.6.1 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.6.2 Point particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.7 Canonical transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.8 Alternative formulations of canonical transformations . . . . . . . . . . . . 113
5.9 Poisson brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.10 Poisson brackets and (canonical) transformations . . . . . . . . . . . . . . 120
5.11 The Poisson bracket as a differential operator . . . . . . . . . . . . . . . . 123
5.12 Poisson brackets and symmetries . . . . . . . . . . . . . . . . . . . . . . . 125
5.13 Hamilton-Jacobi theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.13.1 General strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.13.2 A solution strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.13.3 Connection to the action . . . . . . . . . . . . . . . . . . . . . . . . 137
5.13.4 Angular and action variables . . . . . . . . . . . . . . . . . . . . . . 138

6 Selected special topics 143
6.1 The Kepler problem revisited . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.2 Spinning top . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.2.1 Euler’s angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
6.2.2 The equations of rotation . . . . . . . . . . . . . . . . . . . . . . . 147
6.2.3 Precession . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.2.4 Top in a gravitational field . . . . . . . . . . . . . . . . . . . . . . . 148
6.3 Continuum mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.3.1 Systems of many oscillators . . . . . . . . . . . . . . . . . . . . . . 150
6.3.2 Continuous systems of oscillators . . . . . . . . . . . . . . . . . . . 156
6.3.3 Continuous systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.3.4 Applications of the Lagrange formulation . . . . . . . . . . . . . . . 160
6.3.5 Hamilton formulation . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.4 Chaos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.5 The geometry of phase space . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.6 The relativistic string . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.7 First-order and second-order constraints . . . . . . . . . . . . . . . . . . . 171

iv Contents

6.8 A preview of quantum physics . . . . . . . . . . . . . . . . . . . . . . . . . 173

Chapter 1


Theoretical mechanics is, in a sense, a somewhat ancient topic. Essentially completely
formulated in its modern form in the 19th century, it has matured into a mathematically
consistent and closed theory. Even the advent of special relativity only required a minor
modification of the underlying vector spaces in the mathematical formulation, and could
therefore be technically easily accommodated. Of course, the interpretational impact has
been of much more significance.
The natural question is therefore why it is necessary to discuss it then at all. The
answer to this is that all modern models in physics draw from the conceptual structure
of classical mechanics. General relativity introduced again modifications of the arena.
Quantum physics modified the nature of coordinates themselves. But the basic formulation
tools of theoretical mechanics, especially the Lagrangian formulation of chapter 4 and the
Hamiltonian formulation of chapter 5, are still mathematical cornerstones of these theories.
It is not the concept which changed, merely the entities on which it is applied, at least
from a mathematical point of view. Of course, again the shifts in physical interpretation
had a much larger impact.
Thus, theoretical mechanics remains both mathematical and in terms of conception
still a cornerstone of even the most modern areas of physics. Understanding theoretical
mechanics in these formulations is therefore forming the foundation on which these are
However, these formulation as Lagrangian and Hamiltonian mechanics in chapters 4
and 5, as powerful as they are, are very dissimilar from the usual concept of Newton’s law,
even if they embody the same physics. In fact, at first sight it is far from obvious that
reformulating mechanics mathematically in this way could be anything but obscuring. It
is only when making the transition to quantum physics and general relativity that their
conceptual importance becomes really evident.


Classical mechanics. from the rather hands-on treatments of the books in the two classic series on theoretical physics of Nolt- ing and Greiner.. This reformulation is especially useful for non-relativistic quantum physics. and it is therefore highly recommendable to choose accordingly to one’s own taste. To provide a smooth transition from the experimental view on mechanics to the theoret- ical formulation. ordinary Newtonian mechanics. as well as the more abstract classical treaty of Goldstein. some of the concepts will also be relevant in statistical physics. which is the arena of particle physics. It may appear odd at first sight that this more special case. Finally. the first step will be to give a more theoretical perspective on Newtonian mechanics in chapter 2. The number of books is essentially legion. This lecture does not follow any of these books particularly. Lagrange’s equation of the second kind in section 4. are therefore role models for the future. Lagrange’s equations of the fist kind in section 4. There are numerous textbooks on the topic of theoretical mechanics. as here the subject of study.2 If one would embark on either topic without first going through the mathematical reformulation of mechanics would require to cope at the same time with quite different mathematical and conceptual problems.7 are extremely useful when it comes to problems in applications of classical mechanics. . is already well acquainted from experimental physics. is treated after the more general case. The next step is then to introduce the ideas of special relativity in chapter 3. the microscopical explanation of thermodynamics. but already in parallel for the case of special relativity. Afterwards. as it does not lend itself so easily to special relativity. and to some extent special relativity. the theo- retical approach to physics problems is best outlined. but draws instead from many sources. both having their own advantages. Any real recommendation is too much depending on personal taste and difficulty preference. In this way. The first version of reformulation of mechanics is then Lagrangian mechanics in chapter 4. sometimes also called analytical mechanics. so that one can concentrated on the more abstract theoretical formulation. another reformulation is given in terms of Hamilton’s mechanics in chapter 5. But this formulation is actually easier to understand from the Lagrangian formulation as then from starting outright towards it. This allows to introduce the central concepts later then not only for Newtonian mechanics. up to the modern treatment of Bartelmann et al. However. these methods will then be used to treat some special topics in classical mechanics in chapter 6.4 are the natural way to formulate problems embodying special relativity and quantum physics simultaneously. There are two versions of this.

Comparison between these derived results and experiments then decides whether the basis of the derivation is actually useful to describe experiments or not. The main aim here is to provide a new perspective on it. to identify this basic set. there are some cases where they do contradict experiments. The resulting basis is called a (standard) model or a theory. the theory turns out to be the limit of a more general theory in a particular case. g. Note. In this case. the movement of the earth around the sun. It can never prove it to be correct. However. Newton’s laws. E. while the constructions still awaiting experimental tests are usual considered as hypothesis. of course. the exposure to physics was mainly by means of understanding experiments. and not (yet?) resolved. Only when experiments become sensitive enough they can detect this situation. e. however. Of course. any theory or model can only be considered to be an adequate description for the time being.. but are not really refuted. e. nor actually in the day-to-day research. g. A special limit theory is by no means useless. This also includes. nobody will do 3 . The basic idea is to start from a set of fixed rules. Newtonian physics is familiar from experimental physics. the aim of this exercise is to end up with the minimal basis to describe all experimental results.Chapter 2 Newtonian mechanics The aim of this first chapter is not necessarily to introduce new physics or new physics concepts. however. to even pose this question requires to be able to connect some limited set of basic principles with experiments. So far. This is often considered as the basic laws of nature. g. There are fine distinctions between these names which. Though often theories are refuted entirely based on contradiction to experiment. Whether they are indeed ingrained in reality in some way is a highly non-trivial questions. This is the task of theoretical physics. When building a bridge. Newtonian mechanics will be the limit of special relativity for small speeds. play no role in this lecture. Thus. After all. The theoretical approach is somewhat different. that any experiment can at most falsify a theory. The next step is then to derive consequences of these laws.

4 2. . while time acts as a parameter to identify the position of the particle along its path. this path is described by a vector-valued function ~r(t) in a vector space. the vector space will be the ordinary R3 . is by no means useless.1. equipped with the usual scalar product and norms to obtain a Hilbert space. sometimes also R2 or even R1 . usually only the second differential. e. Quite often the more general theory is of less practical use. and there is no answers to why a particle behaves in a certain way. more often called an effective theory. Similarly. d~v (t) d2~r(t) ~a(t) = = dt dt2 plays also an important role. The others may appear. Usually. their paths will be indexed. the path can be obtained by a single or even more integrations. it is useful to first consider the description of (point) particles. This vector space describes the position of the particle in space. To separate this from the actual reasons of their movement this is often called kinematics. which is also called a trajectory. Zt Zt′ ~r(t) = ~r(t0 ) + ~v(t0 )(t − t0 ) + dt′ dt′′ a(t′′ ). it is possible to obtain the path by twofold integration.1 Kinematics Before starting with physics. Kinematics the statics using special relativity. but ordinary Newtonian mechanics. dt i. however. the path of particles will be infinitely often differentiable. the behavior of a particle can be described (though not ex- plained). Given the acceleration. the initial conditions. In the following. The starting point for the description of a particle is the path it follows during an interval of time. so this is a well-defined quantity. t0 t0 where the quantities r(t0 ) and v(t0 ) are the position and speed at time t0 . The speed ~v of a particle is defined to be d~r(t) ~v (t) = . sometimes also called dynamics. With these definitions. if the speed or higher derivatives are given. In mechanics. In general. Thus such a limit theory. but are not central quantities. the rate of change of the position of the particle along its path. This is a pure description. the central entities of mechanics. If there are multiple particles. 2.

2 Newton’s laws 2. e.1 Preliminaries While the previous section provided the tools to describe the movement of a particle. the kinematics depend on the relative motion of observer and particle. then there is a source of change not identical to just a change of coordinates. If now the observer moves with respect to the particle. If the constant speed in zero. without forces. e. but more often nowadays just as the model. which moves. the particle will not change its kinematics. Classical mechanics is the theory which is based upon Newton’s laws. something changing the kinematics. given the insight that all models have their limits. there is no coordinate system in which the particle behaves as expected. To give this statement a more precise meaning requires the notion of an inertial system. it is necessary to introduce a few more concepts. the acceleration comes about. This history of physics will not be traced out here. They were found in a long sequence of interplay between theory and experiment.2 The first law To formulate Newton’s laws. In analogy to mathematics.2. even if it is the observer. i.. and all laws of physics have been created based on a multitude of experimental insights.Chapter 2. However. The first law of Newton is that if no forces act on a particle then there exists a co- ordinate system in which its acceleration and all higher derivatives of its path vanish identically. Thus. 2. As alluded to earlier. Therefore. no matter how. g.. e.2. If the observer would be accelerated. it did not give any reason how. i. no matter how many spectacular things later were predicted by them. it may look like the kinematics change. inertial systems are restricted to such systems which . but it should be noted that the formulation of the model is by no means a trivial exercise. and thus the particle either remains at rest or moves at constant speed. if there are forces. Newtonian mechanics 5 2. This is the question of dynamics. There is one loophole to be fixed. e. this frame is called the rest frame of the particle. I. The importance of the concept of an inertial system follows from the following idea (also called gedankenexperiment). the first step in theoretical physics is to define a basis. The first is to define a force as the origin of dynamics. it will appear to be moving. A particle can be observed. and therefore. a set of laws which describe the dynamics. these are sometimes also called axioms.

positive value. to define a new quantity. but without derivation. furthermore. In addition. It can at best explain a multitude of experimental observations with a very limited number of external inputs. This also implies that forces add like vectors. the momentum of a particle. for the sake of convenience. It is an intrinsic property of a particle.11. like the Coriolis force. If this is not the case. In classical mechanics. 2. d~r(t) p~ = m~v = m . Newton’s laws may at most move relatively with a constant speed with respect to the coordinate system of the particle. also forces are something motivated and defined by experiment. and so on for more particles. This is sometimes also called the fourth law. This is one of the essences of theoretical physics. However. This mass is an additive property.3 The second law Newton’s first law identifies what happens in the absence of forces. Note that this is indeed a definition of forces. it is possible to formulate the second law as d~p F~ = = m~a(t) (2. The definition is that a force is a vector. and the total force is the sum of all the individual forces. since this is also not something derivable. m(t) = m. dt This will make the analysis especially of mass distributions. Having these two concepts. emerge. as well as the generalization beyond classical mechanics. and it is not quantized (comes in portions). this requires first to define forces a little bit more. In this sense. version of this law is: If there are no forces. sloppy. e. this property is immutable in time. that is the mass of two particle M is just the sum of the individual particles’ masses M = m1 + m2 . The abbreviated. With this. the particle moves at constant speed (which includes zero speed). F~ (t). Newton’s second law describes what happens in the presence of forces. There is no explanation of the origin of this mass in mechanics. at least currently: It cannot explain everything in the sense of requiring nothing external. large numbers of particle with small masses. and for any given problem the value of this mass has to be determined by experiment. every particle is assigned an inertial mass m. the next step is. which is a property of said particle. which is possibly time-dependent.1) dt . simpler. This mass can have any arbitrary. This case will be discussed in section 2.6 2. not an explanation.2. so-called pseudo forces.1. i.2.

like the speed. Forces involving the speed are also often called frictional forces. in addition. but it may (and in general will) depend on the position of the particles. Though not explicitly noted. Newtonian mechanics 7 i. even higher derivatives. As said. This equation is also called the (Newtonian) equation of motion. as they usually appear in the context of friction phenomena. is is strictly speaking not a force field. this is a definition. or. In practice. but at most explicitly on time. the force changes the momentum and equals the acceleration up to a factor of the mass. This is added by the third law. if thinking of an ensemble of particles. the force is often also considered a force-field. except that it fits well with experiment.4 The third law The third law is of a substantially different nature than the two first laws. g. the second law takes the form d~p dm(t) F~ = = m(t)~a(t) + ~v (t) . This happens especially often if there is an external source of the force. not the one in terms of the acceleration or the speed..2) dt dt i. it may still be useful to think of a time-dependent mass. e. When pouring particles on a scale. (2. not only implicitly through the position of the particle as a function of time. The force may. The two first laws describe how particles are affected by the presence or absence of forces. This also elevates the momentum to be the central kinematical quantity in classical mechanics. Because the force can then be seen as a function of other vectors. e. and sometimes dependent on the speed. the majority of cases involve only forces which are position-dependent. While the first law and the third law describe only general features of systems with forces. it is this equation which actually describes the impact of forces on particles. but do not make any statement of the origin of the forces. g. 2.Chapter 2. and has usually a well-defined value for every point in space and time. e. as well as derivatives of the position. If this is the case.2. e. then the target . in principle. and actually far beyond. though the name force is still used for brevity. so there is no reason for it. the force is in general not a constant. the definition in terms of the momentum is the basic one. Its statement is that if the force on a particle emanates from another particle. Only if the force does not depend on the position. the acceleration. Though mass is an immutable concept of a particle. The equation is often also referred to as Newton’s law or as the dynamical equation. depend also explicitly on the time. the mass on the scale changes.

3) |~r1 − ~r2 |3 . the impact of the third law can be taken exactly into account. the gravitational force between them is Gmg1 mg2 (~r1 − ~r2 ) F~ = . all known forces can. An important approximation is very often that it is assumed that a force is external. rather than using them. though the back reaction occurs.8 2. which neglect all those aspects which play no role. any action induces a. for all practical purposes. i. Especially. say. the actual derivation of. The probably best known of these effective forces is the gravitational force. but it can also be shown how it becomes irrelevant for the sun being much heavier than the earth. e. and therefore the force on the particle does not change.2. and investigate.. Again. I. However. there is no reason for this at the present time. just how they act on particles. but require still the force F~ to actually describe motion. Note that still no statement is made about how any of the involved particles creates the forces. Newton’s laws do not explain the origin of the forces. |~r1 − ~r2 |. F~s = −F~t . this statement applies pairwise to each possible pairing of particles. equal in magnitude but opposite in sign. rather than a. Therefore. are too weak to make any practically measurable difference. An example for this.8. which of these forces can be deduced from other forces. (2.3 Gravity As noted before. It is the realm of fundamental physics to deduce these forces from experiment. it is so weak that. If there are more than two particles involved. There. more or less. in principle. the forces involved in standing on a floor from these elementary theories is practically far too involved and too complicated. This is then considered as an effective theory.3. 2. rather than the full forces of these theories. They are therefore sometimes called a dynamical principle. e. Given the distance between two bodies. in the realm of classical mechanics only such effective forces play a practical role. but it requires for any possibility that this balance of action and reaction is satisfied. it is much better to use effective forces. i. reaction. the origin of the force is not changed. e. since it is an axiom. be deduced from them. Though not completely covering all experimental observations. fundamental one. discussed in detail in section 2. Gravity particle acts always with the same but opposite force on the source particle. The most basic ingredients known today are general relativity and the standard model of elementary particles. will be the movement of the earth around the sun.

Newtonian mechanics 9 where G ≈ 6. ~g always points to the center of the earth and has a value. (2. experimental finding that for a particle of inertial mass m and gravitational mass mg m = mg holds. m This is an ordinary differential equation of second order. Then ~er F~ = m1 Gm2 2 = m1~g r where the so defined vector ~g is called the gravitational acceleration due to the body 2. It is a remarkable. where human activities usually take place.2) then becomes for a constant-mass particle2 1 ~ F = d2t ~r(t). g. on the surface of the earth. There is no explanation of it. 2 For brevity. . i. and therefore can be taken to be roughly constant for the couple of kilometers. The equation of motion (2. Since the radius of the earth is large. differential operators d/da will be abbreviated from now on by da . from the deepest ocean trench to the traveling altitudes of planes.4 Point particle The simplest possibility is a constant force. This feature. they are a property of a particles. and highly non-trivial. and ∂/∂a by ∂a . i. since experimentally extremely well supported. Set one of the particles fixed at the coordinate origin. is the basis for general relativity.3). of about 9.. it is again an axiom. E. e. 2.8 m/s. depending on latitude1 .4) m m 2m 1 Because the earth is not exactly spherical. F~ . It is found experimentally that the gravitational masses are always positive. also known as the equivalence principle.6741(4) m3 /(kgs) is Newton’s constant. gravitational and inertial mass are the same. They act as a so-called charge. The quantities mgi are the so-called heavy or gravitational mass. However.Chapter 2. though. There is an interesting special version of the gravitational force (2. It can be solved by integrating twice on both sides. a number obtained from measure- ment. and must be deduced by measurement. in the following no more distinction between the gravitational and the inertial mass will be made. Just like the inertial mass. they are features of particles which determine the in- fluence of a force on this particle. e. and thus one of the most basic foundations of modern physics. Z Z Z  ′  ′ ′′ 1 ~ ′ t ~ t2 ~ ~r(t) = dt dt F = dt F + ~v0 = F + ~v0 t + ~r0 . ~g changes only very slowly when moving just a little bit vertically. ~r2 = ~0.

then an arbitrary solution for this equation is given by any solution of the inhomogeneous equation to which any linear combination of the homogeneous solutions can be added. are very common in classical mechanics. situations where (2. i. Point particle where the integration constants ~v0 and ~r0 describe the position and speed of the particle at time t = 0. If. e. If there is any inhomogeneity on the right-hand side. this is no longer true. t2 ~ ~r(t) = (F1 + F~2 ) + ~v0 t + ~r0 . An interesting observation is the following. The equation of motion (2. Since the equation of motion is an ordinary second-order differential equation there are always two such initial conditions required to fully specify the solution of the equation. An interesting example is the situation with friction. and therefore this superposition principle will play an important role. If the force vanishes.2) is a linear differential equation. This is the typical behavior of the exponential function. and are the so-called initial conditions of the motion. i=1 then any linear combination of two solutions (of which there are N for an ordinary differ- ential equation of order N) again solves this equation. g. Note that this is a particularity of linear differential equations. and the solution is thus m αt ~r(t) = ~v0 e− m + ~r1 . In this sense. is of the type XN ai dit~r = ~0. m This requires that the derivative of the first derivative must again be proportional to the first derivative. 2m This is called a superposition of the two movements. for each movement separately. If a differential equation is linear and homoge- neous. However. Then F~ = αdt~r. Newton’s first law is actually just a special case of Newton’s second law. α . F~ = ~0. with or without inhomogeneous term.10 2.. e. adding two forces F~i will yield that the movement of the particle will be just the sum of the effects of the two forces.4. The reason is of mathematical nature.2) is then α d2t ~r − dt~r = 0. Because the equation is linear. though the definition of an inertial system will play a quite central role later. ~r2 would appear. the particle moves at constant velocity. where α is a friction coefficient. where the consequences of the initial conditions can be taken to be also the sum of two initial conditions.

being the primitive of F (x).6) total derivatives appear. and can be chosen at will. has been introduced. However. the derivative of the potential is the force. a deeper reason behind this arbitrariness. There is.5) It can now be recognized that both sides can be rewritten as derivatives m  Z 2 dt (dt x) = dt F (x′ )dx′ = −dt V (x). Since on both sides of (2. The appearing minus sign is a matter of convention. (2. Note that this integration constant.Chapter 2.7) i. Its relevance will be . (2. it is best to postpone a deeper discussion of it until more conceptual progress has been made. cannot be dismissed. the equation can be integrated to yield m (dt x)2 = E − V (x). Multiplying this equation by dt x. e.5 The potential An interesting concept can be found when considering the movement of a particle under an arbitrary. The force can then be obtained by F (x) = −dx V (x). Note that any integration constant in V (x) does not play a role. Either being at rest from the beginning. Newtonian mechanics 11 where the two initial conditions give again the speed and position of the particle at t = 0.8) 2 where E is an integration constant. which generates the force. which emerges from a time and not a spatial integration. but the potential is known. It will happen very often that not the force. (2. (2. since nature should be somehow uniquely determined. 2. This primitive is called the potential. as the time derivative removes it immediately. which will become evident during the lecture.6) 2 where the quantity V (x). Such a kind of arbitrariness seems to be at first quite astonishing. however. This again shows that any constant terms in V (x) do not play a role. It is best to start with a one-dimensional example first. but only position-dependent. or its movement being exponentially damped. The equation of motion then reads md2t x = F (x). force. The two solutions are now both terms. this yields m(dt x)(d2t x) = F (x)dt x.

This is called an apparent non-conservative forces. It is therefore called the work which has been done. these are taking the role of the two initial conditions of position and speed at time zero used before. Before doing this. the name potential. the force is said to be conservative. It corresponds to the difference in potential. Also. the potential for the constant force is F~ ~x. Thus. if it is possible. x1 This quantity is the necessary integrated force to move something between the two points x1 and x2 . In general. The potential discussed in more detail below. A criterion when this is possible will be developed later.5. this feature may be hidden if only an effective description not in terms of the elementary forces is performed. i as can be seen by performing the steps above for every component. It is an amazing. investing work raises the potential. In general. it is also possible to use the definite integral Zx2 W12 = dxF (x) = V (x1 ) − V (x2 ). it is possible to solve this equation by separation of variables. While so far only the indefinite integral has been used. as a brief remark the situation in more than one dimension is a little more involved. and the force becomes even in one dimension non-conservative. and actually doing some work is reducing the potential. Conversely. and not understood. this is not possible. and many examples are known in daily life. it is in general not possible to find a potential. yielding an implicit result Zx(t) dx′ t − t0 = q .12 2. (2. The amount of stored work for a particle at a position x with respect to zero potential is called the potential energy. Before continuing. In the example of the point particle in section 2.4.9) 2 x(0) m (E − V (x′ )) which after evaluation of the integral can then be solved for x(t) to get the final solution. However. feature of nature that eventually on the most basic level all experimentally established forces are conservative. The potential therefore describes the amount of work which can be done. This solution is parametrized by the constant E and the initial condition x(0). However. if the force depends explicitly on time or on speed. any potential V (~x) will create a force field   F~ (~x) = −∂xi V (~x). Thus. In general. In the general . this means that a single function must create as many different functions as there are dimensions.

i. On the other hand. it will actually be deflected there. E is called the total energy. If the extremum is a minimum. In more than one dimension this kinetic energy. E is then the potential energy. At these points. (2. if the particle at rest is only slightly disturbed. as it can also cope with the situation of a time-dependent mass. combining the potential energy V (x) and a contribution from the motion of the particle m(dt x)2 /2 = p2 /(2m) which is called the kinetic energy. it will return to the minimum. If the potential further increases to one side. a so-called inflection point of the movement. as it is a positive quantity. Note that the prediction of this behavior did not need the solution of the equation of motion. in general. the force to either side of the minimum always point in the direction of the minimum.Chapter 2. a particle set at such a position at rest will remain so indefinitely. In fact. This is called a stable equilibrium. Newtonian mechanics 13 case.8) yields m E = (dt x)2 + V (x). Thus. Thus. solving for E of (2. is given by m 2 p~2 T = (dt~r) = . (2. this will be a line integral. Thus. when E = V (x) for some point x. The second derivative may not vanish. being the derivative of the potential. Because of the general solution (2. 2 2m Once more. the constant E is entirely giving by a quantity obtained from the motion of the particle. the force points away from the maximum. and therefore the particle does not necessarily stops there. since no force acts.9). Another interesting situation arises if the potential has a minimum or a maximum at some point x. If in equation (2. often abbreviated as T . and where p is again the momentum. the sum E − V of a particle must be positive at every point in space. a particle needs to have a positive or zero kinetic energy. Thus. the force. Furthermore. Z~x2 W12 = d~xF~ (x) = V (~x1 ) − V (~x2 ).10) x1 ~ but will remain a scalar quantity. necessarily vanishes.8) the potential is set to zero. At a maximum. the definition using the momentum is the more general one. e. Any slight disturbance of the particle at rest will therefore accelerate it . and move around it if there is no dissipation.11) 2 For a particle at rest. the energy will become a very convenient tool to simplify calculations later. the kinetic energy has to vanish. however. since otherwise there is no solution.

potentials.10) implies that if a potential exists W11 = 0. Note that a potential can always be used to find a force. The potential away from the maximum. when this is possible.12) 2 If the potential depends on time or speed. There is a powerful theorem in functional analysis. C for any closed curve C.5) m  dt E = dt (dt x) + V (x) = m(dt x)(d2t x) − F (x)dt x = 0. 3 If the force is very weak around the maximum. This is thus an unstable equilibrium3 . which is much simpler to check than the integral condition. Such situations are sometimes called metastable.14 2. the acceleration away will be very slow. if the potential is not always continuously differentiable. g. To test. Again. One of the probably most fundamental statements in mechanics is that the energy is x-independent. which follows from (2. e. though there is no firm definition of when an unstable situation becomes metastable. or at least if the curve can be contracted to a point without crossing any singularities. This condition also implies ∂xi ∂xj V (~x) = ∂xj ∂xi V (~x). note that (2. in one dimension any force. this equiv- alence is only true. the energy changes with a rate of f (x. 2 (2. no matter the path. This is not so simple in more than one dimension. e. . which is time and speed independent. and thus requires that the potential is (at least) twice continuously differentiable. dt x.5. but also time-independent for conservative. but this force may not everywhere be well-defined. t). for any combination of i and j. which guarantees that this is equivalent to the requirement4 ǫijk ∂xj Fk = 0 for all i. i. has a poten- tial. if not stated otherwise. dt x. or Z d~xF~ (x) = 0. t)dt x.8). 4 Note that here and hereafter the Einstein convention is used that over any pairs of indices in a given term a sum is performed over their full range. However. it can be written as F (x) = dx V (x) + f (x. This follows from (2. if the force is continuously differentiable. t-independent. if it is maximum of higher order. e.. i.

i. In its simplest form in one dimension. g. will create such a force.Chapter 2. with ~a = 0.6 Harmonic oscillator Probably the most important problem in mechanics is the harmonic oscillator. m where the two initial conditions ~r0 and ~v0 are used to fixed the position and speed at time t = 0. but as will be seen. e. in more general cases this will be very useful. but is in fact a very important technical trick. it is best to start with the homogeneous case first. Newtonian mechanics 15 2. Such functions are given by sine and cosine. It is in this case where the usage of the exponentials becomes interesting. will turn into itself when differentiating twice. α d2t ~r(t) = − ~r. obeying Hooke’s law. this is the situation if the force is negatively proportional to the distance of the particle from the origin. (2. The associated potential is α~r2 /2. This yields a quadratic equation for a with the solutions 1  p  a± = −β ± β 2 − 4αm . respectively.. e. i. which looks at first somewhat unusual. m E. There is an alternative way to solve this equation.14) m m To solve this equation. 2m . described by the equation of motion β α d2t ~r(t) + dt~r(t) + ~r = ~a sin(σt + φ). Therefore.13). up to a negative constant. and insert it into the equation of motion. There is no advantage in this particular case. a solution is ~v0 ~r(t) = ~r0 cos(ωt) + sin(ωt) (2. A possible solution is also given by      ~r0 ~v0 iωt ~r0 ~v0 −iωt ~r(t) = ℜ + e + − e . to yield the two solutions. The solution of this equation requires two functions which. Make the ansatz exp(at). a spring. 2 2iω 2 2iω By expanding the exponentials using Gauss’ formula. it becomes evident that this is exactly the same as (2. Another interesting feature is obtained when considering the more general case of a driven accelerator with friction.13) r ω α ω = .

Harmonic oscillator The solution for the homogeneous case will then be      a−~r0 − ~v0 a+ t a+~r0 − ~v0 a− t ~r0 (t) = ℜ e + e . If the equation of motion (2. the motion grows beyond any bound. If the argument exactly vanishes. However. But this is no longer really friction. and the a± are com- plex. however. and therefore totally dominates the result. there is no oscillatory behavior. the motion becomes that of an ordinary har- monic oscillator. this corresponds to a superposition of an exponential damping. while a− an exponentially decreasing one. as it provides energy to the system. If β 2 < 4αm. after an initial stage. The important addition is the driving term d. (mσ − α) + β σ which exhibits again the structure of being the solution to the homogeneous equation plus an explicit solution of the inhomogeneous equation. Thus. close to this value. what happens is that the external force applies just exactly such that it always accelerates the movement a little bit further. or just a resonances. this is here an educated guess. though with a frequency depending on the interplay of damping and the force and a exponentially decreasing amplitude. the movement will be exponentially damped. In this case. This is called a resonant behavior. If β is positive. the amplitude grows beyond any bounds.6.16 2. This happens. the solution a+ will again yield an exponentially increasing motion. Thus. If β = 0. the movement becomes exponentially increasing.14) ~a no longer vanishes. the result is ~ + d(0) ~r(t) = ~r0 (t) − d(t) ~ ~ m  d(t) = ~a 2 2 2 2 βσ cos(tσ + φ) + (mσ 2 − α) sin(tσ + φ) . and the denominator vanishes. Thus. Thus. the argument of the squareroot becomes negative. If the argument is positive. a− − a+ a+ − a− The resulting path then depends on the relative sizes of the involved constants. If β is negative. and an oscillation. . which one will eventually win. as it has the same form as the driving term. It depends on the initial conditions. Note that the friction force can exponentially increase the motion. the time-dependence can be factored out. such a term destabilizes the motion. instead of damping it occasionally.~ since for ~r0 just the previously stated behavior emerges again. Physically. but in the end. This can be seen to be the necessary inhomogeneous part for the solution by entering it into the solution. depending on the value of a± . There are three distinct cases. It is a reasonable ansatz. and degenerate. if the friction term enhances motion. this term diverges if mσ 2 = α.

. if also ~r0 shows an oscillatory behavior. If β is zero. since gravity. Furthermore  0 = ǫijk drj Fk = ǫijk (drj f )rk + f δjk = ǫijk rk (drj f ) 5 Sometimes also the value is determined when it drops to 1/e or to ln 2. especially in astrophysics. (2. it actually jumps from zero to −π at the resonant frequency. (2. 2. . i.. e.3). Similar (damped) resonances are a feature exhibited by many physical system. It is trivially zero for α = 0. are called central forces. depending on the context. forces on particles of the type F~ = f (~r. These are then different numbers than for the factor 2. but still a strong enhancement at mσ 2 = α. from celestial mechanics to particle physics. which are prototypical. having the same direction as the vector ~r. The value of this phase shift depends on the value of α when all other quantities are fixed. and that the direction of the force is towards this center The further crucial assumption is that there are no lateral forces. The first condition is that the function f is at most depending on ~r. so that the prefactor only involves the distance. This width is 2(β 2σ 2 + 2m2 σ 4 )1/2 . Note that φ only acts as phase. the width is defined as the amount of deviation of α from the critical value α = mσ 2 where the function drops to a half5 . Their structure also entail many further consequences.. and tends to −π for α to infinity. This is called a damped resonance. it modifies only to some extent the relative behavior of the two oscillations.Chapter 2. These forces assume that the source of the force is at the center of the coordinate system. As a first step. t)~r. but delayed by a so-called phase shift γ. and this jump becomes smoothed out the larger β becomes. is the probably best known example. More interesting is that the maximal amplitude is reached no longer when ~r0 reaches it maximum. it is useful to find the condition under which there is a potential for these forces. to avoid any dissipation. It can be shown to be negative for all values of α. Newtonian mechanics 17 If β 6= 0. becoming −π/2 at the resonant frequency. Considering the prefactor as a function of α. They shall therefore be treated in some detail. e. there is no divergence.15) i. Under- standing it well for the harmonic oscillator is therefore understanding a prototype for a multitude of physical phenomena.7 Central potential Generically. These forces are probably the most important type. as then all oscillation is just given by the driving force. dt~r..

(2. where r = |~r| for brevity. and the angular momentum is conserved. e. though it is always constant. This is the reason why it will be so relevant in this section. the actual value of the angular momentum is not intrinsic to the system. Geometrically.17) 2m The right hand side has dimension area per time. (2. If there is no torque. . which will only happen if f depends only on |~r|. the planets fulfill this so-called area theorem. but can be.18 2. this area is constant. Thus.2).16) The so defined quantity ~l is called the angular momentum. 2. a conservative central force must be of the form f (r)~r. which can be derived from its magnitude 1 ~ |l| = |ǫijk ri dt rj ek |. But the position vector and the speed need not be parallel. as its time derivative vanishes. rj drj f (|~r|) = dx f (x)drj |~r| = dx f (x) . because either the force itself vanishes or has a vanishing vector product with the position. g. and form a vector product with ~r. whose time evolution is deter- mined by the torque M. but depends on the coordinate system. (M (2. angular momentum. Take the equation of motion (2.1 Angular momentum For the following. ~ This is called the angular momentum law. so does the torque. depending on the coordinate system. |~r| Thus. This can be seen from the free particle. the torque always vanishes. For a central force. As will be seen. A useful first consequence is the geometrical interpretation of the angular momentum. Since the force vanishes. However.7.15). the angular momentum is conserved.7. Central potential This condition can only be fulfilled. If angular momentum is conserved. it is useful to first introduce a further concept. is physical. The rate of change. however. it is the area which the vector pointing from the origin to the particle covers per unit time. yielding ~ )i = ǫijk rj Fk = mǫijk rj d2t rk = dt (mǫijk rj dt rk ) = dt (ǫijk rj pk ) = dt li . as it is uniquely given by the torque. if the derivative is proportional to rj . a statement about the value of the angular momentum requires to also provide the coordinate system.

2 Effective potential For a conservative central potential both the energy (2. i. it is possible to restrict the description to a plane.5. They come about as Newton’s law is originally a second order differential equation. (2. thus x = r cos φ and y = r sin φ.19) m as a consequence of the conservation of the angular momentum. Thus. This can be used to simplify the solution of the equations of motion. Its angular position is not relevant.5.7. (2. since ~r~l = dt~r~l = ~0. Thus. Due to the radial symmetry it is convenient to chose angular coordinates for the description.Chapter 2. potential Ve .18) 2 2mr 2 This result has a number of interesting implications.16). rather than of the second kind. Secondly.20) 2 r(0) m (E − Ve (r)) Zr(φ) ldr ′ φ − φ0 = p . . a distance from the origin r (on which the potential depends) and an angle φ. because the angular momentum is conserved the speed and the position are coplanar in a temporally constant plane. First. the situation is analogous for the one-dimensional par- ticle of section 2. This implies m m E = (dt~r)2 + V (r) = ((dt r)2 + r 2 (dt φ)2 ) + V (r) 2 2 2 m l m = (dt r)2 + 2 + V (r) = (dt r)2 + Ve (r). and especially its length l = |~l|.21) r ′ 2 2m(E − Ve (r ′ )) r(0) yielding the (implicit) solution. e. as a function of two more integration constants. the so-called effective. rather than the original potential V . the full solution are now obtained from two differential equations of the first kind. the distance from the origin. i. Newtonian mechanics 19 2.11) and the angular momentum (2. but with the modified. are conserved. both equations can be integrated to yield Zr(t) dr ′ t − t0 = q (2. the energy is entirely deter- mined by the radius r as a function of time. e. The angular motion can be determined from lr 2 dt φ(t) = . since the cross product is perpendicular to its components. In analogy to section 2. (2. First. r0 and φ0 .

or an atomic nucleus and an electron. where it takes the form q1 q2 V (r) = −ǫ . mM V (r) = −γ . a uniform motion in the z-direction can be superimposed with this solution. true if the two bodies are the sun and the earth. The other one is Newton’s law of gravity. Central potential and for two variables thus needs four integration constants. though this is called Newton’s law of gravity. r has to be replaced by |~r1 − ~r2 |.2. Of course. . The first three laws are about inertial masses. r0 and φ0 . only an experimental result that these two masses are identical. This is e. One is the electrostatic force. this will be postponed until later. This law is about the gravitational masses. in this particular case. l. since this only obscures a few things right now. It is. as noted above. this becomes a two-body problem which. and γ is again a system-of-units-dependent constant. r and describes the potential between two bodies of masses m and M. If this is not true. there is a-priori no casual connection between this law and Newton’ three laws (except that they have been discovered by the same person. This is an excellent approximation if one of the bodies is much heavier than the others. Newton). can actually be solved only with slightly more effort. and the approximation will be made. It should be noted that. In both cases it is tacitly assumed that one of the bodies resides at the center of the coordinate system. and this is the potential between two electric charges having electric charges q1 and q2 respectively. 2. quite interesting to study the case of V (r) ∼ 1/r in more detail. if wished. This will be done in section 2.7. It is.8.7. This particular case if of special importance as there are two situations in which it arises. Otherwise. called Newton’s constant. but equally useful.7. it is useful to take the opportunity to demonstrate also a different. way of solving the problem: Another replacement of variables. g.2. a satellite and the earth. In the following it will be furthermore assumed that the body at the origin will not move. Though this problem can be solved with the methods of the previous section 2. being here E. the distance between both bodies. However.20 2. r where ǫ is a constant depending on the system of units chosen.3 Planetary motion This previous result is fully general. however.

A convenient choice is to select α = 0. l2 where the constants α and β have to be fixed by initial conditions. which is far simpler in general.22) l2 This shows the motivation for this approach. the derivative of V with respect to s is just a constant. (2. if β > 0. Newtonian mechanics 21 To start out. Differentiating this equation a second time with respect to φ yields  2  l2 2  l 2  0= 2(dφ s)(dφ s) + 2sdφ s + (ds V (s))(dφ s) = dφ s 2(dφ s) + 2s + ds V (s)) . born from hindsight and experience. The solution is then given by 1 γm2 M 1 s = = β cos φ + 2 = (1 + ǫ cos φ) r l k l2 k = γMm2 ǫ = βk .Chapter 2.6 by φ.22) is already known. Though s is a function of the time t. 2m Though it may appear tempting to directly solve this equation. i.7. Inserting this into (2. Then the chain rule yields ds d 1 dt dt r mr 2 = r =− 2 dφ dt dφ r l where (2. Thus. The term in parentheses is also a linear differential equation of second kind l2  2(d2φ s) + 2s + ds V (s).2.18) yields l  E= (dφ s)2 + s2 + V (s). 2m Furthermore. Out of a non-linear differential equation the task has been reduced to the task of solving two linear differential equations. e. But equation (2. the solution can be read off directly by replacing t in section 2. r is constant as a function of φ. It is one particular version of the harmonic oscillator equation of section 2. and therefore the equation is γm2 M d2φ s + s = . define s = 1/r. This is a perfect circular orbit.19) has been used. and yields γm2 M s(φ) = α sin φ + β cos φ + . which implies that the minimal value of r is at φ = 0. it is in this case better to take a detour.6. it is also possible to consider it rather as a function of the angle φ in the same two-dimensional coordinate systems as in section 2. 2m 2m This equation has one solution with ds/dφ = 0.

the quality of the solution is entirely characterized by ǫ. it would have been possible to determine many qualitative features of the particle’s motion just by studying (2. These are just cone cuts. the potential would have taken the value V (r) = r. It can therefore no longer be an ellipsoid. At the same time. Since in general r0 6= r1 . the motion is periodic. In the whole discussion. If ǫ > 1. Thus. r0 = r1 . the equation has no solution also for r larger than some value r1 . since now s can vanish for φ = π. or greater than 1 this describes an ellipsoid. but is deflected before it can reach it. Since the kinetic energy is always positive. This can be seen from the fact that (1 + ǫ cos φ) remains bounded. Central potential where the last rewriting is convenient to characterize the geometric properties of the so- lution.18). and the path of the particle becomes a hyperbola: The particle tries to approach the central one. and the motion becomes the circular orbit. and for ǫ smaller. This implies that not all φ values are allowed. especially zero. If E is exactly zero this is the smallest possible value for which it is still possible to have an infinite distance. Still. and hence so does r. If the time-dependence . which vanishes for l = 0. the movement is therefore bounded. 2mr r If the energy is greater than zero. Note that the case β = 0 returns exactly the circle case. φ can still reach all values. If the energy becomes negative. Thus. This implies that a particle with positive energy corresponds to the hyperbola. This corresponds again to the closed orbits. then the motion is no longer periodic. Therefore it is said that the angular momentum creates an angular momentum barrier. It is therefore corresponding to the parabola. it follows from (2. Thus. and a hyperboloid. a paraboloid. g. The point of closest approach is given by r = k/(1 + ǫ). this equation has a solution for all r greater than some limiting r0 . It is quite instructive to return to the energy of the particle. then there is no longer a positive solution for r for all values of φ.22 2. If. It is an ellipsoid of which one of the main axes goes to infinity. the time-dependence was not an issue. ǫ is also called the eccentricity of the orbit. even without solving the system. equal. a similar study would immediately yield that the particle can never move to r = ∞. If ǫ = 1.7.18) l2 γmM E≥ 2 − . and therefore the movement is a parabola. Such studies therefore are an extremely important tool. the orbit is an ellipsoid.. A non-vanishing angular momentum increases the point of closest approach r0 . The case of ǫ < 1 (with ǫ = 0 being the circle) is just the ordinary case of a planetary orbit around the sun. For the smallest possible value of the energy. e.

bounded. using (2. Kepler’s third law states that the ratio between the time needed for a complete orbit T squared and the third power of the larger axis is constant. the actual solution is rather involved. as it appears quite often.19). Since the heavier mass is at the origin the planets move along ellipsoids with the sun at one of the focus points. X mi d2t ~ri = F~i + F~ij . Newtonian mechanics 23 is actually of relevance. when looking at the total solar system. . e. this may be the gravitational pull of the galaxy as a whole. by gravitation.8 Mass distributions and center of mass 2. Of course. which involve non-separable function of more than two coordinates. must obey F~ij = −F~ji .Chapter 2. However. There can be also external forces on the individual particles F~i . g. Tl πab = 2m But the value of b is just given by the minimal value of s at π/2.17) is actually already Kepler’s second law if angular momentum is conserved. The area of the ellipse is. j Consider now the sum of all N equations of motion. Because of Newton’s third law the force F~ij with which particle i acts on particle j. Thus follows T2 4π 2 = . The relation (2. g. i. X X mi d2t ~ri = F~i .23) i 6 There may also exist genuine n-body forces. e. 2. It appears that this would require to have the time dependence. where d is the number of dimensions6 . It is useful to discuss this situation in some detail. There are thus dN equations of motion. it can be obtained from (2. as it is here. and thus for simplicity F~ii = ~0. a particle does not act on itself. This only complicates the remainder unnecessarily. For the solar system. a3 γM which completes the proof.1 General properties The previously discussed situation with two particles is a special case of having N particles. Still. from the results the famous laws of Kepler can be deduced for planetary. (2. Let each of the particles have its own mass mi .8. motion. but this is not the case. but can appear in practice.17). e.

The center-of-mass is then ~ = m1~r1 + m2~r2 . Therefore. where it useful to define separately the potential between two particles and of the external forces separately. This implies that the potentials are superimposed like the forces. i.23) then reads ~ = f~.8. e. there are two particles only. i. the center of mass ~ moves like a single particle having the total mass M only under the influence of the R external forces. e. i The equation (2.2 Two-particle systems Of particular practical importance is the situation when N = 2. Especially. X X P~ = mi dt~ri = ~pi X X X ~l = mi ǫijk ri dt rj ~ek = ǫijk Ri Pj ~ek + ~+ ǫijk (ri − Ri )(pj − Pj )~ek = L ~li i where L~ is the angular momentum of the center of mass while the ~li are the relative angular momentum with respect to the center of mass. R m1 + m2 . It does not matter how involved the internal forces are for this. Furthermore. Mass distributions and center of mass and thus the internal forces do not appear. for a two-particle potential it holds that F~jk = ~ei d∆~ri Vjk (∆~r) ∆~r = ~rj − ~rk and thus the forces are obtained by deriving with respect to the connecting vector. they do not appear. It may also be useful to define analogously a total momentum and angular momentum.8. Md2t R which looks just like a single-particle equation of motion.24 2. Define furthermore X M = mi i ~ = 1 X R mi~ri M i X f~ = F~i . 2. the total potential of a mass particle is the sum of all potentials acting on it. if all forces are conservative then in analogy to the single particle a potential energy can be defined.

m1 + m2 In terms of this mass the equation of motion becomes F~1 F~2 F~12 d2t ~r = − + . the equation of motion for the relative coordinate no longer references the external system.Chapter 2. Thus.24) M m ~ − 1 ~r. it can be shown that also the relative angular momentum takes the form ~l = µ~r × dt~r. Thus.7. what is called a closed system. Since the planet (or satellite) is so much lighter than the sun (the planet). and the result is automatically the correct one for the two-body problem. Of course. but this requires to solve only the one-particle problem. with the formulation developed here. Furthermore. Thus. Newtonian mechanics 25 The vector ~r = ~r1 − ~r2 gives the relative distance between both particles. the equation of motion for the relative coordinate is essentially the one in the case of the fixed heavier particle. In particular.3 becomes more transparent. ~ + m2 ~r ~r1 = R (2. By direct comparison. the equation of motion for the relative coordinate is the same as for a single particle with the reduced mass. The equation of motion for the relative coordinate is F~1 F~2 F~12 F~21 d2t ~r = − + − . to determine the actual movement of both bodies. in this formulation the approximation made in section 2. However.7. this is much simpler: It now only necessary to substitute in all results of section 2. the reduced mass is essentially that of the planet (the satellite). m1 m2 m1 m2 The dependence on the masses suggests to define the so-called reduced mass m1 m2 µ= .25). it is possible. If one particle is much heavier than the other the reduced mass is essentially identical to the mass of the lighter particle.25) M The center-of-mass movement is as before determined by the external forces only. both bodies will perform exactly the same type of movement. m1 m2 µ If the external forces vanish. It is now possible to rewrite the position of the particles in terms of relative coordinates and the center-of-mass coordinates only. the substitution can be done for either of the bodies. Thus. ~r2 = R (2. using equations (2.24-2. for a bounded planetary .3 the mass of the planet by the reduced mass.

8. This can happen in either of two cases. but its details too involved to be included. The three(and more)-body problem cannot be reduced to a one-body problem. the force on the sun is so tiny that its movement is all but negligible. see section 2.9. What fast enough means is rather context-dependent.3 Scattering A situation which is of high relevance in many physical application is a closed system such that the two-body potential (or a force) is short-range. .2. However. In such a situation it is possible to define a scattering process in the following way. 2. On the other hand. body problem. It is unfortunate insight that a similar reduction of complexity is not possible if there is more than two objects involved.4: The reduced mass is so close to the earth’s mass that setting it equal to the earth’s mass does not create an appreciable error. This is also the way how Newton’s third law seems to have no impact on the source of the force. They are then send towards each other in such a way as their center of mass is at rest7 . which are necessarily fulfilled. They afterwards escape again to infinity.26 2. E. but even maybe some kind of power-law dependence 1/|∆~r| is sometimes sufficient. if the interactions are conservative. where the value ~r0 belongs to the definition of the potential. Start out with two particles. Mass distributions and center of mass motion both celestial bodies will have an orbit around the same point.8. and stop interacting. the movement in the solar system or of the solar system around the center of the galaxy is such a three. which is the center of mass. where the backreaction is known to be small. a situation arising quite often in practice. there are a number of conser- vation laws. but a full solution is required. Thus. This also justifies to make this approximation in more involved cases. while in the closed system Newton’s third law holds. which are initially so far separated that they are not interacting. in these cases it is possible to simplify the problem. exp(−|∆~r|/r0 ). No matter how the details of this kind of interaction is. One possibility is that the potential has a sharp cut-off. or more. V (∆~r) θ(|∆~r − ~r0 |). but very often requires at least an exponential decay at large distances. g. as announced in section 2. which is 7 This is not necessary but makes the following calculations much simpler. Superimposed to the orbital motion of both bodies can then be a movement of the center of mass. The other is that it decays fast enough. where the scale r0 is once more characteristic for the potential. its violation by setting the external force constant and immutable is a very good approximation.

the linear momentum has to be conserved. Especially. it follows that ′ ′ p~2 ~p2 p~ 2 ~p 2 T = 1 + 2 = 1 + 2 = T ′. and can therefore be selected at will. a so-called inelastic reaction. the so-called scattering angle. leaving only one component. the system is no longer really closed. also be influenced by the initial momenta and the masses. Furthermore. Since in the beginning there is no potential energy. all movement is contained in a plane. The final state of the system is then no longer entirely determined by the scattering angle θ. This can be traded in for the relative angle between the incoming and outgoing momentum. Hence. this implies that the direction is arbitrary. Again. p~1 + ~p2 = p~′1 + p~′2 . While the above describes accurately the situation for the scattering of two particles.Chapter 2. and the coordinate system can be chosen such that all components in the third direction can be chosen to be zero. . By construction. ~p1 + ~p2 = ~0. This scattering angle may. Especially. Still p~′1 = −~p′2 and the momentum conservation holds trivially in the center-of-mass system. the only freedom left by the kinematics is the relative angle between one of the incoming and one of the outgoing momenta. Therefore. In this case. both quantities may depend on the properties of the initial state. Though such a situation can be realized. Newtonian mechanics 27 called elastic scattering. the energy balance is T = T ′ + Q. however. where the primes denote the situation after the scattering. These conditions enforce that the incoming and outgoing momenta both lie along respective lines. where the energy transfer to or from the potential Q can be either negative or positive. Thus. Thus. ~p21 = p~22 = p~12 = p~22 . all its consequences are encoded in this one angle. Dropping this condition of elasticity allows to transmute energy and momentum in the potential to an energy loss or gain. there remain only two components for the final momenta free. no matter how complicated the potential in the interaction area. This also implies that. θ. it is very often interesting. but also by Q. the initial relative momentum is zero. One is constrained by the energy conservation. it is often useful to make more statistical statements. Furthermore the energy is conserved. 2m1 2m2 2m1 2m2 ′ ′ Since ~p1 = −~p2 . as energy and momentum are no longer conserved quantities. the final momentum must also be zero.

Because of the symmetry of the problem.7. sin θ dθ So far. effect to reduce the energy of a particle coming from infinity such that it becomes bounded. This can be rewritten in terms of the cross section.8. is in a plane. 9 This determinism is lacking in quantum physics. it was not necessary to specify the detailed form of the central potential. consider the situation of the gravitational potential of section 2. Using the form −α/r for the potential yields α θ b(θ) = cot . and the cross section will only depend on the azimuth angle θ. Thus. Mass distributions and center of mass how many particles are scattered into which direction per unit of incoming particles. as it is just one of the non-closed trajectories. This quantity is called the (differential) cross section and is defined as Particles going into dΩ σ(Ω)dΩ = . in three dimensions e. for this problem particularly suited. In terms of the impact parameters and the number of incident particles equation (2.7. there will be no dependence on the angle φ. It is useful to define the impact parameter b as the point of closest approach to the center of the potential. As an example. Detecting such probabilities belong to the fundamental experiments with which quantum effects have been detected. . It is connected to the angular momentum by √ l = b 2mE = mv0 b.3. dissipative. where N is the total number of incident particles. where v0 is the speed of the incident particle at infinity. as for the bounded cases no particles are incident or going away8 . and thus by their respective impact parameter. only the non-closed trajectories are interesting.26) now reads 2πNbdb = −2πσN sin θdθ. (2. spherical coordinates. the scattering. Integrating the differential cross-section on dΩ yield the total cross section. by suitably choosing the coordinate frame. where the actual direction of a single particle is given by a probability. yielding b db σ=− . Of course. The number of particles N which are then scattered into an azimuth angle of θ are uniquely9 determined by the trajectories of section 2. sin θdθdφ in the.28 2. g.3. 2E 2 8 Gravitational capture requires thus another.26) Total incident particles where dΩ is the solid angle.

the complete quantum calcula- tion for this potential leads to the same result. but is otherwise not present. but this is due to the special structure of the potential. Newtonian mechanics 29 and thus 1  α 2 1 σ(θ) = . and thus in forward direction. i. there is one worrisome feature of this result: At small angles. deviations will appear already at slightly large than zero angle. At first sight.4 Continuous distribution So far the situation has been treated that the system is made up of a finite number of individual particles. it provides a good description at larger angles. 16 E sin4 θ2 This is Rutherford’s scattering cross section. The mass concentrated in some fixed volume V is then given by Z MV = d3~rρ(r). e. and the approximation of a point-like gravity origin breaks down.8. because the sun also has a finite extension. a continuous body. In reality. 2. outside the body this density will be zero. . Incidentally. Rutherford’s formula should not be used at small scattering angles. However. and not true in general. Therefore.Chapter 2. it diverges.1 can then be readily generalized to yield Z M = d3~rρ(r) Z ~ = R d3~r~rρ(r) Z P~ = d3~r(dt~r)ρ(r).8. The quantities from section 2. This is an artifact of the approximation that the sun (or the gravitational center) only produces the gravitational field. this can actually be generalized to the case where the number of particles becomes infinite. and therefore resolve the structure of it beyond the existence of the gravitational field. However. a function which determines the amount of mass per unit volume at every point in space. Of course. V Replacing the finite volume by the whole of space provides the total properties of the body. and the system becomes a mass distribution. the particle would collide with the sun. Such a continuous body is characterized by a density ρ(r). Of course.

Other quantities are obtained similarly. Most features of continuous mass distributions are therefore just a straightforward generalization of a system of point par- ticles. the axis should be going through the center of mass of the body. around a fixed axis.5 Moment of inertia There is one aspect where the generalization to a mass distribution leads to an interesting new concept. re- spectively10 .30 2. The total kinetic energy is then given by Z 1 1 T = d3~rρ(r)|dt~r|2 = Jω 2 2 2 Z .8. where the coordinate system has been chosen to be such that the rotation axis coincides with the z-axis. Consider a body which rotates.8. Furthermore. but for the sake of simplicity does not translate. In this case the angular velocity is constant and given by ~ω = ω~ez . 2. the center of mass and the total momentum of the body. and thus lies entirely in the x-y plane. The speed of rotation at some point ~r inside the body is then given geometrically by dt~r = ǫijk ωi~rj ~ek . Mass distributions and center of mass yielding the total mass.


2 3 .

~ωi .

J = d ~rρ(~r) .

ǫijk ~rj ~ek .


. .

for a homogeneous cylinder of radius R. E. the moment of inertia would be 1 1 J = MR2 + Mh2 .27) ω where the quantity J is called the moment of inertia. 4 2 0 0 0 where in the last step the mass of the cylinder has been used. the last formula needs to be generalized and can ~ be obtained from dt R. if the cylinder would rotate around an axis orthogonal through its symmetry axis but going through its center of mass. This moment of inertia is a quantity which characterizes the reaction of a body to a rotation around a given axis. Note that its definition is also valid if the z axis is not the rotation axis. g. height h. and density ρ0 rotating around its symmetry axis it is Z ZR Z2π Zh 2πhR4 1 J= d3~rρ(r)(x2 + y 2 ) = ρ0 dρ dφ dzρ3 = ρ0 = MR2 .. (2. . For comparison. 4 12 10 If the mass distribution would be time-dependent.

the positions appearing in (2.Chapter 2. it was assumed that the rotation axis was going through the center of mass. In this case. where R ~ is the (orthogonal) displacement of the center of mass with respect to the axis of rotation and ~s is the distance to the center of mass. The consequences of this is described the theorem of Steiner. So far. This yields Z . Newtonian mechanics 31 Thus. Rewriting this position as ~r = R ~ + ~s.27) are with respect to the actual rotation axis. the amount of kinetic energy in the rotation of an extended body is influenced not only be the mass distribution. This will not be the case in general. but also by the relative alignment of the rotation axis with respect to the body.


2 .

~ωi ~ − ~s)j ~ek .

= J + M (~ω × R) .

3 d ~sρ(~s) .


ω .ǫijk (R ~ 2.

If this is the case. An interesting situation occurs if a potential V (x) has an extremum at some point x0 . The fourth term vanishes since R~~ ω = 0 by construction. the situation is considered if the particle is somehow not at x0 . but rather at some place close to x0 . e. yielding11  V (x) ≈ V (x0 ) + dx V (x)|x=x0 (x − x0 ) + d2x V (x)|x=x0 (x − x0 )2 + O (x − x0 )3 . One describes the rotation of the body as a whole. things become highly non-trivial. . Thus remains the higher-order terms. ω 2 where it has been used that   ~ 2 ~ 2 2 ~ 2 ~ |(R − ~s) × ~ω | = (R × ~ω ) + (~s × ~ω ) − 2 (R~s)ω − (~ω~s)(R~ω ) . and all terms in the 11 If the potential is non-analytic in x0 .the vector ~s is measured with respect to the center of mass. and the other component the rotation of the body in itself. According to (2. the two movements separate. x − x0 is small. the actual moment of inertia has two components. The third term vanishes upon integration. since by construction this is the position of the center of mass for the center of mass being at the displacement position . In the next step. I. since by construction the potential has an extremum. Expand now the potential around this minimum in a Taylor series.7). The second term vanishes. 2.9 Perturbation theory Consider again the one-dimensional case. This will not be considered for now. Thus. The first term is just a constant and does not influence the force and hence the motion of the particle. again. the force then vanishes.

according to section 2. But such a potential is just the one of the harmonic oscillator in section 2. i. Thus in a potential which has a positive second derivative. i. and therefore the development is much slower. Placing it at x0 without initial velocity will actually leave it at rest. Thus. The resulting movement is known and depends on the sign of d2x V (x)|x=x0 . this will only modify the initial conditions of the movement of the particle to this order. or moves freely if it has a non-zero initial speed. than any given order in the perturbative series. return to the equilibrium position. this situation is called a metastable equilibrium. There remains the situation if the extremum is actually a saddle point. any movement will not be exponential. the potential is constant. a particle close to the minimum will perform harmonic oscillations around the minimum. if just the initial x is close enough to x0 . Such an equilibrium position is therefore called unstable. If the speed is so large that the kinetic energy is of a similar size. it will move somewhere.6. If it is unstable. The next order does not change it.9. it will move away from the equilibrium position. If the equilibrium is stable. If the second derivative is actually negative. If it is metastable. If the sign is positive. this implies that a particle which is positioned at the equilibrium position of a potential will react to any infinitesimal external force in one of three ways. the second derivative vanishes. only the quadratic term contributes. and higher orders have to be taken into account. e. however. e. if there should appear any damping of the movement it will be returning to and staying at the equilibrium position eventually. The reason is that the full . However. the movement is no longer fully described by this order. exponentially away from the position x0 . and thus a minimum. Similarly. The ultimate fate of the particle is then determined by higher orders in perturbation theory. the motion of the particle to this order in perturbation theory will be an oscillation around the minimum. If the kinetic energy is small.6. In the next order. and the movement can be analyzed in perturbation theory. as the second term in the Taylor series vanishes as discussed before. In this consideration the initial speed has not been an issue. the particle with move. Perturbation theory Taylor series becoming increasingly smaller. Especially. or larger. It is thus possible to consider the particle to only be slightly perturbed. but will eventually be influenced by higher orders of the perturbative series. if damping is present. and therefore the particle remains at the position x. by discussing the effects of the terms in the Taylor series order by order. Then again the higher orders will determine the fate of the particle. To zeroth order. Thus the equilibrium position x0 is called stable. it will oscillate around it or. Higher orders in the perturbative series will not alter this behavior qualitatively.32 2. and any small movement outside the minimum will only yield a small oscillation.

then it is possible to define coordinate systems. Two coor- dinate systems shall have coordinates ~r and R. Newtonian mechanics 33 potential has possibly not the infinite rising barrier of the harmonic oscillator. and ms . Especially.10 Galileo group The concept of an inertial system was of central importance in formulating Newton’s laws in section 2. . some which move at constant speed with respect to the first one. the corrections are small since they are suppressed by powers in mi /ms . There are also other inertial systems. Then the kinetic energy is sufficient to escape this potential well (or feel the distortion away from harmonic). also the particle moves at fixed speed. It is worthwhile to reinvestigating this concept a little more. Thus. The basic starting point is to require that there are coordinate systems in which New- ton’s first law is valid. but may show a different behavior in a different direction. e. ~ What will be assumed is that the time in . and thus a description to this order of perturbation theory makes no sense. Then the above said has to be weighted with the direction of the external displacement or perturbation. If the potential is stable/metastable/unstable in this direction. as well as their respective masses m1 .2. g. Thus with the positions ~r1 . ~r2 of the two planets and ~rs of the sun. and it therefore moves with constant speed. an equilibrium position may have different characteristics in different directions. the same is true for the movement of the solar system around the milky way. the dominating part is a two-body problem. for the case of two planets around the sun. the gravitational interaction between the two planets is small compared to the sun. These concepts can be equally well applied to the situation with multiple extrema. m2 . E. 2. and the remainder can be treated in a perturbative fashion rather accurately. pertur- bation theory can also be used if the potential can be Taylor expanded in a different way.Chapter 2. assume that there is one coordinate system in which there are no forces acting on a particle. it is useful to work in the language of linear algebra. but may flatten out. In those. g. the particle will act accordingly. which a not inertial systems. Besides the possibility to apply perturbation theory for equilibrium positions. a subtlety arises. |~r1 − ~rs | ms ms that is. In this case. by having them being accelerated with comparison to the first. When generalizing to more dimensions. it is possible to write the potential for one of the planets as   m1 Ms m1 m2 V1 (r) = γ +O . To systematize this. If this is the case.

Thus. and hence Λ = 1. transforms into each other trivially t = T. Also. In the other coordinate system. and the difference between both coordinate systems is entirely encoded in the translation ~r0 (t) and the matrix-valued rotation Λ(t) ~ + 2(dt Λ(t))dt R d2t r(t) = d2t ~r0 (t) + Λ(t)d2t R(t) ~ + (d2 Λ(t))R ~ (2. Then if for a particle of mass m md2t ~r = 0 is valid.34 2. the explicit form may differ.30) ~r0 = ~r01 + ~r02 . t and T . dt = dT . and all coordinate system reachable by Galileo transformations. However. the most general possibility of transformations between two inertial systems is ~ ~r(t) = ~v0 t + ~r0 + R(t). . however. A time-independent rotation. (2. parametrized by the two vectors ~v0 and ~r0 . Start with two inertial systems. If two consecutive Galileo transformation are performed. up to a rotation. and thus t the two coordinate systems cannot be accelerated with respect to each other. (2. is not different from an ordinary rotation of coordinate systems in linear algebra. it will move as ~ ~r(t) = ~r0 (t) + Λ(t)R(t). Galileo group both frames.10.28). Note that the forces on a particle are due to (2. This group can also be extended by including time-independent rotations and reflec- tions. they are equivalent to a single Galileo transformation with ~v0 = ~v01 + ~v02 (2. which will be lifted when introducing special relativity in chapter 3.31) and (2. this implies that the rotation Λ(t) cannot be time-dependent. but the force are still equal up to rotations and reflections. especially if friction is involved. then so must be mdt R~ = 0.28) Thus also the time differentials are the same. It is this part. the so-called Galileo group. as otherwise there will be some paths looking like being accelerated. ~ Given the path R(t) in one coordinate system. This is called a Galileo transformation. Galileo transformations therefore form an (Abelian) group.29) t the requirement of both being inertial systems d2t ~r = d2t R~ = ~0 implies d2~r0 (t) = 0. In this case no longer an equality of the forces is true.2) then identical in all inertial systems. and will therefore not be considered for now.

or by the forces necessary to modify the path of the particle.2) and (2. When considering a force- free particle.. 2.32) The first term is the force in the other coordinate system. Com- bining (2. they can be measured. which is essentially given by (2. g. Newtonian mechanics 35 2. To describe the effect. The second vector. Though small in this case. the particle experiences an additional force due to this acceleration. Nonetheless. They also create pseudo forces. and may be bend.11. it will move along a straight line in its inertial system.11. To do so. These appear.1 Pseudo forces And interesting situation arises when the condition of inertial systems is relaxed. In an accelerated frame. ~ with time-dependent coordinates and time-independent unit vectors. e. The origin of these forces can also be understood differently. The other two terms come from a time-dependent rotation of the coordinate systems with respect to each other. up to a (time-dependent) rotation. including rotations of the body itself. It is therefore a relative acceleration.29) yields for a fixed-mass particle   F~ (t) = m Λ(t)d2t R(t) ~ + d2~r0 (t) + 2(dt Λ(t))dt R t ~ + (d2 Λ(t))R t ~ . It is therefore called a pseudo force. and therefore the same as with transformations between inertial frames. The three other terms are different. ~r is the distance to the center of mass in this coordinate . The first term is called the Coriolis force and the second one the centrifugal force.32). this path is no longer of constant speed. As a consequence.11 Rotating bodies revisited 2.2 Tensor of inertia The same approach can be used to discuss the rotation of an extended body at given angular velocity ω(t). Thus. this can either be done by the coordinate transformation. Sometimes these are also called inertial forces. split the coordinates of every element of the body in two parts. it must be included when describing the movement of the particle inside an accelerated coordinate system. R(t). The first term corresponds to an acceleration of the whole coordinate frame without rotations. The first part describes the movement of the center of mass of the body. ~r(t) represents the movement of any element of the body in the center-of-mass frame.Chapter 2. Since this force has no physical origin but emerges just from a an accelerated coordinate system it is not a true force. (2. when a particle is moving on the surface of a rotating body like the Earth.

2 2 The first term is the kinetic energy due to the center-of-mass movement. The total kinetic energy is Z Z   1 3 2 1 3 ~ 2 2 ~ T = d~r ρ(~r)(dt~s) = d~r ρ(~r) (dt R) + (~ω × ~r) + 2(~ω × ~r)dt R . Note that  (~ω × ~r)2 = ~r2 ~ω 2 − (~r~ω )2 = ri2 ωj2 − (ri ωi )2 = ~r2 δij − ri rj ωi ωj and it is thus possible to split the expression into a product depending only on the body and one only on the rotation. This result is a prototypical one.8. This may seem to be rather strange at first. i. the third term never contributes. (dt R) 3 since the integral is proportional to the center of mass in the center-of-mass frame in which ~r is measured. This is a matrix-vector product. 12 The name tensor is justified. This implies that the unit vectors are time-dependent. . The third term is only relevant if the center of mass does not move. It describes that the a time derivative is split into two components. Rotating bodies revisited system. which yields ~ i + ri dt~ei (t) = dt R(t) dt~s(t) = dt Ri (t)E ~ + ~ω (t) × ~r(t) since the time change of the unit vectors describing the body is just the angular velocity. and another one relative motion to it. This view leads to a generalization of the moment of inertia of section 2. dt = dR t + ω×. It shows how changes split into outer and inner components. Thus. It is thus also called a covariant derivative.11. However. The total movement of any element of the body is thus given by ~ + ~r(t) = Ri (t)E ~s(t) = R(t) ~ i + ri~ei (t). this is used to define the (symmetric) tensor of inertia12 Z  Jij = d~r3 ρ(~r) ~r2 δij − ri rj . e. especially in general relativity and particle physics.36 2. by construction  Z  ~ ~ω × d~r ρ(~r)~r = 0. The speed of any of the points of the body is obtained by a time-derivative. and therefore by construction zero. Hence. but the generalization of this so-called operator identity plays a very crucial role throughout physics. as it can be shown that the so-defined matrix indeed transforms in the prescribed way as a second-rank tensor. One describes the action on the center of mass.5. the time derivative acting on the total movement is a composition of two elements.

~ + J~ω = L (2.33) Thus. If the body is an ellipsoid. having its eigenvalues on the diagonal. which are the main axises of the body in question. These eigenvalues describe the moments of inertia around rotations along the respective axes in this system. T = M(dt R) 2 2 and thus as a combination of the motion of the center-of-mass and the rotation of the body. these are just the usual three main axises of this ellipsoid. ~ls as a straightforward and analogous calculation shows. . ~ls = M R ~ × dt R ~ + ~l. This tensor of inertia is also determining the angular momentum of the total movement. it is also possible to rewrite the kinetic energy as 1 ~ 2 + 1 ~ω~l. Since the tensor of inertia is symmetric.Chapter 2. Newtonian mechanics 37 The kinetic energy is then given by 1 ~ 2 + 1ω T = M(dt R) ~ T J~ω . it can be diagonalized. 2 2 and the splitting between translation and internal rotation becomes manifest again.

Strong gravitational fields. including optical. it turns out that classical mechanics based on these postulates is not able to describe all situations. High speeds. as can be shown using statistical (classical) mechanics. which becomes quantum field theory. Of course. However. classical mechanics was essentially describe by Newton’s laws of section 2. 1. supple- mented with forces. This is the purview of general relativity. Note that a full inclusion of electrodynamics in quantum mechanics requires quantum field theory 5. together with their resolutions or combinations of resolutions. as those cases to be discussed here do. phenomena are not described by classical mechanics. at all of the above. Note that thermodynamics and hydrodynamics are actually only the application of classical mechanics to large ensembles of particles.2. supplementing Newton’s laws with Maxwell’ laws this is possible. This requires the combination of quantum physics and general relativity is the. also electromagnetic.Chapter 3 Special relativity So far. which necessarily induces large speeds and thus includes special relativity 4. This is the purview of quantum mechanics 2. At very short distances. and therefore do not require fundamentally new laws. no matter how complicated the forces are assumed to be. However. This requires the combination of quantum me- chanics and special relativity. not yet fully formulated. This is the purview of special relativity 3. quantum gravity 38 . The situations where failures arise are. Both together form as postulates the framework of classical mechan- ics. at short distances and high speeds. but does not lead to anything conceptually new on the sides of mechanics.

Chapter 3. Special relativity 39

Besides electrodynamics and gravity, there are also two more known sources of forces, the
weak and strong nuclear forces. All known forces can actually be derived from these four,
though a unifying theory of all of them is not yet available.
Relaxing all of these conditions, and covering everything listed, essentially sums up the
complete master study of (theoretical) physics. As a first step, here one of those will be
included, those which is the least complicated one: Special relativity, i. e. the modifications
necessary to classical mechanics when including large speeds.

3.1 The speed of light and Minkowski space
3.1.1 The equivalence principle
The starting point of special relativity is a reevaluation of Galileo’s transformations of
section 2.10. Consider a system which moves with a constant speed with respect to another
one. This implies that the coordinates in one system are related to the other by

~r′ = ~r + ~v t,

where ~v is the relative speed of the two systems. These are inertial systems, and thus
Newton’s laws hold equally in both of them. But speeds measured in one system differ by
speeds in the other system, since
dt~r′ = dt~r + ~v . (3.1)

This should be a true statement for any moving object when described in either coordinate
This is where a contradiction to experiments arises. While the relation (3.1) holds to
very good accuracy experimentally at low speeds, it does not do so at large speeds. In
fact, the deviation is there also at low speeds, but it becomes quickly so tiny, as will be
quantified later, that it is below any reasonable experimental uncertainties if the speed is
just low enough. Thus, something does not fit.
As it turns out, Newton’s postulate are the problem. More precisely, their statements
about inertial systems and that Newton’s second law (2.2) holds equally in all inertial
systems connected by Galileo transformations. The reason is that Newton’s laws make
a very deep assumption about space-time: That space-time has an Euclidean structure.
However, as it turns out, nature is not of this form, but has a more complex structure,
which only at low speeds looks Euclidean.
In principle, the only necessary thing is to replace the space-time structure accordingly,
and then derive the results from this.

40 3.1. The speed of light and Minkowski space

It is, however, more instructive, to pursue a slightly different path, where the space-time
structure stands at the end. As always, there are several equivalent ways of formulating
the basic postulates, and each form implies the other.
This different approach will start rather from a physics motivation.
Consider again Newton’s laws. It may look like that the second law, equation (2.2), is
the most important part. Just because so far this was the one made most reference to.
But this is not quite the case. The probably most important statement is that physics is
the same in all inertial systems. As the choice of inertial system is done by an observer,
this implies that physics is independent of any observer. While this appears obvious at
first, this is actually a very deep statement. And a postulate. Changing (2.2) somewhat
would just change the resulting trajectories. Changing this statement would give up any
notion (or even hope) of something like an objective reality1 .
So, this will be the first part of the new set of postulates: Physics is observer indepen-
dent. This implies that the further postulates should have the same form in every inertial
frame. In particular, this implies it is impossible to make statements about the frame from
within the frame: There is no absolute frame. Only relative differences between frames
can be identified. This is known as the equivalence principle.
So far, this seems not be something different than Newton’s requirement that physics
should be the same in all inertial frames.
The big difference comes from an experimental observation: All massless particles
move at a fixed speed. It is called the speed of light c, as light in the form of photons
has been the first observed particles to be massless. Since this speed must therefore
be the same in all frames, this implies that (3.1) cannot be correct. Or more aptly
put: Galileo transformations are not appropriate to transform between frames. Since
this transformation comes from the assumption of an Euclidean space-time, it is this
assumption, which must be wrong. It now remains to work out the consequences. This is
entirely based on the equivalence principle and the experimental observation of a unique
speed of massless particles in every frame2 . That massless particles may create a problem
can also be seen from Newton’s law (2.2), as it implies that massless particles cannot be
described by it. Though this can be alleviated by the formulation using (2.1), it still leaves
a certain feeling of uneasiness. Also this will be fixed in special relativity.
Whether such a thing can exist at all is an unsolved question of science philosophy. In physics, it is
the description of observations, which is the core duty. And so far all observations available very uniquely
point to an observer-independent description.
Historically, it was not the speed of massless particles, but of the speed of light. However, this would
require electrodynamics, which will be avoided here, as considering massless particles lead to the same

Chapter 3. Special relativity 41

3.1.2 Minkowski space
So, it is now clear that something is not right with the way inertial frames, and transfor-
mations between, are defined. This leads to the question how this can be fixed. Consider
the distance traveled by a massless particle with its fixed speed3 c. In a given reference
frame, the distance traversed in time t will be

~r2 = c2 t2 .

Something which would be nice to keep is that space is isotropic - the x, y, and z directions
are all the same. The question is, how does this change to a new coordinate system. There
are several possibilities. E. g., the left-hand side could change by changing the metric,
additional terms could arise, and many more. Experiment would have to decide, which is
correct. Though instructive, considering all possibilities would lead far beyond the scope
of this lecture. Therefore, with hindsight, the only change will be to give t a non-trivial
transformation property, i. e. the time could also differ in both coordinate systems. This
deviates from the starting point of the Galileo transformation (2.28).
In another coordinate system then the relation
′ ′
~r 2 = c2 t 2

holds, where the important piece is that c is the same in both equations. Since zero is
zero, both equations can be combined to yield the equality
′ ′
~r2 − c2 t2 = ~r 2 − c2 t 2 . (3.2)

In the case of a Galileo transformation, a similar equality holds, but there c would be
transformed and t would be fixed.
Nonetheless, such a relation as (3.2) looks quite familiar from linear algebra. It reminds
of an invariant length, but of a vector space with a non-Euclidean metric. This is actually
correctly inferred.
Define now space-time as a vector space, the so-called Minkowski space, M4 as a vector
space having a non-trivial metric4
 
−1 0 0 0
 0 1 0 0
µν  
g = .
 0 0 1 0
0 0 0 1
Which is, of course, the speed of light.
It is equivalently possible to switch the signs of all entries, g → −g, without changing the physics.
This will be discussed more in detail below, and is thus a convention.

42 3. Rotations are matrix-valued operations U. while the Latin indices still enumerate the spatial components starting at 1.1. it is useful to repeat a few results about vector spaces with non-Euclidean metric. as above. The only difference is that. This gives rise to define covariant and contravariant tensors. e. which are quite interesting nowadays. Since it has been the question of coordinate transformations which led to this point. The question is now. ~x y z Conventionally. As it will turn out that this is indeed correct. of course. However.10 a0 also automatically shifts the origin of time. the components of this vector are enumerated with Greek literals starting from zero. no matter whether ~x goes to Λ~x. Given that the coordinates transform as Λx. a translation works as usual. While for special relativity this does not make any difference. whether rotations in such a space are compatible with the equivalence principle and the speed of light. or Λ−1~x. xi = (~x)i . which leave the scalar product invariant. Leaves rotations. this is true. The scalar product is then x2 = xT x = xµ g µν xν = −(ct)(ct) + (~r)(~r) = ~r2 − c2 t2 . Here it is once more important to remind of the concept of tensors. in contrast to the Galileo transformations of section 2. the first important step is that of how they work in Minkowski space. It will therefore not be used here. Most interest- ing is. a replacement xµ → xµ + aµ shifts the origin. 5 In older texts this is often done by having an ordinary Euclidean metric but making the time com- ponent (or all space-components) purely imaginary. .2) is a preserved length. The speed of light and Minkowski space Then this defines a four-vector5   ! ct ct   x x= =  . whether the standard linear algebra for such a space correctly reproduces physics. and thus in such a space indeed (3. i. in Euclidean spaces ~xT Λ−1 Λ~x = ~xT ~x. x0 = ct. First of all. this is not well suited for generalizations.

. it will be defined that the transformation matrix Λ has its indices mixed. xµ . Likewise. To separate both quantities.2) is correct.Λµνnn M ν1 . What has not yet been defined is how the transformation matrices Λ should look like..µn = Λνµ11 . Tensors of higher rank are then defined by their transformation properties on their indices. If some mixtures of Λ and Λ−1 appear.Λνµnn Mν1 . . that is the doubly appearing.. This will happen if ′ v T w ′ = v T ΛT gΛw = v T w is satisfied.µn = Λµν11 . and Λµν = (Λ−1 )νµ . such that (3. Special relativity 43 any object which transforms in the same way is called a covariant tensor (of rank 1)..4) and thus the Λ must leave the metric g invariant. Any object transforming like Λ−1 x is called a contravariant tensor (of rank 1).. (M ′ )µi . indices are appearing once up and once down. the transformation rules read (x′ )µ = Λνµ xν (x′ )µ = Λµν xν . of rank n and (M ′ )µi . and thus summed.Chapter 3.. a contravariant tensor of rank n. covariant tensors will have their index down... (3.νn ..νn . it is called a covariant tensor.3) Thus.. Note that this implies ΛT gΛ = g.. it is called a mixed tensor. If it transforms by a product of Λ. and contravariant tensors up xµ . (3..

It is therefore said that g raises an index. This is not surprising. It can likewise be shown that a matrix with entries defined as solutions to gµν = g µρ g νσ gρσ 6 Mixtures can be deconstructed as consecutive transformations. a contravariant vector differs from its covariant version by a minus sign in the first component. as g µν vν = v µ .44 3.  0 0 1 0  sinh ξ 0 cosh ξ 0  0 0 1 0  0 0 0 1 0 0 0 1 sinh ζ 0 0 cosh ζ (3. But this implies that gv transforms like a contravariant tensor. The other three are odd. J2 =  . J3 =   0 − sin α cos α 0 0 0 1 0  0 0 cos γ sin γ  0 0 0 1 0 − sin β 0 cos β 0 0 − sin γ cos γ       cosh η sinh η 0 0 cosh ξ 0 sinh ξ 0 cosh ζ 0 0 sinh ζ  sinh η cosh η 0 0  0 1 0 0  0 1 0 0        J4 =  . and any invariance there will simply require that the transformation matrices Λ are rotation matrices.       1 0 0 0 1 0 0 0 1 0 0 0 0 cos α sin α 0 0 cos β 0 sin β  0 1 0 0        J1 =  .5) all other matrices satisfying the condition (3. Note first that Λ−1 ΛT = 1 for the matrices J1 to J3 . For the other three Λ−1 = ΛT . Thus. J5 =  . as for ordinary rotations. when ignoring the time. From this follows that. since ΛT = Λ for them.1. The speed of light and Minkowski space It turns out that there are six basic matrices satisfying the condition (3. as the spatial components of the metric g are just an ordinary Euclidean metric. . The first three can be recognized as rotations in three-dimensional space.4) can be written as a product of these six elementary matrices. J6 =  .4). where the first line applies to ordinary rotation and the second one to the new class of matrices. But before continuing with them it is useful to derive some more general statements. depending if Λ is constructed from the first or the second three6 Λ−1 gv = Λ−1 gΛ−1Λv = ΛT gΛΛv = gv ′ Λ−1 gv = ΛT −1 gΛ−1Λv = (Λ−1 )T g(Λ−1)Λv = gv ′.

Vectors with negative norm are called timelike. 0 0 where these vectors have negative. respectively. vectors with zero norm are called lightlike. smaller or equal zero. and positive norm. for reason to be come clearer later. negative. The consequence of such a metric is that scalar products are no longer positive semi- definite. and can be thought of the inverse of the metric. without metric. zero. This also implies that a scalar product is vµ g µν vν = vµ v µ . Especially v T gw = a can now be greater. Special relativity 45 has the effect that gµν v ν = vµ reverses the change from contravariant to covariant. and furthermore gµν g νρ = δµρ . or zero length. of a covariant and a contravariant vector. Thus. Examples for all possibilities in two dimensions are   1 0   v1 =   0 0   1 1   v2 =   0 0   0 1   v3 =  . and vectors with positive norm are called spacelike. This also implies that the ’angle’ v µ wµ |v||w| . vectors can be classified by the fact whether they have positive.Chapter 3. and therefore equivalent to the product. even if v = w.

Thus care should be taken to stick with a single convention. this is equivalent. and in fact this expression needs no longer to vary between -1 and 1. Consider two coordinate systems moving relative to each other at a fixed speed v along the x-axis. say.6) 2 1 − vc2 v β = c 1 γ = p . the solution for sinh η is v c sinh η = q = γβ (3.  y  z and likewise J5 and J6 . Rather.46 3. this will reverse also the signs for the classification of timelike and spacelike.1. This leaves to understand the transformations induced by J4 to J6 . Thus. e. It remains to determine how η. (3. ξ. Applying it to a vector delivers. Using q cosh η = 1 + sinh2 η. a matter of convention. and this expression equal to cosh η. However. whether the negative or the positive norm is conserved. and ζ are related to the parameters of the transfor- mation. as with rotations before.   ct cosh η + x sinh η x cosh η + ct sinh η    J4 x =  . The origin of the new coordinate system must satisfy vt cosh η + ct sinh η = 0 as the two move with respect to each other. Note that nothing of these changes when multiplying the starting point (3. except that the necessary metric would have reversed signs. as it does in Euclidean metric. 1 − β2 This leads finally to the Lorentz transformations   γ(ct − βx) γ(x − βct)   J4 x =  . The speed of light and Minkowski space between two vectors needs no longer to be a real angle.7)  y  z .2) by a minus sign. I. they mix spatial positions and time. which is very different from ordinary rotations. this should be considered to be a generalized angle η.

and the J4−6 describe movements along the three coordinate axises with constant velocities. Special relativity 47 If the speed is small. It is not yet defined how the dynamics.2 Measurements of time and distance To define the kinematics requires to give meaning to distances and time differences.  y  z and therefore reduces to the Galileo transformation. They can be supplemented by operations like space and/or time inversions. Combining the above Lorentz transfor- mation. note that in (3.3).7) not only space changes. to ultimately define speeds and accelerations.concurrence as known in the Galileo group is lost. in accordance with the requirement that at low speeds conventional Newtonian physics should reemerge. with these translation leads to the so-called Poincare transformations. in relatively boosted systems two events will not take place at the same time .Chapter 3. Both together form the so-called Lorentz group. which is sometimes abbreviated to SO(1. since xi = γ(x′i + βct′ ) . β ≈ 0 but cβ = v. The J1−3 are ordinary rotations. this reduces to   ct   β≈0 x − vt J4 x =  . This will be done in the following. but also time. Consider now some rigid object with an extension l = x2 − x1 in its rest frame. should be formulated. and is still an admissible transformation. in analogy to the conventional rotation group. and even the kinematics. all of this is essentially only replacing the arena of mechanics. An observer moving along the x-axis will define in her frame of reference the length of the stick by the difference in endpoints at the same time in her coordinate system. so-called boosts. As noted. The matrices Ji therefore define the possible norm-conserving transformations. It is thus best to use the inverse of (3. To understand why this is not obvious.7). 3. However. Especially. This requires to understand how distances are perceived in different coordinate systems. the so-called proper Lorentz transformations. this does not correspond to the same time in the original coordinate system. It should be noted that a shift by a constant (four) vector is satisfying all the above criteria.

Therefore. This has nothing to do with the finiteness of a maximal speed: Using a Galileo transformation and a finite speed to transfer the information from the different points of the beginning and ending of the stick will yield a different result. This phenomena is called time dilatation.2). Multiplying.2). but one accelerated and then coming back at some point in the future. yields J4 (η)J4 (ξ) = J4 (η + ξ) = J4 (ζ).48 3. 7 Note that these are statements only in frames moving with fixed speed to each other. This makes statements about how speeds add to each other. This is as for ordinary rotations in Euclidean space: Performing a rotation. Both spatial length and time are only components of a 4 vector. This phenomena is called length contraction. . are shorter . they are not separately invariant. Lorentz transformations only keep the norm of a 4 vector invariant. In the end. as γ −1 is always smaller than 1. both time dilatation and length contraction should not come as a surprise. The reason is that spatial distances. Another interesting feature is what happens when performing two boosts after each other. J4 twice for two different speeds. The reason for this is the opposite sign of spatial and temporal parts of the metric. the stick is shorter in the moving frame. And just like as with rotation. That this leads to actual shorter distances is because of the minus sign in the metric. and thus also some opposite effect has to be expected for both time and space here.2. It is important to note that this is a reflexive phenomenon: Not only will the observer in the moving frame think sticks are shorter and times longer in the other frame. when starting at the same point. as it is not possible to have an absolute speed. Measurements of time and distance Calculating the difference yields l′ = x′2 − x′1 = γ −1 x2 − βct′ − γ −1 x1 + βct′ = γ −1 l.a Lorentz transformation only preserves the length element (3. dues to (3. In the same way time differences measured at the same point change. Thus. merely a relative speed7 . the components of a vector are not individually invariant. In both frames. e. say. g. This is the equivalence principle at work. an observer in the other frame will see the same for the primed frame. but they appear longer. need general relativity to accurately describe. but only the total length. Although the indefinite norm has some influence on this naive picture. and reduction in one component’s absolute value has to be compensated by an increase in another component’s. Accelerated frames. physics looks the same. but not a spatial length element. and become ∆t′ = t′2 − t′1 = γ(t2 − t1 ) = γ∆t.

this will make only a difference when introducing fields. and once more shows how Galileo transformations emerge as a limiting case of Lorentz transformations. at small speeds it reduces to its Galileo form (2. the borderline is the case of zero length.Chapter 3. The second is that this implies that the total speed can never exceed the speed of light. the only solutions for β > 1 require the βi to be complex. which has a timelike distance to the reference point. Finally. which cannot influence each other. and thus must be the same in all frames. This implements the experimental observation that there is no speed possible faster than the speed of light. this justifies the name spacelike for these vectors. since causal influence is a physical feature.6). But this is nothing than the statement that the connecting vector of both events has a positive norm. All points 8 In a more refined version. a physical not sensible result. In fact. which is the part of Minkowski space. as it must be. . e. As the length of the vectors are invariant under a Lorentz transformation. which shows that expressed in these quantities boosts are additive. While in principle more exact. Using (3. and thus the connecting vector is timelike. it is actually that no information. This fact can be visualized by the so-called light-cone. This is because their distance is so large that no object can reach them quickly enough. The probably most profound is that there are events. as for any values of βi ≤ 1 the total speed also satisfies β ≤ 1. can be transported faster than the speed of light. i. On the other hand. by construction. carried by energy. In fact. causal connections remain invariant under the change of coordinate systems. as there is always an intervening space. the total speed is also the speed of light. First. This observation has far-reaching implications. and thus lightlike vectors. if both speeds are at the speed of light. for any fixed point. This happens. Thus. two events can interfere if the opposite holds. if (ct)2 < (~r)2 . Special relativity 49 which uses that cosh(η) cosh(ξ) + sinh(η) sinh(ξ) = cosh(η + ξ) cosh(η) sinh(ξ) + sinh(η) cosh(ξ) = sinh(η + ξ).31). points localized in space and time. In a fixed coordinate system. They are said to be causally disconnected. this happens as two points cannot communicate earlier than a signal traveling at the speed of light can travel from one to the other8 . this implies for the total β value β1 + β2 βi ≪1 β= = β1 + β2 1 + β1 β2 This so-called Einstein’s addition theorem of velocities has a number of remarkable prop- erties.

as time was an immutable concept. which is called the eigenzeit. To find one. Note that by convention the eigenzeit is . the position along the trajectory has been identified by the time. Now. both being actually Euclidean covariant tensors of rank 1.3. and thus relativistic kinematics and. consider an infinitesimal move along the trajectory. The right-hand side now defines a quantity τ . So far. It is given by a sequence of four-vectors x. To achieve that physics is invariant under rotations was to formulate all equations in terms of tensors of the same type on the left-hand side and the right-hand side. the eigentime is the ordinary time. and the same in all frames.often called covariant . time is itself part of the trajectory. as it describes the propagation of a particle along a line in the space-time world. Relativistic kinematics in the past to the event can therefore causally influence it. This was possible. it is best to be inspired by the classical case and how it reacted to rotations. Thus. The length of this move is given by 1 (dxµ )2 = −(dτ )2 c2 where the prefactor is chosen by convention. It should be noted that the trajectory xµ (τ ) is also called a world-line. ultimately. will the event itself can causally influence any event in its future light-cone. Also these light-cones are invariant under Lorentz transformations. which uniquely identifies a position along the trajectory. however.50 3. where t0 is the time coordinate of the event. and a new trajectory parameter seems more appropriate.3 Relativistic kinematics With this all necessary ingredients are available to describe the motion of a particle in special relativity. as the advance is uniform in it. this reduces to 1 1 (dτ )2 = − 2 (dxµ )2 = 2 (dct)2 = (dt)2 . and which by construction exactly describes the change in trajectory. 3. where past is defined by t < t0 and future by t > t0 . In the same way. When changing to the rest frame. sometimes also called eigenframe. dynamics. The basic ingredient in the formulation of Newton’s laws was to require the involved quantities to be vectors. justifying its name and its normalization. the generalization will be completely in agreement with Lorentz invariance . The first task is to describe the trajectory of a particle. c c Thus. Newton’s law (2.2) is actually an equation involving two tensors of the same rank. To achieve this aim. in the rest frame.if particular care is taken to involve only tensors of the same rank and type on both sides of the equations. It is therefore no longer a suitable concept.

. as it reduces at low speed to the conventional one. due to (3. Special relativity 51 taken to be the positive value.8) dτ dt γ~v ~v The last equality shows why this is a useful generalization of the concept of speed. and which. reduces to the non- relativistic one at low speeds. The eigenzeit can now be used to generalize the concept of speed to the four velocity ! ! dxµ dxµ γc βi ≪1 c uµ = =γ = = . the first postulate is that the mass of a particle. and especially does not change under Lorentz transformations. (2. 1 − β2 Note that this is also true at low speeds.8). (3. To generalize Newton’s law. as then ~v 2 can be neglected compared to c2 on the right-hand side. the time interval dt is the one measured in a system moving with respect to the particle’s restframe. it is best to start from the more general formulation using the momentum. 1 − β2 uµ uµ = γ 2 (~v 2 − c2 ) = −c2 = −c2 . which is used to describe its trajectory.1). 3. (3. as a direct generalization of the non-relativistic one. This also implies that the four-speed is a timelike vector. It is an interesting feature that the length of the four speed is actually constant. The relativistic momentum is given by pµ = muµ .Chapter 3. This is ! ! dpµ γ β~ F~ βi ≪1 β~ F~ = Kµ = = .9) dτ γ F~ F~ By construction. It then remains to postulate the relativistic version of Newton’s law such that it is an equation in terms of four-vectors. m. is still a well-defined quantity. This again makes the eigen part of the name eigenzeit clearer: τ is the time in the restframe of the particle.2). this has the correct low-speed limit. Note further that for any arbitrary frame r s  2 1 1 d~r dτ = 2 ((dct)2 − (d~r)2 ) = dt 1 − 2 = γ −1 dt c c dt Thus.4 Relativistic version of Newton’s laws To formulate a relativistic generalization of (2.

even if it is a scalar quantity.12). Going back to conservative forces in (2. This also implies that energy is conserved. As this kinetic energy is proportional to the mass.4. It seems to make no sense that a particle at rest has kinetic energy. ~ v. Thus. as this statement holds component-by-component. It can be transformed to energy. The reason is that it should rather be interpreted as the total energy of the particle. An important consequence of this is dt p1 + dt p2 = K 1 + K 2 = K 1 − K 1 = 0. Conversely. There is however still the question of the constant in (3. the total four momentum is conserved in a closed system. thereby changing the identity of particles.52 3. In particular. This is especially different. this is also called the rest energy of the particle.10). as here only the energy is conserved. (3. which then again can be conserved to mass. This is a very important statement. Relativistic version of Newton’s laws However.11) giving a connection between energy and mass. even at rest. It is therefore useful to reinterpret it as the total energy of a particle E. this implies that βi ≪1 m~v 2 T = γmc2 = mc2 + (3. and this is a contribution to this total energy every particle has. Thus. and requires further scrutiny. dt T = F~ Thus. and therefore the possibility to convert one into the other: Mass is no longer a conserved quantity. as it implies the equivalence of energy and mass. What remains in place is Newton’s third law of action and reaction. the fourth component of (3. Four-momentum conservation supersedes energy and momentum conservation of non-relativistic mechanics.9) looks strange.9) is associated to the change of kinetic energy in the non-relativistic limit.10) 2 is the relativistic generalization of the kinetic energy. this kinetic energy is conserved in the absence of forces. From this follows E2 pµ pµ = −m2 c2 = p~2 − 2 c or that the energy of a relativistic particle is given by p v =~0 ~ E= p~2 c2 + m2 c4 = mc2 . the fourth component of the four momentum is the kinetic energy. but this does not need to . the fourth component of (3.

3). Note finally that a free particle will again have a constant four-velocity. (3.11) shows that in the massless case the four-momentum becomes light-like rather than space-like. as it requires also new postulates for the forces. but only accidentally. as it also contributes to the fourth component.9) no longer implies that the acceleration ~a needs to be proportional to the force. Rest mass is no longer individually conserved. Thus. While the formulation (3. treating dynamical problems is far less trivial.9). and it is not yet clear. how massless particles should act. It should also be noted that (3. Considering forces like the one of a harmonic oscillator or the gravitational force. So far. That is of particular importance for the relativistic generalization of scattering processes. As a consequence. Moreover. there is a conceptual problem in applying (3. as this corresponds to an eternal and immutable force field. the force itself needs to have the speed of light as maximum propagation speed build in. it assumes that the information about the propagation of a particle is instantaneously transmitted to the other particle.6. which starts from scratch from covariant quantities. It uses the positions of every particle at a given instance of time. This is inconsistent. Rather. it is actually not a very convenient one. which does not change over time.9) is certainly a possible formulation of relativistic dynamics. But the situation becomes more complicated when considering the two-body problem with the original gravitational force (2. (3. they entail the problem that these forces act everywhere simultaneously. Thus. the results will not be compatible with physics. However. a particle can experience lateral accelerations with respect to the force.9). still the mass appeared throughout. It is therefore useful to first introduce a particular formulation of classical mechanics. now the corresponding formulations will be developed in classical physics. It is certainly possible to use the one-particle version in (3. if four momentum conservation implies it. and vice versa. Thus. and their fully special relativistic form then discussed in section 5. and thus for a massless particle the momentum is proportional to the energy. Special relativity 53 be in form of mass. In addition. and thus move along a straight line in the three-dimensional subspace. as the maximum speed in special relativity is the speed of light. in which special relativity can be more easily embedded.9). This implies that it describes a movement of a particle at the speed of light. rather than that of a massive particle with a finite speed.Chapter 3. . Such effects can be measured. Thus. though its is formally possible to use this force in (3.11) implies for massless particles E = c|~p|. and represent a genuine effect entirely due to special relativity. To attack this problem also a different formulation is better suited.

this will emphasize concepts which act on a more fundamental level. these two chapters provide the basis for the generalization of mechanics to quantum mechanics. without dealing either with much more complex problems or with quantum physics most of the following will be rather experienced as a complication rather than a simplification. if existing.. However. Rather. g. and therefore the formalism would become obscure.1 Constraints The first basic insight is the realization that most particle movements are not as simply described as before. would one introduce the following with suitable complicated problems. The following chapters do not aim at any more elaborate problems with more compli- cated potentials and/or systems until chapter 6. 54 . In fact.Chapter 4 Lagrangian mechanics In the previous chapters the basic concepts of mechanics have been formulated. e. potentials. a particle moving through a winding tube under the influence of gravity. this and the next chapter 5 aim at a reformulating of mechanics. the sheer complexity of the problems would overlay everything. All of the basic physical mechanisms have been collected. While this provides at first sight no new physical systems. in more complicated situations the methods to be discussed in the following turn out to be superior to the ones introduced so far. 4. All the conceptual ideas are there. Finally. Furthermore. The only thing in which other problems of mechanics would differ is by other versions of the forces or. It is therefore a precarious problem to exhibit the reason behind this formalism. Consider. The downside of this is that the motivation for many of the ideas in the following is not at all obvious at first.

E. are no longer independent. In general these constraining forces are essentially impossible to determine. g. Hence two of the coordinates are actually functions of the third. Therefore. In a sense. That the right-hand is zero can always be achieved by a subtraction. following the pull of gravity. but is confined to a one-dimensional path through the tube. Such constraints actually eliminate redundant information. To describe this using Newton’s equation (2.2) would require to know the particularities of these forces. since the consequences of them are known: The particle remains in the tube. the choice of z was arbitrary. The consequence of that is that the three coordinates. Otherwise the particle would fall through the tube walls. To formulate them would not require to fix the geometry of the tube. g. however. there is only one way to go (either forward or backward). which appears to make the solution of the problem using the techniques of chapter 2 essentially impossible. The necessary insight to deal with this problem is that it is actually not necessary to really know these forces in detail. it is better to formulate the conditions more generally as a so-called set of constraints fi (~r) = 0 where the index i runs over the number of restrictions. In fact. This also shows that the constraints are not unique. effectively its movement is reduced to one dimension. The first constraint requires the particle to move in the plane of the circle. e. f2 = |~r| − R would equally well be possible. The number of independent constraints is. unique. on top of the gravitational force. Since the particle moves along the tube. Lagrangian mechanics 55 The particle cannot move in all directions equally. in the above example it would be from one to two. To have such a path requires that the walls of the tube acts with a force on the particle.Chapter 4. the constraints could be thought of as . y. there would be two constraints. In this case. and z. if need be. A simpler example would be the movement of particle which is required to move on a circle with radius R. which has a determined number of dimensions. x(z) and y(z). So. The second constraint then further enforces it to move along the circle. f1 (~r) =z= 0 f2 (~r) = ~r2 − R2 = 0. Of course.. the two additional coordinates did not carry any information relevant to the movement of the particle: These information could be completely reconstructed using the one remaining coordinates as well as the constraints. In the previous example. the constraints reduce the full vector space of coordinates to a subspace. x. which are so- called constraining forces.

Constraints. The first map will transform from a vector space with constraints to a vector space without constraints and less dimensions. ~rn ) = 0 if there are n particles.. In general. there are a few more words to be said about the type of constraints. Fluid dynamics.. a particle which moves in an elevator.13 to transform into a second vector space in which the problem becomes trivial. In fact. is again a different issue. all of the constraints did only involve the coordinates. where the matrix cij is symmetric with cii = 0. It is actually this idea which will be formalized in the following. the dn-dimensional coordinate space.1) 1 Continuous mass distributions will usually imply constraints for the center-of-mass. . this is not necessary. many problems of section 2 did involve multiple particles1 . though this will not be obvious. The second map will then transform to a vector space with the same number of dimensions. All non-trivial information will then be contained in the maps between the three vector spaces. as well as those which guarantee internal rigidity of the body. . fj (~r1 . Correspondingly. Constraints the construction principle for a one-dimensional vector space of non-trivial geometry. t) = 0 which is then called a holonom-rheonom constraint. Thus. but with trivial metric.. to be touched upon in section 6. which itself moves along the z-direction with constant speed v would have the constraint z − z0 − v(t − t0 ) = 0. g.56 4. ~rn . It thus applies only to the non-relativistic case. .3. the aim will be to make a further step in section 5. Alternatives can be that they include explicitly the time2 . but non-trivial metric. So far. and this distinction makes no sense any more.. E. the constraints only involved a single particle. if a problem involves several particles the constraints can involve the coordinates of some or all particles. This rigidity would then be encoded by M(M − 1) conditions (~ri − ~rj )2 − c2ij = 0. Before embarking on this process. consider a rigid body made up of M discrete particles with individual masses. As an example.. However. which are of this form are called holonom-skleronom. So far.. is reduced to a dn − N dimensional space if there are N constraints. where d is the number of space(-time) dimensions.1.. 2 In a relativistic treatment time and space are treated equally. (4. fj (~r1 .

The dn − N coordinates of this space are called generalized coordinates. E. and would thus only vary between 0 and 2π. any set of constraints which do not specify a system with only independent coordinates is also not holonom. in the case of the circle it would be equally valid to choose the double angle or half . However. and some non-holonom constraints may remain. it is appears reasonable to indeed make the trans- formation from the ordinary position space to a new space. they are also no longer holonom. The first condition is the. just their number. Note that the number of generalized coordinates does not correlate with the number of involved particles. In such a case the constraints do not reduce the number of degrees of freedom. In the above example of a particle moving on a circle a choice for the generalized coordinate is the angle along the circle. The second condition is that there is a unique relation between the generalized co- ordinates and the original coordinates: It must be possible to reconstruct the position of the particles in position space from the configuration space at every instance of time unambiguously.2 Generalized coordinates Once the set of constraints are known. which would be the case for ordinary coordinates. Especially all coordinates can range over the full domain of definition. This space is known as the configuration space.Chapter 4. 4. one holonom and one non-holonom constraint. g. especially speeds or acceleration. they may be of mixed type. g.1) could also be formulated as dt z − v0 = 0. Finally. This can be helpful in practice. However. In such cases the following machinery can therefore possibly only be applied partially. the so-called con- figuration space. This said. a formulation involving only vectors is often better to make the symmetries of the problem. Lagrangian mechanics 57 This shows that constraints can be as well formulated using only the individual coordinates. rather than with vectors. statement that all coordinates are independent in the sense that no constraints involving them exists. as noted above. Note that the above example (4. if any. the generalized coordinates are not necessarily unique. To qualify as a generalized coordinate two conditions must be met. If the constraint involve also other quantities. dn − N-dimensional space. they do not need to be a usual coordinate in the sense that they vary from −∞ to +∞. Note that if there are multiple constraints. It is a. e. A particular case of non-holonom constraints are those which are inequalities. The following construction can therefore only be applied to a very limited extent. more explicit. somewhat redundant. and could therefore also be considered as a non-holonom constraint.

once such generalized coordinates have been defined. e. While for every set of values of the generalized coordinates θ and φ there is a unique set of ordinary coordinates x.2. the generalized coordinates are not necessarily coordinates in the usual sense. Much mored different coordinates will be found later. approximated by a sphere. the behavior of this system is then fully described by these new equations of motions. These will be in general again second-order differential equations. E. the sys- tem is completely described by them. However. . g. Then. it is possible to derive them with respect to the time. the trajectory both in position and configuration space remains unambiguous. In case of the angle. After a second derivative the generalized accelerations are created.58 4. At the poles the map to the configuration space is not unique. it is useful. g. As an example for generalized coordinates take the movement of a particle on the surface of the earth. a quantity measured in meters has been transformed into an angle. and z. thereby eliminating a number of them as they are trivially satisfied by the constraints and having only dn − N in the end. the reverse is not true. or even the time. Furthermore. yielding generalized speeds. Then there is one constraint ~r2 − R2 = 0 and thus there are two generalized coordinates. Of course. y. requiring therefore 2(dn − N) initial con- ditions. the behavior of the system is uniquely characterized by the so-obtained trajectory in the configuration space. A useful choice are the angles θ an φ which are defined as z θ = cos−1 R −1 y φ = tan x and from which the original coordinates can be reconstructed as x = R sin θ cos φ y = R sin θ cos φ z = R cos θ This result also emphasizes an important statement. as this case already indicates. Generalized coordinates the angle. Since the generalized coordinates uniquely determine all ordinary coordinates. Thus. the values of the generalized coordinates and speeds at some time t0 . a generalized coordinate may equally well be a speed or acceleration. As long as it is an unconstrained and unambiguous description of the system. when rewriting the equations of motions in terms of the generalized coordinates.

1. these variations are not associated with a speed. The virtual displacement have to be compatible with the constraints. R). E. The force F~i has two components. though the translation is unique. as they are better suited for this problem. This is best done by first refining the definition of a real movement. In contrast. it is useful to introduce the concept of virtual movement. i. 0. change by an amount dr = (dr/dt)dt. The symbol δ is used instead of d to distinguish both cases. generalized or otherwise. Such variations are also taken to be infinitesimal for now. e. for a potential which is symmetric around the z-axis it is useful to transform from the usual Cartesian coordinates to cylinder coordinates. To do. a virtual displacement δr is defined to be an instantaneous change of the coordinates. and vice versa. θ) = (t. δt = 0. The precise value of dr is determined by the equations of motion. it is possible to define the work done by a variation for a particle i as δWi = F~ δ~ri . Though this is usually done already in Newtonian mechanics. A real movement is what happens to a particle when it moves along its trajectory. 4. the process of introducing generalized coordinates can also be useful even if there are no constraints. e. e. and thus they can be used like ordinary differentials.. This is. In the example of the particle in a tube under the influence of gravity in section 4. the case if the generalized coordinates are better suited for a particular problem if they are adapted to the symmetry of a problem. Though not at first obvious. Such a virtual displacement is also called a variation. Especially. (φ. g. If time proceeds by an infinitesimal amount dt. ~ The force Z ~ is here the constraining force. it is for more involved problems simpler to use the machinery to be developed below. g. F~ = K ~ + Z.3 The principle of d’Alembert The aim is now to find the equations of motions for the generalized coordinates without knowing the constraining forces explicitly. K~ is the gravitational force and Z ~ are the forces exerted by the walls of the tube on the particle. ~ are all other i. .Chapter 4. The force K forces. Though no power is obtained by such a variation. Lagrangian mechanics 59 It is an interesting observation that a non-trivial trajectory in configuration space in this case. the coordinates r. and are not constrained by the equations of motion. This emphasizes that the usual intuition of position space may not be easily translated to configuration space. 0) translates to a trivial one in position space (0. the forces which act such that the constraints are fulfilled.

The formu- lation of d’Alembert embodies already one of the central goals of the current chapter: It no longer references the constraining forces. This gives another interpretation of the constrain- ing forces: They can be considered as the loss of the forces K ~ i compared to the attained accelerations dt p~i . e. This has been formulated using Newton’s law. This will take some time. The principle of d’Alembert Since Newton’s law (2. Then the principle of d’Alembert states that these loses do not perform work if a virtual displacement compatible with the constraints is performed.3) i i. The first new postulate is then X ~ i δ~ri = 0. In fact. Any displacement δr along the curve will then not require any work by the constraining forces.4) the so-called principle of d’Alembert.2). But the total work done by them is zero. Zδ~ An equivalent formulation of (4. as the forces to keep the particle on the curve need to act transversely to the curve only. any move along the curves coincides with the constraints.2) t i follows. Of course. this will be turned around. it makes sense in an intuitive ways: If one displaces a particle in a way where it does not violate the constraints. ~ i − dt p~i )δ~ri = 0. Now. while it still is an equation of motion which can be solved. no force has to act. . it is already useful for sufficiently simple problems. and a new basic postulate will be made. Now comes the basic step.60 4. a subset of the particles. and further steps. Therefore. (K (4. it is only an intermediate stage. since ~ r = 0. to maintain the constraints. As an example. md2t ~r = F~ . i. and they will ultimately be derived from them backwards. still applies X ~i + Z (K ~ − mi d2~r)δ~ri = 0 (4. consider a particle which moves along some arbitrary curve and which is forced on the curve by constraining forces. and thus no work to be done. This appears again obvious. but it is hard to generalize to more complex one. Z (4. will replace in the end Newton’s laws.3. they can do work on a part of the system. e. constraining forces do no net work.3) is obtained by using (4. This. Thus. Though this postulate cannot be derived.2).

Since the usual coordinates are differentiable functions of the generalized coordinates.4): The displacements δ~ri are not independent.4) to the generalized case. The general case is more involved. Lagrangian mechanics 61 4. Furthermore   ∂~ri ∂ 2~ri dql ∂ 2~ri ∂ ∂~ri ~ri ∂dt~ri dt = + = dt ql + = . Note that the generalized coordinates can usually not be assembled in vectors like the ordinary coordinates. To simplify matters in the following the mass of the particles will be considered as constant in time. To make it more useful.7) ∂dt qj ∂qj and thus the total time derivatives drop out. the virtual displacements in terms of the generalized coordinates are given by ∂(~ri )j δ(~ri )j = δqk . Now the coordinates can depend both implicitly through the generalized coordinates and explicitly on time.6) ij ∂qj ∂qj where in the last step a zero was introduced.4.1 Reformulating d’Alembert’s principle To make progress to a more convenient formulation. (4. (4. the next step is to eliminate also the constraints by switching to generalized coordinates. but restricted by the constraints.5) ∂qk where it was used that.5) X ∂~ri (dt ~pi )δ~ri = mi (d2t ~ri )δ~ri = mi (d2t ~ri ) δqj i ij ∂q j X   ∂~ri  ∂~ri  = mi dt dt~ri − (dt~ri )dt δqj . As a first step consider X (4. virtual displacements are just ordinary differentials. This leaves to transfer the momenta in (4. it is helpful to recognize the limitations of (4.8) ∂qj ∂ql ∂qj dt ∂t∂qj ∂qj ∂ql ∂t ∂qj . apart from the constraints. (4. This is somewhat more involved. and thus ∂~ri dt~ri = dt qj + ∂t~ri .Chapter 4. The index k runs over the number of generalized co- ordinates. ∂qj Deriving both sides with respect to dt qj yields ∂dt~ri ∂~ri = (4.4 Euler-Lagrange equations of the second kind 4. and no time elapses. but leads ultimately to the same results.

4) in terms of the generalized coordinates   ∂T ∂T dt − − Qj δqj = 0 (4. and usually do not even have dimensions of forces. the expression Qi δqi has the dimension of an energy. ∂(~ri )j then the generalized forces are analogously obtained from the same potential when ex- pressed as a function of the generalized coordinates ∂V Qj = − . ∂dt qj ∂qj and thus the work done by the virtual displacement is given by an expression of deriva- tives of the kinetic energies with respect to the generalized coordinates and their time derivatives. Euler-Lagrange equations of the second kind Hence.4) can be rewritten as ~ i ∂~ri δqj = Qj δqj . While d’Alembert’s principle is very general. and in the final expression with the kinetic energy also again time- dependent masses can be used. Using both insights in (4.10) ∂dt qj ∂qj which satisfies our wish to also eliminate all explicit appearances of the constraints. Similarly. the first term in (4. Note that just as the generalized coordinates the generalized forces are no vectors. Still. this form is in many cases far too general.9) ∂qj recovering the eminent importance of the potential. The arguably .4. as at no time it was used what kind of constraints there are. the partial derivative with respect to the generalized coordinates and a total deriva- tive with respect to time commute for the coordinates. But the generalized forces inherit one particular important feature from the forces: If the forces can be obtained from a potential V .62 4. This yields d’Alembert’s principle (4. (K~ i )j = − ∂V . (4.6) yields X  ∂dt~ri  ∂dt~ri  mi dt (dt~ri ) − (dt~ri ) δqj ij ∂d t qj ∂q j X   ∂ 1  ∂  1  2 2 = mi dt (dt~ri ) − (dt~ri ) ij ∂dt qj 2 ∂qj 2   ∂T ∂T = dt − δqj . ~ i δ~ri = K K ∂qj which defines the generalized forces Qj .

In fact. But then the only possibility to satisfy (4. A motion without potential but with ordinary dissipation proportional to the speed is an example of Rayleigh dissipation. as there the generalized coordinates are the usual Cartesian ones. yielding dn − N independent equations ∂T ∂T dt − = Qj .Chapter 4. ∂dt qj ∂qj where it has been used that the potential will not depend on the generalized speeds. where the matrix β allows for some curvature of the dissipative forces. It is therefore equivalent to Newton’s equations of motion (2. all constraints are independent. Then all δqj are arbitrary. Lagrangian mechanics 63 most important special case is that of holonom and conservative systems. the dissipation is proportional and inverse to the speed. Of course. If the constraints are holonom. the potential would depend on the original speeds. which have been experimentally confirmed so far.10) is if the expression multiplying the δqj vanish individually. all fundamental laws of nature. this set of equations describe the system completely. and can freely vary.2). The Lagrange function is the central quantity for brining together special relativity (and possible general relativity) and quantum physics. ∂dt qj ∂qj Since these are as many independent differential equations of second order as there are independent variables. since they are no longer constrained. because of (4. An example for such a situation occurs for so-called Rayleigh dissipation. are of this type. However.2 The Euler-Lagrange equations On the other hand. since the transformation is unambiguous. i. The quantity L= T −V is called the Lagrange function. the principle of d’Alembert takes the form   ∂ ∂ dt (T − V ) − (T − V ) δqj = 0. as Newton’s law it requires 2(dn − N) independent initial conditions to solve. if the system is conservative then. and thus the system would not be conservative. In this case Qi = −βij dt qj . and especially all δqj are independent.4. 4. in more complex systems the effective description as an emergent form may not be of this type. For . e. Otherwise.9).

Nonetheless. they are not yet the simplest form to solve problems in mechanics and.12) ∂dt qj ∂qj the so-called Lagrange equations of the second kind. these equations require 2(dn − N) initial conditions to solve.11) ∂dt qj ∂qj and therefore only involves the Lagrange function and the generalized coordinates. e. since the kinetic energy can depend explicitly on the time.64 4. each term in (4. Since relativistic quantum physics is for many actual problems. it is very useful to find also a better formulation to generalize to non-relativistic quantum mechanics. . and is therefore stated here without proof. Lagrange’s equation of the second kind is what had been sought for.7 those of the first kind. (4. as noted. If the constraint equations are holonom and the system is conservative. However. Euler-Lagrange equations of the second kind classical. while with hindsight its is better to first derive Lagrange’s equation of the second kind first and only afterwards in section 4. non-relativistic (quantum) mechanics it will only play a role as an intermediate step to Hamiltonian mechanics in chapter 5. as they do so often. and there is a function U such that ∂U ∂U Qi = dt − .10) still decouples. such that the sum in (4. and thus yields dn − N equations of motion ∂L ∂L dt − = 0. though this may at intermediate steps again only be obvious with hindsight.4. in some cases it is possible. Still. d’Alembert’s principle for conservative systems takes the form   ∂ ∂ dt L− L δqj = 0. It seems at first sight that a similar simplification is not possible in non-conservative systems. It is worthwhile to emphasize that. or sometimes also Euler-Lagrange equations (of motion)3 . not the best choice to generalize to non-relativistic quantum mechanics. even though the potential energy does not. in solid state physics. (4. so can the Lagrange function. g. as they do no longer contain the constraints nor the constraining forces. serious overkill and far too complicated. if the equations of motions are holonom. ∂dt qi ∂qi 3 That no Lagrange equations of the first kind have appeared yet is due to the fact that historically the equations of the first kind had been discovered earlier in a logical progression.11) is again independent. Again. This is the aim of the remainder of this and of the next chapter. However. Using the Lagrange function. its central importance for any modern formulation of the fundamental laws of nature cannot be overrated. the historical numbers stuck.

yielding only a single relevant coordinate z.. These are two (different) masses mi exposed to a homogeneous gravitational field of strength g.13) fulfills formally the same equations (4. This seems to be a rather drastic condition for the generalized potential. while in truth something will happen when one of the masses reaches the connection point. as well as any kind of generalization to other sets of variables. Many of the following calculations hold still true if the generalized potential rather than the normal potential appears. which fall into this category. most notably when including electromagnetism. and which are connected somehow such that given the z1 = z position of one particle the other can be found at z2 = l − z. In the end.12). and the Lagrange function L= T −U (4. . e. not only on a single one. t) is then called a generalized potential. it nicely illustrates that the constraints do not appear explicitly in the Lagrange function. m1 + m2 I. The system is conservative.Chapter 4. in practice it turns out that there are many systems.3 Examples It is useful to exhibit the basic strategy with examples. the one mass will move up and the other move down with a reduced gravitational acceleration. also 4 The notation f (qi ) will here and hereafter signal that the function f depends on all qi . and especially was not part of the constraints. to abbreviate the notation.4. dt qi ) depends on all qi and dt qi . While the usage of the Lagrange formalism is probably overkill for this problem. First consider the so-called fall machine of Atwood. However. 2 The single Euler-Lagrange equation is then ∂L ∂L dt − = dt ((m1 + m2 )dt z) − (m1 − m2 )g = (m1 + m2 )d2t z − (m1 − m2 )g = 0 ∂dt z ∂z The equation of motion is thus m1 − m2 d2t z = g. however. depending on which mass is heavier. dt qi . The Lagrange function is therefore 1 L=T −V = (m1 + m2 ) (dt z)2 − m1 gz − m2 g(l − z). and the constraints are time-independent. Likewise f (qi . 4. Lagrangian mechanics 65 The function4 U(qi . that in this solution the particles will move to ±∞. This is not seen in the solution as this was not encoded in the original description. Note. This already implements the constraint.

which is more involved. if it is located somewhere on the stick. as it is the positive definite radius. This implies two constraints. These equations boil down to the exponential equation d2t = ω 2 q.4. A possible generalized coordinate is the distance to the origin where the stick is fixed. These constraints are rheonom-holonom. Note that for the choice of suitable initial conditions it is necessary to keep in mind that q ≥ 0. q = r. g.. z = 0 y − x tan ωt = 0.15) z = 0. which can be formulated as. As a second example. Euler-Lagrange equations of the second kind the Lagrange formalism can only yield information about a system to the extent it was accurately described. and are independent of the mass. as long as not (dt q)(0) + ωq(0) = 0 is satisfied. The particle moves exponentially outwards. e. as they explicitly involve the time.66 4. 2ω 2ω which can be inserted into (4. Solving yields (dt q)(0) + ωq(0) ωt (dt q)(0) − ωq(0) −ωt q(t) = e − e . 2 2 This yields the Euler-Lagrange equations md2t q − mω 2 q = 0. the angular component could have been attached to the coordinate y as well. (4. The transformations are then given by x = q cos ωt (4. . Of course.16) to obtain the final result.16) yielding the Lagrange function m  m  L= T −V = (dt x)2 + (dt y)2 = (dt q)2 + q 2 ω 2 .14-4.14) y = q sin ωt (4. Solving the same problem in Cartesian coordinates would have yielded a product of oscillatory and exponential behavior. in which case it is moving to the center. but without any external force and thus V = 0. consider a particle moving along a stick which rotates in the plane with a fixed angular velocity ω.

. t) qi = qi (Q1 . but is somewhat more involved. dt ri ∂ri ∂ri The three Euler-Lagrange equations are therefore exactly Newton’s second law (2.Chapter 4. Only the mathematical framework postulated around Newton’s laws in section 2. Since Newton’s first and third law are special cases of the second law . The general case can also be done.. Such a transformation is also called a point transformation. Qs . consider a conservative system of a single particle without constraints. e. Lagrangian mechanics 67 4.. For simplicity again only conservative .2).12) are not given as vectorial equations. . with s = dn − N the number of generalized coordinates. Consider an arbitrary.4.2) was written in terms of vectors. Lagrange’s equation of the second kind (4. The Euler-Lagrange equations are then m L = (dt~r)2 − V (~r) 2 ∂L ∂L ∂V dt − = mdt (dt ri ) − = md2t ri − Fi = 0. 4. This is not so. But since they are related to the arena of mechanics rather than to the dynamics.they are satisfied as well. . there is a potential V (~r).2) in component form.the action-reaction relations appear in the n-body version . this was to be expected.5 Invariances of Lagrange’s equation of the second kind Newton’s law (2.. which would seriously impede their usefulness. i.2 has to be furthermore carried over as well.4 Equivalence to Newton’s law To see that the formulation is equivalent to Newton’s law (2. invertible transforma- tion of the generalized coordinates qi to new coordinates Qi . as it assigns to any point in the new coordinates exactly one point in the old coordinates. which must stay the same. Qi = Qi (q1 . as can be seen as follows. and specifying the actual coordinate system is only necessary if it is needed to evaluate it component-wise. In this case. qs . Since furthermore the choice of generalized coordinates is not unique.. It therefore had the same form in any coordinate system.. it is an important question whether when changing from one set of generalized coordinates to another the equations change form. t).

∂dt Qi ∂Qi ∂dt qj ∂qj ∂Qi where it has been used that the Euler-Lagrange equations hold in the variables qi .7-4. t) and can therefore be considered as function of both the new and the old coordinates. (4. Then the Lagrange function ∂f Lf = L + dt f = L + ∂t f + dt qi ... Invariances of Lagrange’s equation of the second kind systems with holonom constraints will be considered..17) ∂qi has the same Euler-Lagrange equations... qs .. which is very useful.. its actual functional form may be different. Choose some Lagrange function and an arbitrary (twice differentiable) function f (qi . qs .68 4. However. ∂Qk ∂Qk which will be needed in the following. ∂dt f ∂2f ∂2f = + dt qj ∂qi ∂t∂qi ∂qi ∂qj   ∂dt f ∂ ∂f ∂f ∂2f ∂2f dt = dt ∂t f + dt qi = dt = + dt qj ... Furthermore. In the same way as for (4.. This can be seen by considering the Euler- Lagrange equations for the additional term. . t). qs . as will be seen in actual calculations. To piece together (4. Qs (q1 . it is necessary to note that the Lagrange functions in the new coordinates is obtained by writing L(Q1 .. the Euler-Lagrange equations are indeed form invariant under any transformation of the generalized coordi- nates. t) = L(q1 . ∂dt qi ∂dt qi ∂qi ∂qi ∂t∂qi ∂qi ∂qj . and therefore the parentheses vanishes for every value of j separately. t). .12) in the new coordinates requires three ingredients ∂L ∂L ∂qj ∂L ∂dt qj = + ∂Qi ∂qj ∂Qi ∂dt qj ∂Qi ∂L ∂L ∂qj = ∂dt Qi ∂dt qj Qi   ∂L ∂L ∂qj ∂L ∂dt qj dt = dt + . ∂dt Qi ∂dt qj ∂Qi ∂dt qj ∂Qi Subtracting the first from the third line yields    ∂L ∂L ∂L ∂L ∂qj dt − = dt − = 0. Thus.. Qs .8) it can be shown that ∂dt qi ∂qi = ∂dt Qk ∂Qk ∂qi ∂dt qi dt = . .. There is another invariance of the Euler-Lagrange equations.5. . ... t) depending only on the generalized coordinates and the time. t) = L(Q1 (q1 .

Note that it was important that the function f is continuously differentiable. implying the existence of 3 conserved quantities. Is this the case the corresponding coordinate qi is called cyclic. Its Lagrange function reads m L= (dt~x)2 . are very useful.18) ∂dt qi Even for a particle. the so-called generalized momenta. the pi must be constant. If the Lagrange function does not depend on a given coordinate ∂L =0 ∂qi it follows that ∂L ∂L 0 = dt − = d t pi . Finding thus generalized coordinates in which as many as possible coordinates are cyclic makes problems therefore much simpler to solve. ∂dt qi qi Thus. sometimes also called integrals of motion. as the cancellation would have not worked if it would have been only a partial one. (4. but this terminology will not be used here. These are ∂L pi = = mi dt xi ∂dt xi . As has been seen in section 2. after further tools have been introduced to achieve this goal. Thus the quantity pi is conserved. 4. They are defined as ∂L pi = . because swapping the derivatives would otherwise not be possible. Lagrangian mechanics 69 which are therefore identically zero.7 such conserved quantities. An interesting example for this is obtained for the case of a single free particle.1. This is significant. How to do so systematically will be discussed further in section 5.13. Such an addition of a total time derivative is sometimes called a (mechanical) gauge transformation. as the corresponding motions are trivial. 2 In this case the coordinates are all cyclic. the generalized momenta will in general not coincide with mdt qi . as the more general idea of gauge theories will supersede it in classical field theory and quantum field theory.Chapter 4.6 Generalized momenta and cyclic coordinates To continue. the corresponding generalized momenta. though they may. as will be seen later in section 5. It was also important that the total time-derivative was added to the Lagrangian. it is helpful to introduce a new concept.

A less trivial example is the Kepler problem of section 2. that in section 2. ranging from thermodynamics and statistical physics to solving numerical problems. Note that this result holds true for any φ-independent potential.19) y = r sin φ sin θ (4. Thus. however.3. that of Lagrange multiplicators. during developing them a powerful technique. suitable generalized coordinates are spherical coordinates x = r cos φ sin θ (4.21) as it is a central potential. r is the radial distance between both bodies. In some of these cases Lagrange’s equations of the first kind can be useful. and θ and φ are the azimuth angle and polar angle.3 already the generalized coordinates had been used. usually GN m1 m2 . The reason is. this would have required to solve already the equations of motion.70 4. the movement in a plane is yielded by the Lagrange formalism much easier than before. the situation would be much different. α is the strength of the gravitational interaction.20) z = r cos θ (4. Using the conventional approach.2).7 Lagrange’s equation of the first kind If the constraints are non-holonom.10) in the previously formulated way. it is not possible to go beyond (4. respectively. There is one cyclic coordinate with the conserved generalized momentum pφ = mr 2 (dt φ) sin2 θ = lz which is identified as the z-component of the angular momentum. In this case. Furthermore. 4.7. Solving the remaining equations of motion is in this particular problem not substantially easier than before. The corresponding Lagrange function is m  α L= (dt r)2 + r 2 (dt θ)2 + r 2 (dt φ)2 sin2 θ + . In general.7. 2 r where m is the reduced mass. their purpose is also to implement constraints. the Euler-Lagrange equations are ∂L dt = mi d2t xi = 0 ∂dt xi and therefore coincide once more with Newton’s law (2. Lagrange’s equation of the first kind which therefore coincide with the usual momenta. will be introduced which is very useful to implement constraints in much more general settings. . In fact. but not necessarily for mechanical systems only.7. Would the Cartesian coordinates have been used.

Now. and add them. despite their time-dependence. i. (4. Now assume that there exist values for the λj such that ∂L ∂L dt − − λj fji = 0 ∂dt qi ∂qi for i = s + 1.22) where j runs from 0 to M. qr . the condition (4. under any kind of virtual displacement dt = 0. t)dt = 0... ∂dt qi ∂qi Since the generalized coordinates are not independent. quantities λj should exist .11) still holds.. the separate terms in parentheses are not zero individually. The constraints are hence formulated as differentials and the holonom constraints have already been used to transform to generalized coordinates as far as possible. However. Thus. it is useful to order them. and the remaining constraints have the form fji δqi = 0.. out of the r generalized coordinates only dn − N = s are independent. .Chapter 4. though it was not possible to resolve the generalized constraints... then also any sum of these statements is true.. e. M equations for the dependent variables. r. . ... qr . .. Consider for the following a system where M of the N constraints are non-holonom. This yields λj (t)fji δqi = 0 where the (so far) arbitrary M functions λi are called Lagrange multiplicator for reasons to be seen. r. in general time-dependent...11).. s to be independent and the remaining ones i = s + 1. Thus. but not of the coordinates.23) If this is true. and choose the first i = 1. and therefore   ∂L ∂L dt − − λj fji δqi = 0.. they are still constraints. (4.. r = dn − N + M and i = 1.. It appears at first sight highly non-trivial that some. Since the choice of which are dependent and which are independent is arbitrary. . it is possible to add zero to (4. Therefore multiply all M equations by an arbitrary functions of time. r to be dependent on the former. They are also called constants. Lagrangian mechanics 71 The method of Lagrange multiplicators is particularly suited if the constraints them- selves contain derivatives and cannot be integrated. or there is at least a generalized potential. If the system is conservative. A useful way to write the constraints of this type is fji (q1 . .. as they do not depend on the coordinates. t)dqi + fj (q1 . Furthermore..

.N functions as well as for the M parameters λj . A holonom constraint F can be transformed into a non-holonom one by considering ∂F ∂F dF = dqi + dt = 0. i. Thus. some knowl- edge about them is gained. But since the remaining generalized coordinates qi. as this is exactly the same information content. e. and can therefore be considered to be a generalized constraining force.22). It moment of inertia is J = mR2 . In fact. To actually apply Lagrange’s equation of the first kind requires a dependent . The expressions Q̃i = λj fji . However. s as well. However. the advantage is that more information is gained as well. they are not alone sufficient. Lagrange’s equation of the first kind such that this relation between functions should be correct. Thus it appears reasonable that such values for the λj exist.s are independent. the fact that these qs+1.. However. the two parts of the equations actual express the same statement. where R is its radius.22) have not been resolved. This becomes visible when comparing (4..r are not independent variables.24) with (4. Since the constraints are related to the constraining forces.. Take a hollow cylinder rolling down a hill. act like generalized forces..7. . It is useful to consider an example. Though these are now more equations to solve. and indeed this is the case. and hence ∂L ∂L dt − − λj fji = 0 (4.72 4. The same could be achieved by rather multiplying the terms involving the Lagrange function and summing them and adding single constraints. By superimposing all possibly ways how to express the dependence of the generalized coordinates by summing over all constraints. for practical use this form is more adequate. The constraints (4. Since it can only move along the hill. once formulated using the Lagrange function and once formulated using the constraints. Lagrange’s equations of the first kind can be used as well. so are their virtual translations δqi . it is taken care of that the Lagrange function expresses the same information in a different way. ∂qi ∂t and the derivatives of the constraint therefore take the role of the fji and fj in (4... They thus contain the same information. the generalized coordinates can be chosen to be the the distance along the hill q1 = x. it is necessary to solve N + M. if knowledge about the constraining forces in the case of holonom constraints is required. and still needed to be solved simultaneously.. rather than N equations for the q1. These are Lagrange’s equation of the first kind.. only in different ways.10).24) ∂dt qi ∂qi for i = 1.

The two equations of the first kind of Lagrange and the one obtained from the constraints read ∂L ∂L dt − − λ1 f11 = md2t q1 − mg sin φ + λ1 = 0 dt q1 ∂q1 ∂L ∂L dt − − λ1 f12 = Jd2t q2 − Rλ1 dt q2 ∂q2 −dt q1 + Rdt q2 = 0. f12 = R. It is important to note here that. Lagrangian mechanics 73 further generalized coordinate. 2 The two generalized coordinates are therefore just uniformly accelerated motions. Deriving the constraint equation once more yields −d2t q1 + Rd2t q2 = 0. making it roll. and f1 = 0. even though q2 is cyclic. because this applied for independent coordinates. and the Lagrange parameter is in this case a constant.Chapter 4. Then three times the same unknowns appear linearly in the equations. q2 = θ. this does not yield a conservation law in this case. The cylinder should stay rigid and therefore the constraint is Rdθ − dx = Rdq2 − dq1 = 0. λ1 . and thus a relation between the second-order derivatives of the generalized coordinates. 2 2 where φ is the inclination of the hill. More interesting are the two generalized forces mg sin φ Q̃1 = λ1 f11 = − 2 mgR sin φ Q̃2 = λ2 f12 = . There is only one Lagrange parameter. 2 . and it is possible to decouple them. The one non-holonom constraint has therefore f11 = −1. The Lagrange function of this (conservative) system is then given by m J L= (dt q1 )2 + (dt q2 )2 − mgq1 sin φ. yielding g sin φ d2t q1 = 2 g sin φ d2t q2 = 2R M λ1 = g sin φ. This will be the angle of rotation of the cylinder. Assume that gravity is acting on the cylinder.

. But the considerations in section 2..6 conserved quantities. using the notations of section 2. q1. 2 2 in which four coordinates are cyclic.. ~ 1. Consider two bodies of different masses which interact via a potential V (|~r1 − ~r2 |)...8. In Cartesian coordinates their Lagrange function reads m1 m L= (dt~r1 )2 + (dt~r2 )2 + V (|~r1 − ~r2 |). cos θ)T . reducing its acceleration com- pared to the free-fall acceleration of g. Both are exerted by the forces between the hill and the cylinder.3 q1. This can be seen in the following example.74 4. which makes the cylinder roll. so-called integrals of motions.1..3 = R q4 = |~r1 − ~r2 | q5 = θ q6 = φ.. Conservation laws The first is a force opposing the movement of the cylinder. 4. have been obtained from cyclic coordinates. but there are otherwise no constraints. Though tedious. Introducing these coordinates as generalized coordinates.8 have shown that the movement of the center of mass and the relative movements of the particles decouple us- ing center-of-mass coordinates. which are actually microscopic electromagnetic forces. straightforward evaluation yields the Lagrange function in these co- ordinates M  µ  L= (dt q1 )2 + (dt q2 )2 + (dt q3 )2 ) + (dt q4 )2 + q42 (dt q5 )2 + q42 (dt q6 )2 sin q5 + V (q4 ). while the fourth is the angular momentum along the z-axis with respect to the center of mass. This already follows from the fact that generalized coordinates are not unique.3 and q6 .8. sin θ sin φ. there are immediately four con- served quantities. These are exactly the ones al- ready known. the three components of the momentum of the center of mass. Thus. where θ and φ are the angles in spherical coordinates of ~r = ~r1 −~r2 = r(sin θ cos φ. but not for another. The second is the angular acceleration. the converse is not true. and there may be (many) more conserved quantities than there are cyclic coordinates in the Lagrange function. 2 2 None of the coordinate is cyclic. While cyclic coordinates always yield a conserved quantity.8 Conservation laws In section 4. and a coordinate may be cyclic for one choice. simplifying the problem tremendously.

In this case ∂T ∂adt qi ∂T da T = = dt qi = 2aT. this is a rather abstract conserved quantity. and therefore also for a = 1. a good choice of generalized coordinates can simplfy the problem tremendously.25) and which is therefore conserved in time when the Lagrange function does not explicitly depend on time. which are by far the most numerous ones and the only ones relevant in fundamental physics. ∂dt qi and therefore in this case the Hamilton function is just the total energy. while a much more sophisticated approach will be discussed in section 5. Note that for rheonom constraints this is not true. for skleronom and conservative systems. some general rules will be described which allow to construct many conserved quantities. This implies ∂L H= dt qi − T + V = 2T − T + V = T + V. Then        ∂L ∂L 2 ∂L ∂L 2 ∂L dt L = dt qi + d qi = dt dt qi + d qi = dt dt qi . the total energy is conserved if 5 A function obeying f (ax) = an f (x) is said to be homogeneous of order n. Lagrangian mechanics 75 Thus. finding them is not always simple. ∂(adt qi ) ∂a ∂(adt qi ) which is true for any a. However.13. 4. (4. as the constraining forces can then also contribute energy.8. ∂qi ∂dt qi t ∂dt qi ∂dt qi t ∂dt qi where in the second step Lagrange equation’s of the second kind have been used. At first sight.Chapter 4. However. assuming holonom constraints. This function will reappear also in chapter 5.1 Energy conservation Assume that the Lagrange function does not depend explicitly on the time. The required property of the kinetic energy is therefore homogeneity of order 2. Thus. for skleronom constraints in a conservative system and if the generalized kinetic energy has the property5 T (adt qi ) = a2 T (dt qi ) it is possible to give H a very particular meaning. . Here.18) the Hamilton function is defined as H = pi dt qi − L. Using the generalized momentum (4.

just as at the beginning of this section.3: The potential depended only on the relative distance of two particles. the system must be invariant under translations in the cyclic coordinates. rather on their absolute positions.2 Momentum conservation As may be expected from chapter 3. This also explains that fundamental physics is usually of this type: To the best of our knowledge time has no absolute frame. ∆~ri → ∆~ri + ∆qi~ni . Conservation laws the Lagrange function does not depend explicitly on time. Any transformation t → t + ∆t will leave in this case both the Lagrange function and the Euler-Lagrange equations unchanged. which is just the center-of-mass momentum. This is therefore a particular case of section 4. 4. and there is no distinct time. a very similar argument as in the previous section also holds if the space is homogeneous.7) has been used.26) ∂dt qi ∂dt qi j ∂d t qi j ∂q i where (4. Therefore any shift of the coordinate system ~ri → ~ri + ∆~r would not have changed anything. Thus the corresponding generalized momenta. just as discussed in chapter 3. it is best to consider a conservative system. j . The same statement could therefore be read that if a system is translationally invariant. To better understand this generalization.8. its associated generalized momentum is conserved. The center-of-mass coordinate. e. ~r1 − ~r2 .7. Thus. which is affected. all coordinates must change in the same way under a change in qi . in a generalized coordinate. and thus the derivative is a unit vector ~ni in the homogeneous direction. The absence of explicit time- dependence can also be interpreted in a different way. Corresponding generalized coordinates would therefore be relative coordinates. i. the absolute position does not matter. would therefore be cyclic. as defining any time as t = 0 for the initial conditions is equally good.8.76 4. will be conserved. Homogeneity in space means that everything depends only on relative distances. X pi = mj (dt~rj )~ni = ~ni P~ . Then for a cyclic coordinate ∂L ∂T X ∂dt~rj X ∂~rj pi = = = mj (dt~rj ) = mj (dt~rj ) . the system is homogeneous (or isotropic) in time.6. They are not affected by such changes. just as with the time discussed in case of the homogeneity of time. An example is the central force problem of section 2. (4. Since by assumption.

the choice of coordinate system guarantees in the example that the total angular momentum is oriented along the z-axis. In fact. ∆~ri = ∆qi~ni × ~ri . and actually the total angu- lar momentum is conserved. as only the component L ized momentum. . Deriving this relation with respect to qi then yields X X pi = mj (dt~rj )(~ni × ~rj ) = ~ni ~ mj (dt~rj ) × ~rj = nj L. In this case the derivative in (4.26) will not be along a direction ~ni . Thus.3 Angular momentum conservation A system is called isotropic in space if it does not change under a rotation. Thus. the projection of the total force in the corresponding direction vanishes. the question may at first arise ~ z has appeared as a conserved general- whether it is isotropic. Lagrangian mechanics 77 where P~ is the ordinary momentum of the center of mass. the vanishing of the corresponding generalized force implies that the com- ponent of the total torque along the rotation axis vanishes. this results in three cyclic coordinates. Given the example in the beginning of this section. 4. This is however misleading. The cyclicity implies that the corresponding generalized force vanishes and thus ∂V ∂~rj X 0=− = Qi = F~j = ~ni F~j ∂qi ∂qi j Hence. Therefore.Chapter 4.8. This has also been exemplified in the example at the beginning of the section. The example at the beginning of this section exemplifies this. If the system is isotropic for all possible rotation axises. as the system does not depend on it. the corresponding angular variable will again be cyclic. j j where L ~ is the total angular momentum. If space is homogeneous in all directions. but rather be like a rotation of the coordinates. As before. the total angular momentum is conserved. Here it surfaces that the angular momentum is defined by a coordinate system acting as a reference point. Therefore. the two other components vanish by construction. the corresponding com- ponent of the center-of-mass momentum is conserved. the angular momentum along the rotation axis is conserved. and thus the center-of-mass momentum is fully conserved.

because exploiting conservation laws is very helpful in solving problems. To keep it simple. dt q + dt ∆q) = L(q. Such conserved quantities are of central importance for both theoretical and practical reasons. start with a Lagrange function of a single generalized coordinate. Theoretical. if the coordinate q is cyclic. i. i.4 Noether’s theorem The previous observations can be generalized in the form of Noether’s theorem. The Lagrange function shall now be invariant under the transformation q → q + ∆q. space.8. and that the corresponding generalized momentum is conserved. ∂dt qi . This must be true for any variations of q. dt q) + ∆q + (dt ∆q) = L(q. e. dt q). this implies qi′ = qi + ∆qi   ∂L dt ∆qi = 0. ∂q ∂dt q ∂dt q ∂dt q where in the last step the Euler-Lagrange equations have been used.78 4. ∂q ∂dt q But this implies       ∂L ∂L ∂L ∂L 0= ∆q + dt ∆q − dt ∆q = dt ∆q . and rotation. Practical. the invariance implies a conservation law. e. as they amounted to a continuous shift in time. L(q + ∆q. It is one of the central reasons why symmetries are so important in physics. Noether’s theorem boils down to the fact that if the Lagrange function is invariant under a variation of a generalized coordinate. The previous cases were special examples of this. entails the existence of a conservation law. and thus especially for time-independent ones. this just reproduces what was known before. Of course. dt q). a symmetry where a generalized coordinate is change by a continuous quantity. dt q). Conservation laws 4. It shows that for a holonom and conservative system any continuous symmetry.8. where ∆q is taken to be infinitesimal. Essentially. But in general. yielding ∂L dt =0 ∂dt q Thus. conserved quantities define the properties of a system. L(q. the variation can be used to derive a conserved quantity. Expanding the left-hand side to first order in a Taylor series yields ∂L ∂L L(q.

it involves now how a generalized coordinate reacts to a change. Lagrangian mechanics 79 This does not look very helpful yet. 4. But rewrite ∆qi = ǫfi . but nonetheless very helpful statements. Such an invari- ance is also called a symmetry.5 Virial’s theorem The conservation laws treated so far hold locally in time. These expressions may now be more complicated than just cyclic coordinates. conserved quantities are linked to invariances of a system. this will be an average over a time τ . and will be encountered throughout all of physics. Zt+τ Zt+τ  1 1 dG 1 (G(τ + t) − G(t)) = dt = 2T + ~ri F~i dt.2). In this way. e. the left-hand side can be made arbitrarily . where the fi are some arbitrary functions. by making τ large. the aim is to make a statement about quantities on the average. i. it also often possible to make weaker. whether using the coordinate q or q ′ .Chapter 4. as the system looks the same.8. they are fulfilled at every time t. where T is the kinetic energy. Thus. This is a very fundamental insight that symmetries are connected to conservation laws. This will become very important when deriving thermodynamics as a limit of the mechanics of many particles. Using Newton’s law (2. Then this is equivalent to   ∂L ∂qi′ dt = 0. τ τ dt τ t t If the system is finite in the sense that all positions and momenta stay finite. ∂dt qi ∂ǫ i. Here. e. then G is bounded from above.. Consider for a system of particles with ordinary coordinates the quantity G = p~i~ri . In Virial’s theorem. Such statements are usually describing some average properties of a system. and ǫ is infinitesimal. its total derivative is given by dt G = (dt~ri )~pi + ~ri (dt p~i ) = m(dt~ri )2 + ~ri F~i = 2T + ~ri F~i . As announced. Though these are very powerful statements. so-called statistical mechanics. as a first example of such statements Virial’s theorem will be discussed. This average is obtained by integrating the time derivative of G over time.

This yields Virial’s theorem Zt+τ Zt+τ 1 T dt = − ~ri F~i dt. . Conservation laws small. Then ~ri F~i = r∂r Vi (r). it can even be made zero. by choosing τ to be the period. proportional to some power r n . then r∂r Vi (r) = nV (r). as investigated in section 2. For many cases also the following particular case of a conservative central potential is interesting. and Virial’s theorem takes the form Zt+τ Zt+τ n T dt = − V dt 2 t t and thus the temporal average of the potential energy has a fixed relation to the temporal average of the kinetic energy. where r is now the distance to the center of the potential. If V (r) is. For a periodic system.8. This statement is the starting point for many derivations in the aforementioned statistical mechanics access to thermo- dynamics. 2 t t where the right-hand side is known as the virial of Clausius.80 4.7. as often. While an explicit evaluation of either side still requires to know ~r(t) already knowledge of this relation often is sufficient to answer relevant questions about a system for which the actual trajectories are not important to know.

e. the dn − N-dimensional space of the independent. Consider the solutions of the equations of motion. integrals.13. but also requires non-local aspects. i. generalized coordinates. which are solutions to differential 81 . Though for classical mechan- ics both a differential and an integral formulation yield the same result. The main reason for this is that quantum physics. e.1. it is helpful to perform the corresponding conceptual formation already in the otherwise much simpler classical mechanics. it is not yet the best formulation to extend it to (non-relativistic) quantum physics. respectively.Chapter 5 Hamiltonian mechanics So far. most of the following will apply to the most relevant case of conservative systems with holonom constraints. i. 5. Unless otherwise noticed. introduces correlations which are not local. Any curve inside this space parametrized by a parameter t.1 Hamilton’s principle 5. It will also help eventually to demonstrate that for many cases it is always possible to find generalized coordinates such that the system becomes that of almost trivial mechanical systems in section 5.1 Integral formulation The basic entities to formulate the following are constructed from the configuration space. the formulation of mechanics has been by use of differential equations. i. Though this approach is extremely useful in mechanics. and also of great practical impor- tance. is a trajectory describing the movement of the particles of the system. especially Lagrange’s equation of the first kind and second kind. not describable by differential equations.12). especially in combination with special relativity. e. (4. no matter whether it is a solution to the equation of motion.24) and (4.

2).1) t1 which is called the action. yielding a function only of time L(qi (t). However. . i. i. like with the circle. situation will be noted if occurring. e. This subset of configurations will be called M. Hamilton’s principle equations which are of second order in time1 . the trajectories are dense around the physical trajectories inside the configuration space.1. Otherwise in the following only a single physical trajectory will be relevant. t) = L(t). This is different from functions. by successive virtual displacements. e. holes in a generalized sense. It is now possible to evaluate the Lagrange function on any element of q(t) ∈ M. dt qi (t). and therefore require (2dn − N) initial conditions. as otherwise the constraints (in form of the initial conditions) would be violated. First of all. or more precisely functions. e. to the real numbers. (5. This unusual. but this is deceiving. g. Such a situation occurs if there are. i. Such an object has therefore a different name.. there can be no more independent initial conditions than those for all independent coordinates in the formulation using Newton’s law (2. qi (t1 ) and qi (t2 ). by construction. any other trajectories inside the set can equally well be used to define it2 . it will be required that these trajectories must be reachable from a trajectory fulfilling the equations of motions. but by no means irrelevant. δqi (t1 ) = δqi (t2 ) = 0. Secondly. This number depends on the choice q(t). e. it is always possible to select two times t1 and t2 and specify the initial conditions by the values of all generalized coordinates at those two times. all virtual displacements at the initial and final time have to vanish.82 5. which trajectory is chosen. the so-called physical trajectories. 2 The reason for talking about physical trajectories instead of a single trajectory is to include also cases in which there are two equivalent trajectories. e. Then it is possible to select among all possible trajectories in the configuration space those which are (twice) differentiable in time and satisfy the initial conditions of the problem at hand. 1 Note that this may appear different when. which map values of variables to the real numbers. they spend trivially the same time between start and end points at t1 and t2 . At any rate. In addition. and it is formally possible to integrate this function over time to yield a number Zt2 S[q] = dtL(t). All such trajectories share a number of characteristics. g. It therefore maps a function. once this definition is made. a speed becomes a generalized coordinates. Though in practice it may be awkward. for moving in both possible directions from one point of the circle to the opposite point. These trajectories are not deformable into each other. and is called a functional.

To introduce the concepts.2 Variational formulation The alternative formulation uses again the concept of virtual displacements.Chapter 5. Hamiltonian mechanics 83 After these preparations. This is very different from the Euler-Lagrange equations or Newton’s law. There are actually two different ways to postulate it. which is a wide field. 5. it will be shown later that this yields the same results as the Euler-Lagrange equations in classical mechanics. This postulate already emphasizes the non-local nature of Hamilton’s principle. Furthermore. It seems to be odd to define the solution using a manifold which was defined using the solution. visible: The action does not depend on the coordinate system in configuration space. It requires that the physical path is the one where the change of S under virtual displacements of the path in configuration space. not accidentally. q → q + δq. reminiscent of the way how extrema of functions are found. a functional F depending only on a single function f depending in turn on a variable y. vanishes. which only needed to know about the function and its derivatives at one point to build the solution. but the whole set M. The first version is that the physical trajectory is the one trajectory in M. It is therefore a coordinate-independent classical mechanics. it will be required that the values of y at . which itself depends on a single parameter x. which extremalizes the action. however. only those parts will be discussed which will be needed to calculate δS. which plays an important role in classical as well as quantum field theory. expressed as δS = 0. To know the solution requires not only to know the possible trajectories at all times. but also this will now be seen. it is possible to postulate the principle of Hamilton. it is best to consider a simplified case.1. it is not yet obvious that this formulation is equivalent to the integral version. Here. However. since it is entirely formulated on paths. Again. One advantage is. This will amount to the statement that an extremal trajectory has zero variation of the action. which is. therefore eliminating the human-made coor- dinate systems. Of course. While this seems to be a very compact notation. it will turn out that it is in general much simpler to specify M even without knowing the actual physical trajectory than to solve the system. That this is still the more powerful formulation is hence far from obvious. it is mathematically far from obvious what it means: How is the variation of a functional calculated? Understanding this is the first part of functional calculus.

84 5.1. Hamilton’s principle

the two points x1 and x2 are the same for all functions y. The functional F is then defined
F [y(x)] = dxf (y, dxy, x).

Note that this is a very special class of functionals, a so-called linear functional. It would
be equally well possible to have other forms of the functionals, but these will be of no
interest in this lecture.
It will be necessary that both y and f are continuously differentiable functions3 .
As the question has already been formulate in this way, the task is now to find a criterion
for the extremalization of F as a functional of y. To create a formally well-defined way,
split the function y in two parts,

y(x) = ys (x) + αη(x),

where ys is the (yet unknown) solution searched for, α is a new, continuous parameter,
η is an arbitrary, sufficiently often differentiable function with η(x1 ) = η(x2 ) = 0. Thus,
this is a completely equivalent way of rewriting the class of functions y.
Any infinitesimal change4 of the curve y around the solution curve ys can therefore be
replicated by taking α infinitesimal small, denoted as δα, and some suitable choice of η,


δy = y − ys = ηδα = δα.

it is possible to expand f in a Taylor series to first order in δα   ∂f ∂y ∂f ∂dx y f (y. and this will be necessary in quantum physics.α=0 The corresponding variation of the functional is then defined as Zx2 δF = F (y) − F (ys ) = dx (f (y. dxy. x) + δα η+ dx η . dx ys . x) − f (ys . .2) x1 Since it is assumed that δα is infinitesimal. dx ys . dxy. This is beyond the scope of this lecture. 4 An equivalent way of doing so is to take η as some arbitrary function of α. dx ys . x). x)) . dx ys . x) + δα + ∂y ∂α ∂dx y ∂α   ∂f ∂f = f (ys . αη(x) = γ(α. it is possible to extend the concept beyond these limitations. x) ≈ f (ys . (5. x) + δαdαf = f (ys . ∂y ∂dx y 3 In fact. and taking only the leading term in a Taylor series in γ.

2) yields Zx2   ∂f ∂f δF = δα dx η+ dx η ∂y ∂dx y x1 Dividing by δα now defines a functional derivative of F . Hamiltonian mechanics 85 Inserting this into (5. Performing on the right-hand-side further a partial integration on the second term yields Zx2   . which for this particular functional yields the right-hand-side.Chapter 5.

x δF ∂f ∂f ∂f .


2 = dxη − dx + η . δα ∂y ∂dx y ∂y .

e. this problem introduces the very important concept of a geodesic.4 already δS = 0 is sufficient general. Therefore. any small changes in y does not change the value of the functional. Since η is arbitrary. this implies that the trajectory y which satisfies the Euler equation extremalizes F . this implies δF ∂f ∂f =0⇔ − dx = 0. . Also. In practice. The probably simplest problem is the minimal distance between two points on a plane. it is best to study some examples of the procedure. This is already hinted at by the fact that the Euler equations of a general functional become for the action the Euler-Lagrange equations. or creates at least a turning point. However. as η vanishes at x1 and x2 .1. there may be more than one extrema or turning point in general. It is thus equivalent to the integral formulation. The function y which satisfies the Euler equation is therefore the one at which the functional F is stationary. That this has the same form as a single variable-version of the Euler-Lagrange equations for f = −L is already indicating the connection. Since F maps into the ordinary numbers and is. While this will comparatively not be mathematically too challenging. so any case will do. by assumption. Turning points play almost never a role. e. a continu- ous curve in the trajectories. i. any solution to the Euler-Lagrange equations will yield δS = 0. some to a maximum. most mechanical problems lead to minima of the action. δα ∂y ∂dx y where the equation on f is called the Euler equation. which of these conditions is true requires to calculate the second functional derivative of S. i. a small change in the value of y also creates only a small change in F . However.x1 x1 but the last term vanishes. as will be seen in the section 5. Before doing so.

With the Minkowski metric of chapter 3. r dy c2 = = a. and f is given by s  2 dy f (y.1. dxy. x1 . after resolving for the derivative. Hamilton’s principle The infinitesimal (Euclidean) distance between two points is given by p ds = dx2 + dy 2. Since the problem is rotationally symmetric. and the shortest connection is therefore generally called a geodesic. Another example. However. this is so because the metric is Euclidean. Choosing x. this problem can be parameterized either using x or y. dx The Euler equation is thus   dy d ∂f ∂f d  dx − = q =0 dx ∂dx y ∂y dx  dy 2 1+ dx It is thus possible to integrate the equation once. yielding. Therefore. this time with further practical use. This is the shortest distance between two points in an Euclidean plane. is the question which is the minimal surface of a flexible object rotating around the z-axis.86 5. consider the total surface Zx2 p A[y] = 2π x 1 + dx ydx. just a straight line. the shortest connection will be different for different metrics. dx 1 − c2 where c is an integration constant. x) = 1 + . The solution is thus y = ax + b. the result would be different. the surface of a small strip is given geometrically by p dA = 2πx × 1 + dx ydx To find a formulation as an extremal value problem. and the total distance between two points is thus s  2 Z Z dy S = ds = 1+ dx dx The requirement that this should be the minimal length implies that S now plays the role of F in the general setup.

yielding Zt2   ∂L ∂L dt dt − − λj fji δqi = 0 ∂dt qi ∂qi t1 .7 it is possible to add a zero created from the sum of the constraints multiplied by Lagrange multiplicators. This is fortunate. Then in the remainder it is only necessary to sum over all i. 1 + (dx y)2 since y is cyclic. while the rest remains form-invariant. and therefore to lower the condition of non-holonom constraints and the requirement of only a generalized potential rather than a conservative system. Hamiltonian mechanics 87 The corresponding Euler equation is xdx y dx p − 0 = 0. justifying their name of Euler-Lagrange equations. a Reformulating with initial conditions y(x1 ) = y1 and y(x2 ) = y2 instead of a and b is. not directly possible.2 to multiple variables yi (or qi ). ∂dt qi ∂qi t1 It is not possible to proceed as before. since the variations δqi are no longer independent. In the same way as in section 4. Integrating with an integration constant a and solving the equation for dx y yields dy a =√ . thereby recovering Lagrange equations of the second kind. together with a second constant of integration. and thus with the same parameter. It is more complicated to show the equivalence to Lagrange’s equation of the first kind. and this allows to transform the problem into a first- order differential equation. however.Chapter 5. 5. but different functions. This is essentially done by rewriting these functions as yi = ysi + αηi .1.3 Equivalence to Lagrange’s function of the first kind The equivalence of Hamilton’s principle to Lagrange’s equation of the second kind is obtained by generalizing the derivation of 5. x y(x) = a cosh−1 + b. dx x − a2 2 which can be elementary integrated to yield. Most steps can actually be performed in the same way.1. up to Zt2   ∂L ∂L dt dt − δqi = 0.

4). X  2 ~ mi dt ~ri − Ki δ~ri = 0. ~ i δ~ri = Qi δqi = − ∂V δqi = −δV. (5. Thus 1 δ~ri d2t ~ri = dt (~ri δ~ri ) − (dt~ri )δdt~ri = dt (δ~ri dt~ri ) − δ((dt~ri )2 ). even with just the ordinary action S - the additional part is only used to resolve the residual dependencies. 5. this is still the action. In the same way as in section 4.3) after inserting (5.4) 2 It is perfectly permissible to integrate (5. t1 t1 t1 . These are again Lagrange’s equation of the first kind.88 5. Transforming the remainder of (5.1.5) i 2 t1 The first term is a total derivative and can be integrated directly. yielding Zt2 X  mi ~ i δri 0= dt dt (mi δ~ri dt~ri ) − δ((dt~ri )2 ) − K (5.7 it is possible to use the Lagrange multiplicators to eliminate the part of the sum which contains dependent variables. and therefore are (continuously differentiable) functions of time as well. and therefore Hamilton’s principle is still valid: Extremalize this expression with respect to the paths. ∂dt qi ∂qi together with the determination equations for the Lagrange multiplicators and the un- resolved constraints.1. as the result is proportional to δ~ri . The remainder yields then as Euler equations ∂L ∂L dt − − λj fji = 0. Hamilton’s principle Since the added expression is zero. K ∂qi Combining yields Zt2 Zt2 Zt2 0= δ(T − V )dt = δ (T − V )dt = δ Ldt = δS. the virtual displacements them- selves can be different at any time. (5.4). Furthermore. which are therefore also equivalent to Hamilton’s principle. for a holonom conservative system. it vanishes. However.5) to generalized coordinates yields that the first term is δT .4 Equivalence to d’Alembert’s principle To show that Hamilton’s principle is indeed equivalent to d’Alembert’s principle it is best to start from (4. for the variation of products the same product rule holds as for ordinary differential. The second term is.3) i Though no time elapses during a virtual displacement. which by construction vanishes at t1 and t2 .

or cases in which at least a generalized potential existed. also here the result simplifies back to the original Euler-Lagrange equations for conservative systems. in this case K Thus. It is possible to extend it even further to include non- conservative cases.6) by a single prescription.6) t1 which is the sum of the kinetic energy and the scalar projected forces. Thus. but with the present integrand rather than the Lagrange function. 5. By virtue of the arguments in section 4. Hamilton’s principle is therefore also equivalent to Newton’s laws. . every step is invertible. Though this was done in forward direction. in the form of (5. starting at d’Alembert’s principle and then deriving Hamilton’s principle from it. in the end it requires again to solve some kind of Lagrange’s equation. define the action as Zt2 S[q] = ~ i~ri ) dt(T + K (5. Especially. make the action stationary.1. i.3. Hamilton’s principle is in all forms equivalent to Lagrange’s equations.2 is independent of the mechanics context this only requires to determine the Euler equations for this action. Thus. If Hamilton’s principle should also be true for the non-conservative systems described by this action. ~ i~ri = −V . assuming holonom constraints. and therefore a well-defined quantity for any mechanical system. Since the technology developed in section 5.3 can be used. As in that case. Though it is has the appeal that all situations are no completely covered. though satisfying on conceptual grounds. Hamiltonian mechanics 89 and thus Hamilton’s principle in variational form. nothing has yet been gained on practical grounds. e.Chapter 5.4. For this purpose.1. Hamilton’s principle is as good a starting point to formulate classical mechanics as is d’Alembert’s principle. This yields ∂T ∂T 0 = dt − − Qi ∂dt qi ∂qi Qi ~ j ∂~rj = K ∂qi and therefore coincides with the results of section 4. it is necessary that the stationary points of this action yield the equations of motion.4.5 Beyond conservative forces So far Hamilton’s principle only covered conservative cases. If the constraints are non-holonom the same procedure as in section 5.1.

The reason is that cyclic coordinates imply conserved generalized momenta. This is true for the effective low- speed system of classical mechanics but for the special relativity version. Thus. as no such simple relation exists for the generalized speeds. it may seem natural to use the generalized coordinates and speeds for this. By introducing generalized coordinates using N holonom constraints. Thus. it is called the phase space.2 Phase space and state space Though the concept of trajectories was extremely useful in reformulating mechanics in a more compact way. There. but also the final state of a system. The paths of all particles are uniquely described by trajectories in this space. the time plays a different role than the other coordinates in classical mechanics: Here it parametrizes the trajectories of particle in configuration space.18). Formulating the problem thus in these two quantities will make the whole problem trivial for these directions. the aim in the following is to reformulate the system in terms of the generalized coordinates and speeds. this role is played rather by the eigentime. which both differ by the metric. 5 Systems like fluids without individual particles will be discussed in more detail in section 6. In the relativistic case.2. Phase space and state space 5. which has dn coordinates. However. This can be upgraded to space-time with dn + 1 dimensions by adding time. It is therefore useful to reformulate Hamilton’s principle as an initial value problem. where d is the number of coordinates and n is the number of involved particles5 .90 5. This is especially true when it comes to problems which can no longer be solved in a closed. in practice this is often quite different. If time is included to characterize the progression along the trajectories. And this is true for almost all systems of interest. This is the configuration space. But. thus forming again a space of the same dimensionality. it is usually only known how the system starts. There are again 2(dn − N) of these. this becomes the S + 1-dimensional event space: Every event is localized at one point in this space. it has one essential drawback: It requires to know not only the initial state of a system. . being either Euclidean or Minkowski. it is much more useful to rather use the generalized coordinates and momenta. Though from a mathematical point of view there is no difference between this requirement and the requirement of knowing position and speed at the initial time.3. requires an individual time for every particle. the number of coordinates is reduced to S = dn − N. It is quite useful to reiterate the various concepts of spaces encountered so far. To separate this space from the configuration space before. in fact. and the question is how it will develop with time. they are advantageous. analytic form. the latter as defined in (4. At first sight. The basic space is the coordinate space.

that there are rare multi-particle systems which are described by higher-order differential equations. However. The fact that Newton’s law is of second order is nothing which can be derived. and so on. Lagrange’s equation are not yet of this type. determining the point in state space at which a system is located. The position along the trajectories can be parametrized by the time.Chapter 5. There exists a general mathematical theorem which guarantees that no more initial conditions are needed to uniquely identify a trajectory. a state is called the minimal collection of information necessary to fully describe a system. both in ordinary space-time as well as in event space it is not sufficient to specify a point to specify a system completely. but. Note. a state space which gives uniquely defined trajectories would be of mS + 1 dimensions. as noted in section 2. . Thus. in thermodynamics quantities like temperature and pressure may be part of the minimal information defining a state. a 2S-dimensional space is obtained. Even in event space. in these cases this is also only an effective rewriting of initial conditions of multiple particles in the form of a single effective particle. Returning to classical mechanics.4. trajectories may cross. derived from observations. Also. as they are second order in time. this conversely implies that measuring all the minimal information. which ultimately underlies any classical mechanics problem. any other minimal way is equally good. this concept can be extended to any physical system. however. and three initial conditions would be required. Though the state space has been used to define what a state is in the sense of classical mechanics. 6 That this is sufficient is because Newton’s law (2.2. By doubling the degrees of freedom by adding the generalized momenta. this 2S + 1-dimensional space is called the state space7 . Since a single point is sufficient to fully describe a state this implies that a first-order set of 2S differential equations in time should exist which describes the evolution of a state. This will no longer be possible in quantum physics. e. the position and future movements of any particle of a system is uniquely defined when a position in this space is given. is a differential equation of second order in time. Only by adding generalized speeds the trajectory of a given particle is uniquely determined6 .2). Conversely. if Newton’s law would be rather of mth-order in time. g. Hamiltonian mechanics 91 However. i. leading ultimately to Hamilton’s equations in section 5. It thus fully describes the state of a sys- tem. 8 This is the reason why classical mechanics is a deterministic theory. If Newton’s law would be a third-order differential equation. 7 Consequently. this would no longer be true. is a postulate.. This problem is resolved by the introduction of phase space. This will be done in the following. gives all information about a classical mechanics system8 . E. Thus. Tra- jectories do no longer cross.

the question is how to obtain a function g(u). but now a function of u. In addition to making the procedure more simple for mechanics. dx where the new variable u will play later the role of the generalized momenta. e. consider some function f (x). Since it appears reasonable that the transformation should be invertible. Legendre transformations 5. and will play. As all steps can be performed in reverse order. which is equivalent to f . Note that if d2x f = 0 vanishes. this transformation is reversible. the same approach can be used for a wide range of problems. But. The function u f (x) = αx2 ⇒ u = 2αx ⇒ x = 2α . This procedure is called Legendre transformation. It is noteworthy that just inserting x(u) into f would not be. At first sight. it is possible to invert the various relations to obtain. That is best seen when considering an example. solves the problem. in principle. necessary.3. the searched for function g(u). say. a Legendre transformation would not be possible. is to still use df /dx = u to reexpress x as a function of u. What is. this problem can be put into a larger context and a general recipe for this type of transformations can be given. du Furthermore. consider that g is a solution to dg = ±x. g. df = udx = d(ux) − xdu and therefore d(f − ux) d(ux) − xdu − (dux) = = −x. of course. Since the generalized momenta are known as function of the generalized coordinates and speeds. this seems to be still a function of two variables. Having this in mind. x will play the role of the generalized speeds and f will be the Lagrange function. To formulate it as generally as possible. the so-called Legendre transformed of f (x). The differential of f is df df = dx = udx.92 5. but merely a constant. again an important role in thermodynamics. since u would then not be a variable. the Lagrange function as a function of the new quantities.7) du du Thus. is given by df g(u) = f (x) − ux = f (x) − x . (5.3 Legendre transformations This is not entirely straightforward. Later. dx which.

2α 4α while the Legendre transform is 1 2 u2 u2 g(u) = f (x(u)) − x(u)u = u − =− . but also its derivative. On the other hand. as can be seen also from (5. the result is independent of c. f (x) = g(u(x)) + u(x)x = f (x) − u(x)x + u(x)x. 4α 2α 4α which depends on c. 4α 2α 4α Consider now the function u f ′ (x) = α(x + c)2 ⇒ u = 2α(x + c) ⇒ x = − c. it is useful to consider this more general case. To see that the reverse transformation is indeed unique. This is seen by considering the full case. The reversal of the sign is required to have the corresponding forward and backward transformation. y). The additional term.Chapter 5. u −x = du g ′ (u) = − + c ⇒ u = 2α(x + c) 2α f ′ (x) = g ′ (u(x)) + xu(x) = α(x + c)2 . consider first the example. This type of transformation is therefore not unam- biguously invertible. Especially. Hamiltonian mechanics 93 Performing an insertion yields  u 2 1 2 f (x(u)) = α = u. the Legendre transformation is u2 u  u2 g ′ (u) = −u −c = − + uc. This shift compen- sates for the fact that for the inverse transformation is not only necessary to know how the function is modified. The question may arise. Since in the relevant case the function f also depends on more variables. which should not change. and therefore a continuously infinite number of functions are by this all mapped to the same result.7) ensures that not only the function but also its differential is consistently transformed to the new variables. the differential takes the form . 2α and thus ′ u 2 u2 f (x(u)) = α −c+c = . If now y plays the role of the additional (mute) general coordinate in f (x. what the real significance of the addition is. 2α 4α This coincides with the previous case.


∂f .


∂f .


y)dx + v(x. df = dx + dy = u(x. ∂x . y)dy.

y ∂y .

x .

4. y) has then the differential . Hamilton’s equations The Legendre transform g(u.94 5.


∂g .


∂g .


y)dy. ∂u . y)du + v(x(y. dg = − du + dy = −x(u. u).

y ∂y .

u Since df = vdy + udx = vdy + d(ux) − xdu implies .

∂(f − ux) .


∂y .

= v .

u ∂(f − ux) .


∂u .

5. After transforming the Lagrange function. It will be the Hamilton function which will take over the role of the central quantity from the Lagrange function for most of the rest of this lecture9 . = −x x the desired Legendre transform is given by ∂f g(u. what the equivalent reformulation of the Euler- Lagrange equations in phase space are. the first step is to apply the Legendre transformation to it. this is different in relativistic systems. pj ). in the final result the non-transformed variable plays no active role. y). pi (qj . the next interesting question is how this affects the equations of motion. 9 As discussed previously. except that the new variable may depend on it. Though.25). It is Hamilton function (4. this function is already known. . Using the generalized momenta ∂L pi = ∂dt qi yields H(qi .8) ∂x Thus. with the total energy H = T + V . at that time the generalized momenta was considered as a function of the generalized coordinates and speeds. (5. especially. However. pi . t). holonom system.4 Hamilton’s equations As Lagrange’s function is still the basic dynamical quantity. t) = pi dt qi (qj . y) = f − x . which coin- cides under certain conditions. pj ) − L(qi . y) − ux(u. y) = f (x(u. homogeneous kinetic energy of order 2 and a conservative.

The equations (5. This can be further simplified by noting that the Euler-Lagrange equations imply ∂L ∂L d t pi = d t =− ∂dt qi ∂qi and therefore ∂L dH = dt qi dpi − dqi dt pi − dt. (5. it is a useful first step to consider the total differential of the Hamilton function. as it is not an equation involving a separate variable.12) are first order in time.2. ∂qi ∂dt qi ∂t ∂qi ∂t where the terms involving ddt qi have canceled because of the definition of the generalized momenta. first-order equations are usually easier to solve. They are also called canonical equations. . 10 Note that ordinary differential equations of second order can always be rewritten as a set of twice as much first-order differential equations. rather than the second-order differential equations of Lagrange. The disadvantage is that the number of equations of motion to be solved has doubled. (5.9) ∂pi ∂qi ∂t ∂L ∂L ∂L ∂L ∂L = dt qi dpi + pi ddt qi − dqi + ddt qi − dt = dpi dt qi − dqi − dt. In practice.9) and (5. but rather a consistency condition. and therefore Hamilton’s equations are technically a step forward. The result when solving these equations is then no longer a trajectory in configuration space.Chapter 5. but rather a trajectory in phase space. The last equation is actually not a real equation of motion.12) ∂qi ∂L ∂H − = . The result is therefore not trivial on mathematical grounds. This is not necessarily so for partial differential equations as the equations of motion are.11) ∂pi ∂H d t pi = − (5. as was aimed at in section 5.13) ∂t ∂t which are the new equations of motion determining the dynamics as the Euler-Lagrange equations did before. The most remarkable result is that Hamilton’s equations are now first-order differential equations.10) yields Hamilton’s equations ∂H dt qi = (5. and are therefore describing the evolution of a state in state space with knowledge of a single point. but in practice this is still better10 . Hamiltonian mechanics 95 To find them.11-5. ∂H ∂H ∂H dH = dpi + dqi + dt (5.10) ∂t Comparison of (5.

dt qn . under certain conditions discussed in section 4. the time-dependencies of the qi and pi have to cancel each other inside the Hamilton function in this case.. but depends for s cyclic coordinates only on the q1 . of the Hamilton formalism is that it is more complicated in the relativistic case. The resulting function is then still of the Lagrange type. The disadvantage. and therefore H = 2T − T + V = T + V is the total energy. Note that (5. the energy is in this case conserved. because of (5. Furthermore.4.. Though this is the most relevant case. Therefore. This rather simple relation to the Lagrange function L = T − V comes about as this is true if the kinetic energy is at most quadratic in the speeds. then the correspond- ing coordinate is constant as well. Similar as there. To facilitate the . It is quite useful to study the procedure from the original system up to Hamilton’s equations and the solution of Hamilton’s equations for an example. as a compromise. This implies that if the Hamilton function is not explicitly depending on time. this is by far not always the case. Hamilton’s equations of motions (5. and therefore the existence of a cyclic coordinate does not immediately solve any of the equations of motion. it will not be developed here further. however. H = T + V ..q. In case of a cyclic coordinate this also implies that one of Hamilton’s equations of motion is solved trivially. it is just the total energy of the system.96 5.12) have been used. and thus creates only n − s second-order differential equations and 2s first-order differential equations. as the Lagrange function depends on the general- ized speeds rather than the generalized momenta. Thus. and only in very special cases in classical mechanics. much rarer then that of a cyclic coordinate. Hamilton’s equations Though in general the Hamilton function is the Legendre transform of the Lagrange function. . Then the first term in the Legendre transformation taking care of the differential properties is 2T . and actually constant as a function of time.. stated without proof here.11) implies that if a generalized momentum is cyclic.n and the n − s generalized speeds dt qs+1 . This case is. The actual value of the corresponding generalized coordinate is then determined by the initial conditions. ∂t H = 0. This is again a practical advantage of the Hamilton formalism over the Lagrange formalism.8. As this formalism plays essentially no role in quantum physics. it can be shown that ∂H ∂H ∂H ∂H ∂H ∂H dt H = dt qi + dt pi + ∂t H = − + ∂t H = ∂t H.11-5. ∂qi ∂pi ∂qi ∂pi ∂pi ∂qi Here.12) any cyclic coordinate in the Lagrange function is neces- sarily also cyclic in the Hamilton function.. the so-called Routh formalism performs the Legendre transformation only for cyclic coordinates. This is quite different from the Lagrange case. If the Hamilton function coincides with the energy. . it is a conserved quantity.1.

though Hamilton’s formalism is overkill for it. The kinetic and potential energy of this conservative system is m T = (dt q)2 2 k 2 V = q . p)) = − p + q = p + q = T + V. 2 This yields the Lagrange function m k L= T −V = (dt q)2 − q 2 .11-5. However. 2 2 in which no coordinate is cyclic.Chapter 5. being ∂H p dt q = ∂p = m dt p = − ∂H ∂q = −kq This set of equations can be solved in multiple ways. m 2m 2 2m 2 which coincides with the conserved energy. yielding again a second-order differential equation. In this case. k m with the two integration constants.12). or by inserting one equation into the other. This could be used to eliminate either q or p. The total energy is     kq02 2 k kq02 2 k H=E= cos t + φ0 + sin t + φ0 = kq02 . But it is more instructive to use Hamilton’s equations (5. dt q(q. it is possible to replace p or q by using the conserved total energy. there is a single generalized coordinate q = x describing the position of the oscillator. ∂dt q The Legendre transformation yields the Hamilton function p2 1 2 k 2 1 2 k 2 H = pdt q − L(q. the solutions are   k q(t) = q0 sin t + φ0 m   mq0 k p(t) = cos t + φ0 . and therefore the energy will be conserved. 2 m 2 m . At any rate. this will be the harmonic oscillator. The generalized momentum is ∂L p= = mdt q. it does not depend explicitly on the time. Hamiltonian mechanics 97 comparison to previous results. Either.

5. It is therefore reasonable to start by reformulating the action in phase space. p] = dt(pi dt qi (qj . it is possible to reformulate the action principles of section 5.98 5.5. Likewise. where the relative size of the major axes depends on the initial conditions as well. To do so. Still. as long as the variations of the generalized coordinates and speeds vanish. Performing the variation as in 5. 5. This does not necessarily imply that any variation of the generalized momenta do vanish at t1 and t2 .1) is replaced by the Hamilton function using the inverse Legendre transformation.2 yields Zt2   ∂H ∂H δS = δα dt ωi dt qi + pi dt ηi − ηi − ωi . modification is to change from a formulation in terms of tra- jectories in configuration space to one in phase space. ∂qi ∂pi t1 . Such a closed path in phase space is typical for periodic systems. The question is thus. this is called an aperiodic system. Each of the following reformulation puts emphasis on a different aspect of a mechanical system. If the path is not closed.5 Action principles With the introduction of the phase space. what is the characteristic property of a phase space trajectory to be the solution of a mechanical problem. t1 which is thus a functional of the phase space trajectory.1. pi (t) = pi (t) + αωi (t) with ω(t1) and ω(t2) not constrained in general. but both the coordinates as well as the generalized momenta are bounded. In the configuration space case it was that it extremalized the action. pj ) − H(qk . Thus. the variation under a change of the trajectory in configuration space must vanish. depending on the knowledge about a system.5. pk )).1) only required that the variations of qi and dt qi vanish at t1 and t2 . yielding Zt2 S[q. They therefore center the discussion on different concepts. different of these principles may be useful in practical applications.1 Modified Hamilton’s principle The first.1. Action principles Thus. The trajectory in phase space is a closed ellipsoid. An important insight is that the initial conditions in (5. the energy is completely determined by the initial conditions. perhaps obvious. the Lagrange function in (5.

Chapter 5.14) t1 and considers the following type of trajectory variation: Rather than keeping the time constant for all admissible trajectories. 5. this principle requires that the Hamilton function has the same value for every trajectory. or none of these variations. either. it operates on a different concept of variations. t → t + ατ . it is necessary to also consider that the time changes for a trajectory variation. e. Also. this will only vanish if Hamilton’s equations (5. While it is in spirit similar to Hamilton’s principles. and the symbol ∆ is used to make the distinction between the two types of variations explicit. It is applicable for conservative systems only. start at the same points and end at the same points. this type of variation no longer . To make this sensible.11-5.5. i. δα ∂pi ∂qi t1 Given that the variations η and ω are independent and arbitrary. this implies that the physical trajectories in phase space extremalize the action. Zt2 A= dtpi dt qi . Hamiltonian mechanics 99 Performing a partial integration Zt2 Zt2 Zt2 dtpi dt ηi = pi ηi |tt21 − dtηi dt pi = − dtηi dt pi t1 t1 t1 yields Zt2      δS ∂H ∂H 0= = dt ωi dt qi − − ηi dt pi + . This is the reason why it only makes sense for conservative systems.2 Maupertius’ action principle A different formulation is due to Maupertius. Due to the explicit occurrence of the time. a variation of a trajectory now takes the form ∆q = ∆α(η + τ dt q) = ∆α(δq + (dt q)∆t). Trajectories from the configuration space may be admissible for both. every trajectory can be considered to be parameterized by a parameter s. Conversely. (5. where t(s) + ατ (s) is a (strictly) monotonously increasing function of s. As a consequence. Thus. and thus systems in which the Hamilton function is a constant of motion.12) are fulfilled. Therefore. the physical trajectories in both configuration space and phase space extremalize the action. It is based on a different action formulation. the trajectories should be compatible with the initial conditions.

but not of fixed Hamilton function. However. (5. t1 t1 t1 t1 where it has been used that by construction H is constant along all admitted trajectories. their variations have to be calculated in this way. which are no longer fixed. the integrals are functionals of the trajectories. for an arbitrary function of the coordinates ∆f (q. as it does only so for trajectories of fixed time. Since a time derivative and a variation ∆ do not commute. t1 where the boundary term appears as the consequence of varying t1 and t2 . Especially.5. Varying now the action A yields  t  Zt2 Zt2 Z2 Zt2 ∆A = ∆ dtpi dt qi = ∆ dt(L+H) = ∆  dtL + H(t2 − t1 ) = ∆ dtL+H(∆t2 −∆t1 ). since neither p nor dt q are further constrained. using that this variation still commutes with time derivatives.100 5. yielding Zt2 Zt2   Zt2   ∂L ∂L ∂L dtδL = dt δqi + dt δqi = dtdt δqi ∂qi ∂dt qi ∂dt qi t1 t1 t1 . Action principles commutes with the total time derivative. It is therefore necessary to calculate it.15) ∆α ∂q ∂t ∂t ∂t Especially. This statement holds true for any such variation. Unfortunately. and therefore (5. This yields Zt2 ∆A = δS + (dt2 − dt1 ) dtL∆t + H(∆t2 − ∆t1 ) = δS + (L + H)∆t|tt21 . so neither does a time inte- gration and the variation. t) ∂f ∂f ∂f ∂f = ∆q + ∆t = (δq + (dt q)∆t) + ∆t = δf + (dt f )∆t. dt ∆q = ∆α(dt δq+dt ((dt q)∆t)) = ∆α(δdt q+(d2t q)∆t+dt qdt ∆t) 6= ∆α(δdt q+(d2t q)∆t) = ∆dt q.15) can be used. the first term does not vanish.

t1  .

t2 ∂L .

∂L ∂L .

= δqi .

= .

∆qi − (dt qi )∆t .


∂dt qi t2 ∂dt qi ∂dt qi t1 where in the third step the Euler-Lagrange equations and in the last step (5.15) has been used. ∆qi (t1 ) = ∆qi (t2 ) = 0. . Since the initial conditions remain valid. Recognizing the generalized momenta in the second term and combining everything yields ∆A = (L − pi dt qi + H) ∆t|tt21 = 0. . and the first term drops out.

t1 which must vanish. implying that the kinetic energy needs to be constant as well. (5. At the same time. and thus in a fixed medium fastest. and (general) relativity. as noted above. Thus. Consider a further specialization. it emphasizes that the selection criterion of trajectories still needs to be tied to some physical concept characteristic of classical mechanics. its important message is that the partition of trajectories in configuration space used before is not the only possibility to formulate classical mechanics. which implies constant speed dx/dt. the concept generalizes. The path of shortest .5. It is this principle which lends itself particularly well to generalization of an action principle to optics. It should be noted that in most of the remaining lecture the name action refers to (5. and thus which makes the traveling time extremal. This equivalence is also true for a force-free particle. and therefore Zt2 Zx2 dx 0=∆ dt = ∆ dx.16) dt t1 x1 and thus the requirement to extremalize the path. also known as the principle of soonest arrival. path.14) is. usually minimal. Hamiltonian mechanics 101 for any ∆t. Thus Zt2 ∆A = ∆ dtc = c∆(t2 − t1 ). due to 2T = H + L = pi dt qi = c constant itself. every trajectory satisfying the Hamilton principle will also satisfy this minimal action principle. This implies that the integration kernel of (5.3 Fermat’s principle Another reformulation of the action is the one of Fermat.1) if not remarked otherwise. This is Fermat’s principle.Chapter 5. The following is therefore again of particular interest only for its conceptual implications. In geometrical optics it is the same underlying principle which requires the light to take the shortest. since this is just the definition of the Hamilton function. Though trivial in mechanics. quantum physics. as it is a very special case. This implies that the trajectory is chosen which extremalizes the variation of the time difference. usually making it as short as possible. in which V is also constant. in classical mechanics it is less useful. 5. While this formulation of the action principle is not outright useful. However.

which is a solution to a given problem.102 5.14). This reformulation of the action principle is known as Jacobi’s principle. This reformulates the action as Zt2 Zt2 Zρ2 √ Zρ2 dt 2T √ 0 = ∆ dt2T = ∆ 2T dρ = ∆ 2T ρ=∆ E − V dρ. If V = 0. but still for a conservative system Zt2 Zt2 Zt2 0=∆ dtpi dt qi = ∆ dt(H + L) = ∆ dt2T t1 t1 t1 applies. the energy is constant.4 Jacobi’s principle A further important concept is introduced when any explicit appearance of the time is eliminated from the action principle. ρ1 . and the principle turns into Zρ2 0=∆ dρ. the (diagonal) matrix of masses can be considered also as such a metric.5. dρ d t1 t1 ρ1 ρ1 The final expression does no longer involve the time. and is thus the searched-for expres- sion. This time the potential is not constant. This is best seen by multiplying the expression by dt2 . the generalized metric transforms a length element in coordinate space to configu- ration space. Start again from (5. however. 5. then X X ∂rji ∂rji X T = mi (dt rji )2 = mi (dt qk )(dt ql ) = µkl (dt qj )(dt ql ). If the constraints are skleronom. yielding X X (dr)2 = mi drji drji = µkl dqj dql = (dρ)2 ij kl Thus. i ijkl ∂qk ∂ql kl where the quantity X ∂rji ∂rji µjl = mi . Formally. Action principles length.5. better known as the configuration space metric. has therefore a special name: It is called again a geodesic. ij ∂qk ∂ql though being a function of the qi is sometimes called generalized mass. It is.

An intriguing example for the Jacobi principle is that of electron deflection. because the time no longer appears. as the potential is constant. To illustrate the metric. but where these potentials differ between both media. where it jumps. and finally end up in the second medium at position 3. q2 = θ. Since a single path is investigated. first consider the trivial case of the generalized coordinate being just the ordinary ones for a single particle. Now let the electron start somewhere in the first medium at the position 1. the potential is constant.   1 0 0   µ = m 0 r 2 0  0 0 r 2 sin2 θ and thus still a diagonal matrix. While the principle is useful in itself. hit the interface at some point 2.3 in terms of the time is extremely complicated but without is straightforward.Chapter 5. The Jacobi principle the requires Z3 p p 0 = ∆ 2m(E − V ) dx2 + dy 2 1 Z2 p p Z3 p p = ∆ 2m(E − V1 ) dx2 + dy 2 + ∆ 2m(E − V2 ) dx2 + dy 2 1 2 . ∂qk ∂ql and the metric is only a rescaled unit matrix. The metric is then the matrix. Since there are no forces in the two media acting. Consider two materials.7.21). q3 = φ. just remember planetary motions where the solution in section 2. the system moves along a geodesic in configuration space. it is the concept of the metric in configuration space which also deserves attention. One interesting consequence of this principle is that. Then ∂xi ∂xi µkl = m = mδik δil = mδkl . the ∆-type and δ-type variations again coincide. A little less trivial is the case where the generalized coordinates are the spherical coordinates (4. Hamiltonian mechanics 103 This is a reformulation of Fermat’s principle (5. The interface should be located along the x-axis at y = 0. except at the interface. using q1 = r. say V1 in medium 1 and V2 in medium 2. in which an electron moves under the influence of a constant (electrostatic) potential.19-4. the movement in the individual media is that of a free particle. it is always possible to put everything in the x-y-plane. Thus.16) in configuration space: In the absence of forces.

sin β E − V1 |v1 | and is determined by the electron’s speed in both media. rather than ordinary vectors. as noted. τ ). This does not incidentally look like the refraction law of optics. This yields p x2 p x3 − x2 (E − V1 ) p 2 − (E − V 2) p x2 + y12 (x3 − x2 )2 + y32 p p = (E − V1 ) sin α − (E − V2 ) sin β.6. they can be moved outside the integral.1 Formulation The basis for this will be Hamilton’s principle in its integral and variational formulation. (5. Then the free movement in the medias imply p q p q 0 = 2m(E − V1 )∆ x22 + y12 + 2m(E − V2 )∆ (x2 − x3 )2 + y32 . Therefore. L = L(xµ . To have a genuine relativistic formulation. .6 Covariant Hamiltonian mechanics With the action principles a point has been reached where the attempt can be made to give a formulation compatible with special relativity. the deflection of the electron at the surface is given by r sin α E − V2 |v2 | = = .9) will be shown afterwards. as generalized coordinates.6. Again. the best starting point is to start with a Lorentz scalar to encode the physics. uµ . Covariant Hamiltonian mechanics Since the energy and the potentials Vi are constant. As the Lagrange function has already been a scalar under rotation. Since. the formulation is postulated. as this can be derived in a very similar way from Jacobi’s principle applied to geometrical optics.15) implies  q q  p p 2 2 2 2 0= 2m(E − V1 )dx2 x2 + y1 + 2m(E − V2 )dx2 (x2 − x3 ) + y3 δx The variation δx is non-zero.104 5. and its equivalence to (3. 5. It then needs to depend on Lorentz vectors. where it was chosen that x1 = 0. it is natural to require that the Lagrange function becomes also a Lorentz scalar. and can therefore be divided by. 5. time plays no role in the variation.

6. which are now worldlines.1.17) τ1 which therefore acts again on trajectories. It is now postulated that a free particle of mass m is described by the action Z p Spp = −m dτ −∂τ X µ ∂ τ Xµ . and there- fore the arguments of section 5. (5. Especially. It is then possible to use the eigenzeit τ to define an action as Zτ2 S= L(xµ . Of course. it is possible to also introduce generalized coordinates or add more particles. τ )dτ. but this will not be done here.Chapter 5. and the variation with respect to the trajectories. uµ . (5. and thus there are now four components of the trajectory. Classically. Now there is the world-line with 4 functions Xµ (τ ) of the eigenzeit. uµ is its four-velocity. After all. and not withstanding its physical interpretation it is not part of the coordinates. the eigenzeit was introduced as a parameter describing the trajectory.18) dτ ∂uµ ∂xµ giving four partial differential equations for the four unknown quantities xµ (τ ).19) Performing the variation along the world line δ Ẋµ ≡ δ∂τ X µ = ∂τ δX µ . Hamilton’s principle is then defined in the same way as before: An integral over the Lagrange function. 5. Hamiltonian mechanics 105 where xµ is the position of the particle. relativistic form of Newton’s law. a trajectory is described by the 3 spatial coordinates xi (t) as a function of time t = x0 . To show that this is equivalent to the.2 Point particle Take a particle moving along its world line. and τ is the eigenzeit. this just introduces another variable. leading to the relativistic form of the Euler-Lagrange equation d ∂L ∂L − = 0. (5. From the point of view of mathematics. also postulated. start first with the case of a free particle. here also t(τ ). For the following it is important to note that the eigenzeit strictly monotonously increases along the world-line.2 can be repeated identically.

With τ the eigentime the action is indeed Poincare-invariant. Inserting this expression for the argument of the square root yields   ∂τ (Λµν X ν + aµ ) ∂τ Λµω Xω + aµ = Λµν Λµω ∂τ X ν ∂τ Xω . Covariant Hamiltonian mechanics yields the equation of motion as Z q r  !   δSpp = −m dτ −Ẋµ Ẋ µ − − Ẋ µ + δ Ẋ µ Ẋµ + δ Ẋµ Z q r  !  = −m dτ −Ẋµ Ẋ µ − − Ẋµ Ẋ µ + 2Ẋ µ δ Ẋµ  v ! Z q u u ρ Ẋ δ Ẋρ  = −m dτ  −Ẋµ Ẋ µ − t−Ẋµ Ẋ µ 1 + 2 Ẋν Ẋ ν Z Taylor Ẋ µ δ Ẋµ = m dτ p −Ẋ ν Ẋν where in the last line use has been made of the infinitesimal smallness of δ Ẋµ and the square root has been Taylor-expanded. Defining now the four-speed as Ẋ µ uµ = p −Ẋν Ẋ ν yields the equation of motion after imposing the vanishing of the action under the variation and a partial integration as mu̇µ = 0 This is nothing else then the equation of motion for a free relativistic particle. as the length of dτ xµ is a constant.3.106 5.6. This also justifies the interpretation of m as the rest mass of the particle. this makes the expression invariant. Since the expression in parenthesis is just δνω because of the (pseudo-)orthogonality of Lorentz transformations. This can be seen as follows. A Poincare transformation is given by X ′µ = Λµν X ν + aµ . as discussed in section 3. this shows the invariance of the total action. . which of course reduces to the one of Newton in the limit of small speeds. Note the fact that the division done just normalizes the four-speed. Since the eigentime is in- variant by definition.

The quantity γτ τ is called the world-line metric. For the functions follows then ′ dτ 1 Ẋ µ (τ ′ ) = Ẋ µ (τ ) ′ = Ẋ µ ′ dτ τ̇ Hence the scalar product changes as 1 µ Ẋ µ′ Ẋµ′ = Ẋ Ẋµ . it is also reparametrization invariant. Hamiltonian mechanics 107 Additionally. This is what the equivalence principle is about. τ̇ ′ yielding the transformation property of the integral measure. 2 η Under a reparametrization τ → τ ′ (τ ) it is defined that the functions X and η transform as X (τ ) = X (τ ′ (τ )) dτ 1 η ′ (τ ′ ) = η (τ ) ′ = ′ η (τ ) (5. this metric is only a single function γτ τ (τ ) of the eigentime. and the remaining one is then compensated by the integral measure. i. as it ought to be: Physics should be independent of the coordinate systems imposed by the observer. It is instructive to study this in more detail. e.20) dτ τ̇ .19) was rather tedious. it is possible to transform the eigentime to a different variable without changing the contents of the theory. τ̇ ′2 One power of τ̇ ′ is removed by the square root. This implies dτ ′ τ̇ ′ = dτ dτ ′ dτ = . For the moment. as it will be found to measure distances along the world-line.. consider the tetrad as an independent function in a new action defined as Z ! µ 1 Ẋ Ẋµ ′ Spp = dτ − ηm2 . which is defined as 1 η (τ ) = (−γτ τ (τ )) 2 . To show this invariance for the action (5. Since the world line is one-dimensional. This yields a trivial example of a so-called tetrad η. and it is useful to rewrite the action.Chapter 5.19) take an arbitrary (but invertible) repara- metrization τ ′ = f (τ ). For this purpose it is useful to introduce a metric along the world line. Showing this explicitly for the action (5.

m2 Thus knowledge of X determines η completely.108 5.7.9). and also makes the second expression invariant.18) to work.5 it was shown that the formulation using Lagrange’s equations is invariant of the particular chosen coordinates. The formulation of mechanics in terms of Hamilton’s principle gives this a conceptual more important interpretation. Using the Euler-Lagrange equation this yields d ∂L ∂L Ẋ µ Ẋµ 0= − = 2 + m2 dτ ∂ η̇ ∂η η Ẋ µ Ẋµ =⇒ η 2 = − .20) takes care of the extra factor of τ̇ ′ .18) makes also clear why Newton’s law in special relativity. and that η is thus just an auxiliary function. ′ Thus Spp is indeed equivalent to Spp . By separation ′ of the mass Spp can also be applied to the case of m = 0 directly. Studying now (5.7 Canonical transformations One of the important insights in chapter 4 was that in section 4. since no derivatives of η appear. Canonical transformations The transformation of η (5. cannot so easily be derived. To show that the new action is indeed equivalent to the old. this requires in general a delicate cancellation between the first and second term in (5. where the force is just produced from the second term. This problem will therefore not be study in general further here. 5. Since the objects are now trajectories. rather than the generalized coordinates and speeds. This is in contrast to the classical case. which is only possible in a limiting process for the original action Spp . (3. Inserting this expression into (5. But. Because the fourth component of the relativistic force depends on the speed. there is an additional advantage. can be shown by using the equation of motion for η. the coordinate-system-invariance is the statement that .20) leads to  s  Z µ µ ′ 1 Ẋ Ẋµ Ẋ Ẋµ 2  Spp = dτ  q − − m 2 Ẋ ν Ẋν m2 − m2 Z q ! m −Ẋ µ Ẋµ = dτ − p − −Ẋ µ Ẋµ 2 −Ẋ Ẋνν Z q = −m dτ −Ẋ µ Ẋµ = Spp . The tetrad η is a so-called auxiliary function.

(5.5) that the equations of motion remain form-invariant. Deriving from (4. but now for the new coordinates. pfi . (5. qif = qi . This has a somewhat different impact in Hamilton’s formulation. The actual dynamical objects of mechanics are the trajectories. adding such a function changes the generalized momenta. Furthermore. t) + ∂t f ) ∂H ∂2f ∂H(qk . ∂qi ∂t∂qi ∂qj ∂qi ∂qi . The canonical equation take the form ∂H f ∂H(qk . pfk . t) ∂pj = = − + ∂qif ∂qi ∂qi ∂t∂qi ∂pfj ∂qi ∂H ∂2f ∂2f ∂f = − + (dt qj ) = −dt pi − dt = −dt pfi . and reexpressing the Lagrange function in these new coordinates. exploiting the form-invariance of the Euler-Lagrange equations. ∂qi ∂qi Thus.21) ∂dt Qi the same derivation of Hamilton’s equation can be performed as before. When changing the generalized coordinates and speeds to new ones Q and dt Q. and thus arriving at the same form of Hamilton’s equations (5.17). not the coordinates.11-5. This changes the Hamilton function as ∂f ∂f H f = pfi dt qif − Lf = pfi dt qif − L − dt f = pi dt qi − L + dt qi − ∂t f − dt qi = H − ∂t f.13). while the generalized coordinates remain unchanged. it was already shown in (4.Chapter 5. and this left the Euler-Lagrange equations also untouched.22) ∂dt qi ∂dt qi ∂dt qi ∂qj ∂qi That is. The introduction of Hamilton’s equations requires to reinvestigate these statements. which are arbitrary invertible and continuously differentiable functions of time. in section 4. and thus whether Hamilton’s equations are form-invariant under changes of the so-called canonical coordinates of generalized coordinates and momenta. Defining thus new generalized momenta as ∂L Pi = . (4. pfk .17) the generalized momenta yields   f ∂(L + dt f ) ∂L ∂ ∂f ∂f pi = = + ∂t f + dt qj = pi + . this does not require any deep calculations. Hamiltonian mechanics 109 it does not matter how the trajectories are described. the Hamilton function is shifted by a partial derivative. However. while the Lagrange function is shifted by a total derivative. whether the trajectories in coordinate space can be replaced by states. t) ∂H ∂pj ∂H ∂H = = = δij = = dt qi = dt qif ∂pfi ∂pfi f ∂pj ∂pi ∂pj ∂pi f ∂H f ∂(H(qi .5 it was shown that it is possible to add to the Lagrange function a total time derivative of a function depending only on the generalized coordinates and time. The question is.

it is global. This implies that the generalized momenta are not unique. t) (5. (5. change the generalized momenta at every point in space and time almost arbitrarily. And also with the same speed. the basic object is the trajectory. as the generalized coordinates do not change: The particle still move in the same way. Why this is the case can already be seen for the present case in classical mechanics. It will be necessary to specify the conditions on the point transformation when this is the case. t). This is a much stronger statement as a change of coordinate system.24) Point transformations. Such a local arbitrariness is also known as a gauge freedom. which satisfy some Hamilton function’s h(Qi .21). Here. the equations are indeed also form-invariant. as also the generalized speeds are not changed.7. At the same time. as the generalized speeds are uniquely determined once the trajectories are known. This freedom implies that the possible redefinitions of variables in the Hamilton case is much larger than in the Lagrange case. Pi . This is not a surprise. Any solution with the same trajectory will provide the same physics. Canonical transformations Thus. This is. This arbitrariness implies that some of the information contained in the generalized momenta is arbitrary. While this result is in itself not totally surprising. as the Lagrange and the Hamilton formalism are for any sets of coordinates equivalent as discussed above. And this freedom can be used to simplify problems. As noted above. as this is done for all coordinates in the same way. not true for the generalized momenta. It implies that it is possible to locally redefine the generalized momenta. i. and still get the same physics. e. While it was straightforward than any transformation changing generalized coordinates and speeds left Hamilton’s equations invariant.110 5. as seen here. this result has some far-reaching consequences. Its generalization plays a very important role in physics. the trajectories remain fixed. pj . In the Lagrange formulation. there is more freedom. t) Hamilton’s 11 Note that the Pi do not necessarily fulfill the relation (5.23) Pi = Pi (qj . this arbitrariness is not there. There it was only possible to change the generalized coordinates and speeds in a way which maintained the fact that the generalized speeds were derivatives of the generalized coordinates. That is the idea behind the following concept. this is actually not possible for an invertible and differentiable point transformation of generalized coordinates and momenta11 Qi = Qi (qj . . but can be changed locally. pj .

it is best to consider first an example. lest alone construct it. Qi = −pi (5. but this is not required for a transformation to be canonical. (5. the problem . −Pi . and the general- ized coordinates and momenta are just description without inherent importance of their own. Pi . and are called canonical. Pi . qi . As a first example. since for the Hamilton function h(Qi .Chapter 5. This is a very important distinction.26) is fulfilled.25-5. pi (Qi . where the Hamilton function satisfies the Hamilton equations in the coordinates qi and pi . t). t) = H(qi (Qi . consider a transformation which exchanges generalized coordinates and momenta. Proper canonical transformations provide a constructive way to create the Hamilton function. or even partially. Hamiltonian mechanics 111 equations ∂h(Qi . Hamilton’s formulation does not distinguish between both. Pi . t) = H(−pi . It implies that generalized coordinates and momenta can be exchanged at will. it is called a proper canonical transformation. Improper canonical trans- formations only require the existence of some Hamilton function such that (5. t) d t Pi = − (5. how far can this be driven? Is it possible to find a canonical transformation which makes coordinates cyclic? Which makes all coordinates cyclic? Then.27) Pi = qi .28) This transformation is a proper canonical transformation. No tool is yet provided to proof the existence of such a canonical transformation. If also h(Qi . before trying to construct general tests for canonicity of transformations. Finding (im)proper canonical transformations could therefore be much more complicated. ∂Qi ∂pj ∂Qi This is a quite remarkable result. t) ∂h ∂H ∂qj = = −dt pi = dt Qi ∂Pi ∂qj ∂Pi ∂h ∂H ∂pj = = −dt qi = −dt Pi .25) ∂Pi ∂h(Qi . t).26) ∂Qi are therefore special. To get a better idea of how far canonical transformation can go. Pi . t) dt Qi = (5. t). Pi . emphasizing again that the trajectory is the basic object. So.

Of course. F1 is uniquely determined. t). pi (Qi . then the transformation together with this new Hamilton function is a canonical transformation. pi dt qi − H(pi . First is that the trajectory should remain the same. though they are inter- changeable with the generalized momenta. Qi .33) Thus.112 5. Qi . then it is unique. This implies ∂F1 = pi (5. further developments are necessary. if either F1 or h are known. this will yield the other.30) ∂qi ∂Qi As. t). Hamilton’s principle will be used.22). with respect to F1 .7. It still remains to show that the associated transformation is canonical. a special role. Pi . up to an irrelevant constant. Canonical transformations would become trivial. Replacing the original Hamilton function using (5. t1 t1 12 The meaning of the index 1 will become apparent later. To do so. The answer to this is affirmative.13. It will be shown that if this relation holds. But before this can be achieved. (5. but reexpress the Lagrange function by the Hamilton function.32) ∂Qi ∂t F1 (qi (Qi . though its parametrization may change. Note that this still yields no constructive way of determining h and thus F1 . and will be given in section 5. (5. t). the variables qi and Qi are independent. For this. the prefactors of dt qi and dt Qi must cancel with those of (5. . t) + dt F1 (qi .1). (5. and Qi and Pi . t) = Pi dt Qi − h(Pi . The first step is done by searching for a way how to construct for any canonical trans- formation the Hamilton function. two particular insights are important. note that ∂F1 ∂F1 dt F1 = dt qi + dt Qi + ∂t F1 . Unfortunately. Consider thus the action S of (5. t) − H(qi (Qi . Pi .31) ∂qi ∂F1 = −Pi (5. It therefore seems reasonable to give the generalized coordinates. To this end. t). t) = h(Qi . Pi ). Both together suggest to make the following ansatz. while possible in principle. in practice this is often as hard as solving the original problem.29) yields Zt2 Zt2 S = dt (Pi dt Qi − h + dt F1 ) = dt (Pi dt Qi − h) + F1 |tt21 = s + F1 |tt21 . qi . The first is finding a criterion for the canonicity of a transformation. Qi . Pi .29).29) where the function F1 is called the generator of the canonical transformation12 between qi and pi . as all derivatives are fixed by known quantities. The second is the observation that the definition of the generalized momenta can be shifted by some arbitrary function of the coordinates. It is helpful to first show that if such a function exists. (5.

Hamiltonian mechanics 113 Performing the variation with respect to the new variables yields .Chapter 5.

t2 ∂F1 .

δS = δs + δQi .


24) make the new variables Qi dependent on the old canonical momenta.34) ∂Qi t1 There is now one important aspect to be taken into account. This subtlety can be taken care of by a partial integration Zt2   . (5. The point transformation (5. However. the values of the Qi at these times still depend on the generalized momenta at these times. Therefore. Since the varied trajectories can have different canonical momenta at the initial and final times. there has been no conditions for the generalized momenta. and the variations at the initial and final times of the Qi can therefore potentially not vanish.23-5. The boundary conditions on the path qi required them to have fixed values at the times t1 and t2 . the corresponding Qi belonging to the original trajectories may change at the initial and final times.

t2 ∂h ∂h ∂F1 .

δS = dt (δPi )dt Qi + Pi δdt Qi − δQi − δPi + δQi .


(5.35) ∂Qi ∂Pi ∂Qi t1 t1 Zt2        .

t2 ∂h ∂h ∂F1 .

= dt dt Qi − δPi − dt Pi + δQi + Pi + δQi .


5. The new Hamilton operator is obtained from (5.32) implies that the term involving F1 vanishes. this implies (5. This resolving of the equations may actually be involved.33).36) . given any arbitrary function F1 . ∂Pi ∂Qi ∂Qi t1 t1 The relation (5. a canonical transformation is created. It therefore appears suspicious that the generating function for canonical trans- formations should depend particularly on the coordinates. Hence. On the other hand.32) for them. .31) and (5.25-5. This also implies that if F1 does not depend explicitly on the time the canonical transformation is proper. as it is only a rewriting of the the original Hamilton’s principle in the old coordinates. The new generalized coordinates and momenta Qi and Pi are obtained by solving the equations (5. Since δS still needs to vanish. But this is just the condition for the transformation to be canonical.8 Alternative formulations of canonical transforma- tions As noted in the previous sections.26). generalized coordinates and momenta are exchangeable concepts. it has been shown that the generating function is unique for a given transformation. (5.

coordinate-dependent representation of this map. Pi . ∂F F2 (qi . which of them yields the most simple result. t).3. It depends on the problem at hand. it can be expressed in any sets of coordinates. it must also be expressible in any set of coordinates. t). Pi . t) + Pi Qi (qi . They are obtained in the same manner. it should be sufficient to reexpress the original function F1 and the new Hamilton function h in those sets of coordinates.31-5. t) F3 (pi . Therefore.8. Qi .30). the transformation cannot be done by merely replacing the arguments. as described in section 5. But the corresponding transformations are known due to the relations (5. This will be exemplified in some examples after their equivalence has been established.37) ∂qi ∂F2 = Qi (5. Therefore. To obtain the other three logical possibilities F2 (qi .38) ∂Pi ∂F2 = h − H. The map itself is coordinate-independent. This will be exemplified in detail for one of the functions.3. ∂Qi ∂Pi Comparing again term by term yields the relations ∂F2 = pi (5. Qi (qi . t) F4 (pi .39) ∂t .8) of F1 on the variable Q1 to P1 . using (5. The function F2 is obtained by a Legendre transformation (5. ∂F2 ∂F2 dt F2 = dt qi + dt Pi + ∂t F2 = dt F1 + Qi dt Pi + Pi dt Qi = pi dt qi + Qi dt Pi + h − H. The trans- formation describes an invertible (and differentiable) map of the state space into itself. Rather. t) − Qi = F1 (qi . Qi . as has been discussed in section 5. However. F2 . while for the others only the relevant results will be given. Pi . (5. This implies for the relevant total temporal derivative. t).33). and it is therefore not necessary to show that it is unique. It is not possible to obtain a generating function containing only the old or the new coordinates. Pi . Note also that there is no general optimal choice which of the Fi are best suited. Thus. they will again be obtained by a Legendre transformation. Pi . The function F1 is a particular. Alternative formulations of canonical transformations This discrepancy can be resolved by again separating the concepts cleanly. as the differentials and invertibility of F1 play important roles. ∂Qi Note that F2 is uniquely determined by F1 .114 5. as such a function would contain no information on the relation between both sets of variables. t) = F1 (qi .

Hamiltonian mechanics 115 and thus a very similar form as (5.31-5.37-5.Chapter 5. The first is to formalize the transformation exchanging generalized momenta and co- ordinates using this framework. it is best to express the Lagrange function. they need to be the same. To show that this also represents a canonical transformation.29).31-5.33) and (5. Their origin will be discussed in more detail in section 6. yielding ∂F3 = −qi ∂pi ∂F3 = Pi ∂Qi ∂F3 = h−H ∂t ∂F4 = −qi ∂pi ∂F4 = Qi ∂Pi ∂F4 = h − H. This is achieved by F1 = −qi Qi since ∂F1 pi = ∂qi = −Qi ∂F1 Pi = − ∂Q i = qi .13). following. yields again that the transformation is canonical.36).5. it is useful to consider a few examples to show how these principles are applied.33). (5. .34- 5.11-5.39) stem from those in (5. as being proper is a coordinate-independent statement. As they are deciding whether a canonical transformation is proper. This result can be entered in the action.33).31-5. In exactly the same way the analogue of (5. but using (5. It is not surprising that the partial derivatives with respect to time are always the same. and proper canonical if it does not depend explicitly on time.33) are obtained for F3 and F4 . After these more formal developments. using F2 .37-5. L = pi dt qi −H = Pi dt Qi −h+dt F1 = Pi dt Qi −h+dt F2 −Pi dt Qi +Qi dt Pi = −Qi dt Pi −h+dt F2 .31-5.39) instead of (5. ∂t The occasionally appearing minus sign here and in (5. and using the same line of arguments as in (5.

consider F4 = −pi Pi . i. 2m 2 where it does not matter whether this is indeed a harmonic oscillator or whether the problem in suitable generalized coordinates just takes the form of a harmonic oscillator. there exists an inverse transformation.28). This is also a proper canonical transformation.27-5. The choice fi = qi is then an identity transformation where the new and old coordinates coincide. ∂Q 2 sin2 Q 13 Incidentally. this matrix is also invertible. and generated by F2 = fi Pi ∂F2 Qi = = fi ∂Pi ∂F2 ∂fj pi = = Pj . Consider the generating function mω0 q 2 F1 = cot Q 2 ∂F1 p = = mω0 q cot Q ∂q ∂F mω02 q 2 P = − = . The identity transformation is therefore also a canonical transformation13 . Alternative formulations of canonical transformations in agreement with (5. To give an example of a reformulation. This implies that a change of coordinate system also changes the generalized momenta. Since a coordinate transformation must be invertible. e. which yields the same transformation. It is often useful to change the coordinate system of the generalized coordinates. of a canonical transformation.116 5. this shows that canonical transformation form a group: There is an identity. consider the harmonic oscillator with Hamilton function p2 mω02 2 H= + q . It can also be obtained as a double Legendre transformation of F1 . ∂qi ∂qi The last equation is a system of linear equations with a transformation matrix Aij = ∂fi /∂qi . It is also proper. Finally.8. at least in principle. . yielding Pi = (A−1 )ij pj . to demonstrate the value. as F1 does not depend explicitly on time. two canonical transformation are again a canonical transformation. Qi = fi (qj ). and the transformations are associative and closed.

the so-called Poisson brackets. the coordinate Q is cyclic. t) where f (qi .41) Since the generating function does not depend on time the new Hamilton function is H(q(Q. (5. P ).5. Therefore. P )) = P ω0 cos2 Q + ω0 P sin2 Q = ω0 P.40) mω0 p p = 2mω0 P cos Q. and the solution is P (t) = P (0) Q(t) = ω0 t + Q0 . Reinserting this into (5.13.41) yields then the familiar solutions of the harmonic oscilla- tor from section 2.9 Poisson brackets A transformation can be most directly be identified to be canonical using a new concept. Another interesting example is the transformation (4. This transfor- mation is obtained from the generating function F2 (qi . p(Q.40-5. t) = qi Pi − f (qi . and thus the same result as in section 4.6. Thus. which will be developed in the following. t) is the same function as in (4. the Poisson brackets are again a concept with an importance which will only become fully unveiled in quantum physics. Pi . a good choice of a canonical transformation can substantially reduce the com- plexity of a given problem. This will be further developed in section 5.17) of section 4. the canonical transformation lead to trivializing the problem. The Poisson brackets will also be very useful to formulate many results . Hamiltonian mechanics 117 Solving these equations for q and p yields r 2P q = sin Q (5. 5. This yields ∂F2 Qi = ∂Pi = qi ∂F2 ∂f Pi = ∂qi = Pi − ∂qi h = H + ∂t F2 = H − ∂t f. Though it is in the end a very compact way of expressing the conditions of canonicality.Chapter 5. Thus.17).5.

how this quantity changes with time. g}q. Poisson brackets in a very compact way. though it can be argued whether this can be considered to be an advantage. H} + ∂t f. This result motivates to define the Poisson brackets as the following prescription.43) Thus.. the generalized coordinates. which are needed for any concrete realizations of the trajectories. This question is: Given a classical mechanical system. akin to. The answer to this question is ∂f ∂f ∂f ∂H ∂f ∂H dt f = dt qi + dt pi + ∂t f = − + ∂t f. it is arguably better to show that it emerges automatically from a very general question. ∂qi ∂pi ∂qi ∂pi The name brackets originates from the curly braces on the left-hand-side. the main focus were on the trajectories of the individual particles.42) ∂qi ∂pi ∂qi ∂pi ∂pi ∂qi where in the second step Hamilton’s equations of motion (5. An interesting question is. solving the system and inserting the results for the generalized coordinates and momenta would answer this question. if the generalized coordinates and momenta are referred to. Of course. Given two functions f and g depending on two set of variables pi and qi . how often particles collide. other questions may be much more interesting. (5. So far. In the context of classical mechanics the index pair p. But it is very often the case that it is already sufficient to know the answer as a function of the generalized quantities. how many particles appear in which area of the system. then writing {f.9. f (pi . e. g}. Since any mechanical system is uniquely characterized by a state.11-5. say. the Poisson brackets with respect to the two sets of variables is defined as ∂f ∂g ∂g ∂f {f. qi . This list can be extended arbitrarily and infinitely. Especially.p = − . any such observable f can only depend on the state. However. t). how often does a particle orbit another particle before escaping to infinity. The Poisson brackets are therefore a mathematical operation. the time evolution of a mechanical quantity is determined by its Poisson bracket with the Hamilton function.12) have been used.42) can therefore be expressed using the Poisson bracket as dt f = {f. While it is possible to just define the Poisson brackets and then show its usefulness. there may be many quantities of interest. multiplication. g. E. up to any explicit time dependence. (5. this provides a result independent of any particular initial conditions. This gives the Hamilton . The result (5. q is usually dropped.118 5. i. momenta. and the time.

(5. Especially. H} = ∂qj ∂pj − ∂qj ∂pj = δij ∂p = j ∂pi ∂pi ∂H ∂H ∂pi H dt pi = {qi .8. for quantities with dt f = 0. i. pk } = ∂qi ∂pi − ∂p ∂qi ∂pi = δji δki = δjk . ∂qi giving a reformulation of Hamilton’s equations (5. modifications of the relations (5. A special case are also the so-called fundamental Poisson brackets ∂qj ∂qk k ∂qj {qj . . as it can therefore be considered as the source of evolution in time. therefore the name funda- mental.45) ∂qj ∂pk k ∂qj {qj . as these are the most elementary Poisson brackets since no further function is involved.44-5. pk } = ∂qji ∂p k ∂pi − ∂p ∂qi ∂pi = 0 (5.47) and is a very compact test for conserved quantities. only mixed fundamental Poisson brackets do not vanish. The time evolution equation (5.46) These are the Poisson brackets of the sets of variables itself.46) can serve as the fundamental postulates of quantum physics.12 below. besides cyclic coordi- nates. e. Therefore. pi ).43) has a particular interesting implication for constants of motions.44) ∂p k ∂pj {pj . This will be put in a different perspective in section 5. i. found so far.43). H} = ∂qj ∂pj − ∂qj ∂pj = −δij ∂H ∂qj = − . e. H} + ∂t H = ∂t H.12) using the Poisson brackets. In this case 0 = dt f = {f. qk } = ∂qi ∂pi − ∂q ∂qi ∂pi = 0 (5. This is also emphasized by ∂qi ∂H ∂H ∂qi ∂H H dt qi = {qi . the statement of section 4. This criterion is often much easier to check than any other criteria.Chapter 5. Thus. For the Hamilton function itself this implies dt H = {H. as it does not require the explicit solution of the equations of motion. (5.11-5. Hamiltonian mechanics 119 function an extremely elevated conceptual importance. Hamilton’s equations can be considered as special cases of the more general time evolution equation (5. H} = 0 ⇒ dt f = 0. Though these seem to be comparatively simple statements. for any not explicitly time-dependent quantity it follows that {f (qi .1 is recovered that the Hamilton function is conserved if it does not depend explicitly on time. H} + ∂t f.

e. just as in section 4. Pl ). this implies {Qi . Consider     ∂Qj ∂H ∂H ∂Qj ∂Qj ∂H ∂Qk ∂H ∂Pk ∂Qj ∂H ∂Qk ∂H ∂Pk dt Qj = − = + − + ∂qi ∂pi ∂qi ∂pi ∂qi ∂Qk ∂pi ∂Pk ∂pi ∂pi ∂Qk ∂qi ∂Pk ∂qi     ∂H ∂Qj ∂Qk ∂Qk ∂Qj ∂H ∂Qj ∂Pk ∂Pk ∂Qj = − + − ∂Qk ∂qi ∂pi ∂qi ∂pi ∂Pk ∂qi ∂pi ∂qi ∂pi ∂H ∂H = {Qj . i. H(pi .120 5.5. Pj } = 0 (5.11-5.7.10 Poisson brackets and (canonical) transformations One of the most important features of the equations of motions. Pk } = 0. Pk )) = H(Ql . qj ) are invertible relations. Studying this feature in detail for the Poisson brackets. and thus especially of the dynam- ical time evolution formulated in (5. as well as Qi and Pi . Qj } = {Pi .48) {Qi . Poisson brackets and (canonical) transformations 5. (5. In particular. there are uniquely defined functions qi = qi (Qj . is that they do not depend on a particular coordinate system. the change of variables is canonical. such that the relations Qi = Qi (pj . i. it can be shown that Hamilton’s equations (5.5.10. Pj } = δij . and that they keep the same form in any coordinate system.43). Pj ). Pj ) pi = pi (Qj . just as in section 4. Consider two sets of generalized variables qi and pi . will lead back to the canonical transformations of section 5. Pj ).13) hold in the same form in the new variables.49) The time evolution of Pj furthermore shows. . Qk } + {Qj . Pk }. Especially. in exactly the same way. pi (Qk . ∂Qk ∂Pk Since the same set of Hamilton’s equation must hold for the new variables. as discussed already for the Lagrange equations of the second kind in section 4. {Pi . qj ) Pi = Pi (pj . qi ) = H(qi (Qj . e.5.

g.. the Poisson brackets are still evaluated with respect to the generalized quantities pj and qj . it is useful to consider first a check using the generating functionals. Given. Pl }. ∂Qi where it should be kept in mind that. e. ∂Ql ∂Pl Before continuing. also the Poisson brackets remain form-invariant under a change of variables. e. canonicality would imply ∂F1 pi = ∂qi ∂F2 Pi = − . In particular. Consider some new variables Qi and Pi . If they are obtained from a canonical trans- formation. it still needs to be answered what happens in the general case. when setting f either to the new generalized coordinates or momentum. g} = − = + − + ∂qi ∂pi ∂qi ∂pi ∂qi ∂Qj ∂pi ∂Pj ∂pi ∂pi ∂Qj ∂qi ∂Pj ∂qi     ∂g ∂f ∂Ql ∂Ql ∂f ∂g ∂f ∂Pl ∂Pl ∂f = − + − ∂Ql ∂qi ∂pi ∂qi ∂pi ∂Pl ∂qi ∂pi ∂qi ∂pi ∂g ∂g = {f. it is useful to note two particular cases of this result for future use. Pi} = .p = − + = {f. Hamiltonian mechanics 121 Though this implies that the fundamental Poisson brackets have the same form. ∂Qi Differentiating both equations once more with respect to any of the other variables leads . no matter the choice of coordinates. ∂Qi ∂Pi ∂Pi ∂Qi i.Chapter 5. in line with the conventions. Qi } = − ∂Pi ∂g {g. g}Q. This feature can now be used to check if a given variable transformation is canonical. These two results can be used also in the original expression. ∂g {g. the dynamical equation (5. Ql } + {f. F1 . g}q. there must exist also the corresponding generating functions Fi . To prepare this.P . In a similar fashion as before it follows that     ∂f ∂g ∂g ∂f ∂f ∂g ∂Qj ∂g ∂Pj ∂f ∂g ∂Qj ∂g ∂Pj {f.43) takes the same form in any system of generalized quantities. yielding   ∂g ∂f ∂g ∂f {f.

10.122 5. Poisson brackets and (canonical) transformations to .


∂pi .


∂ 2 F1 ∂ 2 F1 ∂Pj .


= ∂Qj ∂qi = ∂qi ∂Qj = (5.50) ∂Qj .

q.t.Qk6=j ∂qi .

qk6=i .t.Q.


∂pi .


∂Pj .


51) ∂qj . = (5.

t.Q.qk6=j ∂qi .

t.qk6=i .Q.


∂pi .


∂Pj .


(5.52) ∂Qj . = .

q.Qk6=j ∂Qi .t.

it is always possible to evaluate the Poisson brackets for the new variables in terms of the old ones. This completes the proof.43) holds. for any quantity (5. it will be shown that if (5. At any rate. Rather. as discussed in section 6. Thus. and thus that Hamilton’s equations take the same form also in the new coordinates. An extension to general canonical transformation is possible. Since it does not matter which variables are used.8.q. . ∂Qk ∂Pk Completely analogous follows ∂h ∂h ∂t Pi = − {Qk . if and only if (5.49) hold. Thus. Also. and does not yield them. ∂Qi ∂H ∂Qi ∂H ∂t Qi = {Qi . though this partially enters regions of function theory not yet available. the result implies that certain derivatives of the new and old variables have to coincide. only one possibility will be investigated here. However. then the transformation is canonical. and implies that also generating functions exist. Irrespective of whether a transformation is canonical or not. as it already hints at a certain structure. see section 5. ∂Qk ∂Pk Both results together imply that the transformation is (proper) canonical.48-5. Pk }. it is possible to reformulate it in much more convenient ways. Qk } + {Qi .Qk6=i Similar expressions can be obtained from any of the other Fi s.49) hold.t. just that the derivatives are then with respect to other variables. The following will be done for proper canonical transformation for simplicity.5. in practical calculations this test is rather unwieldy. all of the relations have to be fulfilled. Pk }. Pi } + {Pi . So far it has also been show that if the transformation is canonical they hold. The fact that second derivatives of the generating functions have to coincide is actually conceptually significant. but tedious. the next step is to prove the reverse. In fact. the proof is not constructive.48- 5. Using Poisson brackets. H} = − ∂qj ∂pj ∂pj ∂qj     ∂Qi ∂h ∂Qk ∂h ∂Pk ∂Qi ∂h ∂Qk ∂h ∂Pk = + − + ∂qj ∂Qk ∂pj ∂Pk ∂pj ∂pj ∂Qk ∂qj ∂Pk ∂qj ∂h ∂h = {Qi . However. and make it an equivalence.

qj ) ∂c {c.45) are special cases of the more general statements ∂f ∂f ∂f ∂f {f. The Poisson brackets also vanish if either or both of the arguments are constant. f }.Chapter 5. ∂c ∂g(pj .55) ∂qi ∂pi ∂qi ∂pi ∂qi ∂pi ∂qi ∂pi From this (5. f3 } = − ∂qi ∂pi ∂qi ∂pi     ∂f2 ∂f3 ∂f3 ∂f2 ∂f1 ∂f3 ∂f3 ∂f1 = f1 − + − f2 ∂qi ∂pi ∂qi ∂pi ∂qi ∂pi ∂qi ∂pi = f1 {f2 . g} = − =− − = −{g. f3 } + {f1 . g(pk )} = − =0 ∂qi ∂pi ∂qi ∂pi which hold true for any sufficiently often continuously differentiable functios. It is thus (5.44-5. f3 }f2 .53) follows immediately without calculation. (5.57) . ∂(c1 f1 + c2 f2 ) ∂f3 ∂f3 ∂(c1 f1 + c2 f2 ) {c1 f1 + c2 f2 .56) and products expand like ∂(f1 f2 ) ∂f3 ∂f3 ∂(f1 f2 ) {f1 f2 . (5. f3 } = − ∂qi ∂pi ∂qi ∂p    i  ∂f1 ∂f3 ∂f3 ∂f1 ∂f2 ∂f3 ∂f3 ∂f2 = c1 − + c2 − ∂qi ∂pi ∂qi ∂pi ∂qi ∂pi ∂qi ∂pi = c1 {f1 .54) ∂qi ∂pi ∂qi ∂pi They therefore act non-trivially only on functions. which will be pursued further in section 6. The Poisson brackets are antisymmetric under the exchange of its arguments   ∂f ∂g ∂g ∂f ∂g ∂f ∂f ∂g {f. f3 }. g(pj .53) ∂qi ∂pi ∂qi ∂pi ∂f (qj ) ∂g(qk ) ∂g(qk ) ∂f (qj ) {f (qj ). (5.11 The Poisson bracket as a differential operator The Poisson brackets can be used to establish a more abstract mathematical framework in mechanics. Before laying the foundations for it. the results (5.5. qj ) ∂g(pj . The Poisson brackets are a linear operation for any function fi and constants ci . First of all. Hamiltonian mechanics 123 5.46) which holds the non-trivial information. it is useful to collect a few properties of the Poisson brackets. g(qk )} = − =0 ∂qi ∂pi ∂qi ∂pi ∂f (pj ) ∂g(pk ) ∂g(pk ) ∂f (pj ) {f (pj ). f3 } + c2 {f2 . f } = − =0 (5. qj )} = − = 0. (5.

{g. the Poisson bracket is just one example.54). where the differential operator is Df . g} = Df g. are linear operators. and other generalizations. h}} = f. of a whole class of mathematical operations.58) ∂qi ∂pi ∂qi ∂pi and thus the product rule for ordinary differentials applies also to Poisson brackets. as it depends on the function f .57).55). (5. ∂x g}. They are therefore considered to be a differential operator. and serves here as a prototype. B] = AB − BA also satisfies the same conditions. the so-called commutator of two square matrices A and B is defined as [A.11.56). and obey a product rule. In fact. ∂qj ∂pj ∂qi ∂pi ∂qi ∂pi ∂pj ∂qj ∂qi ∂pi ∂qi ∂pi creating in total eight terms. It is possible to create double Poisson brackets. Just to give another example. h}} + {g.53-5. appear throughout physics.59) which is known as the Jacobi identity. These. . g}} = 0. (5. where the concept of a constant is taken by the unit matrix.124 5. − ∂qi ∂pi ∂qi ∂pi     ∂f ∂ ∂g ∂h ∂h ∂g ∂f ∂ ∂g ∂h ∂h ∂g = − − − . {g.   ∂f ∂g ∂g ∂f ∂x {f. The Poisson bracket as a differential operator and likewise for other expressions by usage of the antisymmetry (5. sometimes also written as Df = {f. ·}. Such generalizations of differential operators play an important part of formal theory developments. f }} + {h.   ∂g ∂h ∂h ∂g {f. The Poisson brackets have thus a number of features familiar from differential opera- tors: They yield zero when acting on a constant. In contrast to an ordinary differential operator it is a parametrized differential operator. expressed as {f. Explicitly performing such calculations. g} + {f. (5. {f. and especially in quantum physics. {h. Note also that. for x any variable. it is possible to show that {f. (5.59). (5. each term having one second derivative and two first deriva- tives. which all satisfy the properties (5. g} = ∂x − = {∂x f.

F2 = qi Pi + ǫG(qi . the generating function needs to be only a small deviation from the identity transformation as well. where δx should for now only indicate that the quantity δx is small.38) it follows that ∂G Pi − pi = δpi = −ǫ ∂qi . {H. and is not connected to displacements.8. Consider an infinitesimal transformation of the coordinates and momenta Qi = qi + δqi Pi = pi + δpi . 5. ∂t f } + {H. Hamiltonian mechanics 125 To show the usefulness of the derived relations for the Poisson brackets consider the following question: Given two conserved quantities. {f. g}} = {∂t g. g} = 0. H}} + {g. e. H} + ∂t {f.12 Poisson brackets and symmetries Poisson brackets give another conceptual view on invariances and symmetries. Thus. Of course.59) 0 = {f. Because of (5. small enough to make ǫG infinitesimal as well. Pi ) where G is some function and ǫ an infinitesimal parameter. and thus the conservation laws of section 4. g}} it follows that {{f.37-5. it is not possible to generate with this infinitely many non-trivial conserved quantities. f }} + {H. g. f } + {g. {f. due to the Jacobi identity (5. is it possible to find a non- trivial further conserved quantity from it? In fact. g}. At some point the Poisson brackets will give again just combinations of known conserved quantities. If this is the case. {g.Chapter 5. f and g. the Poisson bracket of any two conserved quantities is again a conserved quantity. But often enough it is helpful for a construction of additional con- served quantities.

! ∂G ∂ ∂G(qi . Pi ) .


∂G(qi . pi ) + (Pi − pi ) = ǫ ∂Pi ∂Pi ∂Pi . pi ) Qi − qi = δqi = ǫ ≈ǫ G(qi .

Pi =pi ∂pi where the last approximation is good. . as a small change multiplied by ǫ is already smaller than ǫ.

G} .61) states that the integral of motions associated with a symmetry generate changes of quantities. As has been discussed in section 4. this will be the key to identify symmetries of the systems and the all the dynamical features of the theories. δqi δpj ) − u(qi . It is interesting to consider the case that G = H and ǫ = dt a small interval in time. Poisson brackets and symmetries So far.60) together with (5. . This statement is actually a very deep one. This implies time evolution is equivalent to an (infinitesimal) canonical transformation. its Poisson bracket with the Hamilton functions vanishes.43). δH = ǫ {H. the function G will be seen to generate the transformation of any quantity under the corresponding symmetry transformation. this implies that any coordinate transformation generated by an integral of motion leaves the Hamilton function unchanged.61) and thus also the Hamilton function itself changes in general. It is therefore valid to say that the Hamilton function creates the evolution in time. this is general. If the function G is a constant of motion then. Consider now some arbitrary function of u. Thus. It is one of the key properties in the connection of symmetries and physics.126 5. G} . the change of a quantity u under some (infinitesimal) coordinate transformation G is determined by the Poisson bracket of this quantity with the coordinate transformation. which is quite similar to the statement (5. δpi . which changes as δu = u(qi + δqi . In the present context. pi ) + δqi + δpi + O(δqi . this canonical transformation actually pushes the generalized coordinates and mo- menta forward in time.8 an integral of motion is usually associated with some symmetry of the Hamilton function. because of (5. pi + δpi ) − u(qi . In fact.12. By consecutive such infinitesimal transformations ultimately the whole time evolution can be build. (5.47). Thus. This leads to dqi δqi = dt ∂H ∂pi = dt dt = dqi δpi = −dt ∂H ∂qi = dp dt i dt = dpi . pi )   ∂u ∂u 2 2 = u(qi .60) where in the second line a Taylor expansion was performed. but leave the Hamilton functions invariant. (5. Thus. Especially. (5. pi ) qi ∂pi ∂u ∂u ≈ δqi + δpi qi ∂pi   ∂u ∂G ∂u ∂G = ǫ − qi ∂pi ∂pi ∂qi = ǫ {u. In quantum physics.

Chapter 5. For simplicity. according to (5. G is then G = Lz . The corresponding symmetry transformation is.8.62). On the other hand. Hamiltonian mechanics 127 However. if G is an integral of motion. this function will have vanishing Poisson brackets with the Hamilton function. a full general proof of the connection is beyond the present scope. which requires G = pj . the integral of motion is connected with invariance. Thus. Lz } = {ypz − zpy . realizing the aforementioned statement. indeed the Poisson bracket with the function G generates the transformation.62) δpi = 0. G} = δθ − + (x → y. As a second example consider Cartesian coordinates and a rotation around the z-axis. z) = −δθyi . G} = 0. an infinitesimal translation in qj . in the corre- sponding coordinate. Furthermore. note   ∂xi ∂G ∂G ∂xi ǫ{xi . It will thus need to be sufficient to do so for some examples. the z component of the total angular momentum. ∂xj ∂pxj ∂xj ∂pxj and likewise for yi or the momentum. These changes are created by G = xi pyi − yi pxi . Again. Thus. as this is the transformation performed. First consider that some coordinate j is cyclic and the transformation δqi = ǫδij (5. It is also very interesting to consider what happens for u = Lx . in this case the rotation. ∂xi ∂pi ∂pi ∂xi . xpy − ypx} ∂(ypz − zpy ) ∂(xpy − ypx) ∂(ypz − zpy ) ∂(xpy − ypx ) = − = −(zpx − xpz ) = −Ly . this implies {H. provided ǫ = δθ is identified. because the system has symmetry around the z-axis as investigated in section 4. This yields {Lx . or translation symmetry. reduce this to a single particle. The corresponding change is δxi = −yi δθ δyi = xi δθ δzi = 0. Since H does not depend on pj . the generator of this transformation is the conserved quantity.

is the fundamental space of (classical) mechanics. the shape of the surface. there exists a canonical transformation such that the problem becomes trivial. while this is possible in principle. This is known a Liouville’s theorem.57) it also follows that {L2 . and thus yielding trivial Hamilton’s equations.13 Hamilton-Jacobi theory It will now be shown that it is possible to find for any mechanical problem. However. the conservation is only true if the system has rotational symmetry. Lj } = ǫijk Lk . . the time-evolution along any trajectory in phase space is generated by the Hamilton function. The drawback is that. the knowledge that it is possible in principle helps often substantially in gaining conceptual insights. this surface will not open when evolving in time. 5. These results holds in general. Using (5. even if practically tedious. The advantage is that. Trivial in this case means that it is either transformed to a known problem. Hamilton-Jacobi theory Generalizing by permutation. this yields {Li . and trajectories on the inside cannot move outside. as noted in section 5.128 5. Note that the quantum versions of these expressions will play a very central role in quantum physics. Thus. and it will play an important role in statistical mechanics. or to a situation in which all coordinates becomes cyclic and the Hamilton function is explicitly time-independent or finally such that all generalized coordinates become constant. it is an important development. this is equivalent to an infinitesimal rotation and translation in phase space.2. with is symplectic geometry. Also. and vice versa. As a consequence. trajectories in phase space do not cross. Li } = 2Lj {Lj .13. and only then the components of the angular momentum generate rotations. Since trajectories are continuous functions. it is often as complicated or more complicated than the original problem. This implies that a volume in phase space is surrounded by a dense set of trajectories as a surface. An interesting consequence can be obtained from these insights. If no explicit time-dependence exist. the density of trajectories needs to remain constant over time. Li } = 2ǫijk Lj Lk = 0. As has been shown. may change. Thus. However. Note that similar considerations do not apply to either position space or configura- tion space. phase space. and thus of the volume. which is formulated by a Hamilton function.

this implies that there will be as many initial conditions as derivatives. possible using any of the Fi . leaving N.Chapter 5. This ambiguity is due to the fact that point transformations of the generalized momenta are always a possible canonical transformation. The generalized coordinates will only be time-independent. the third option is the most attractive one. This is necessarily the case if it is possible to find a canonical transformation such that h = H + ∂t F = 0. In this context.13. ∂qi This is an implicit. since it is often most easily achieved by first transforming to either of the second or third option and then back to the first one. the mathematical theory of partial differential equations is much more involved. of course. F2 is also called Hamilton’s action function S. Furthermore. t + ∂t F2 = 0. the solution for F2 is only unique up to a constant. but not with respect to the Pi . thus the number of generalized coordinates plus one for the time derivative. As usual. and will be the starting point. and it will require substantially more work to find a solution14 . this implies N + 1 integration constants. in the equation only the derivatives of F2 with respect to the momenta qi enter. it is of at most first order in the variables of F2 . Thus. The following becomes most simple when searching for a generating function of type F2 . Hamiltonian mechanics 129 5. though it is. This yields the Hamilton-Jacobi equation   ∂F2 H qi . Unfortunately. However. in contrast to ordinary differential equations. in general non-linear as the generalized momenta enter quadratically. differential equation for F2 . One is used up for the additive and irrelevant constant. The second option cannot always be reached if there is explicit time-dependence. as then all (partial) derivatives of the new Hamilton function vanish identically. But for any problem relevant in (conservative) mechanics the existence of such a solution can be taken to be guaranteed. This yields the following strategy to create a trivial solution: 14 Whether a solution for partial differential equations exist is in general a much more involved problem than for ordinary differential equations. for reasons to become clear at the end of the derivation. However.1 General strategy Explicit time-dependence of the Hamilton function plays an important role in which pos- sibility of the three enumerated ones can be achieved. . if the Hamilton function is constant. Given N generalized coordinates. while the first one is often actually more involved. . since for any canon- ical transformation only the derivatives matter. This freedom will be used to identify the new generalized momenta with the N initial conditions.

In principle. This then gives the qi and pi as a function of the initial conditions and the time. ∂αi Since F2 satisfies the Hamilton-Jacobi equations. solving the problem While this procedure guarantees a result the step 2. 2m ∂q 2 There is. where the αi are the initial conditions. βi and time only (4) Solve ∂F2 pi = ∂qi to obtain the pi as a function of the αi . even for this comparatively simple. yielding the generalized coordinates qi as a function of the αi .2 A solution strategy To outline the Hamilton-Jacobi strategy it is useful to consider the standard example of the harmonic oscillator with the Hamilton function p2 mω 2 2 H= + q . βi and t. 5. αi .130 5. this is possible for any canonical trans- formations. it is admissible to identify in the following the αi with the Pi (3) Solve ∂F2 Qi = = βi . t). 2m 2 The Hamilton-Jacobi equations then takes the form  2 1 ∂F2 mω 2 2 + q + ∂t F2 = 0. Since this does not constrain the dependence of F2 on the αi .this is the tricky part. This may again not be possible analytically in general.13. by definition the Qi have to be constant. This yields F2 (qi . but also the steps 3-5. Hamilton-Jacobi theory (1) Write down the Hamilton-Jacobi differential equation (2) Find the solution F2 . may in general only be possible numerically. Again. but is possible in general (5) Determine the constants αi and βi from the initial conditions for the qi and pi . . In the next subsection. case no obvious solution. Then solve for the generalized coordinates qi . this may not be possible analytically. some examples in which it is possible analytically will be discussed.13.

2m dq 2 Since the left-hand side does not depend on time and the right-hand side does not depend on q.Chapter 5. which has to be identified with the second integration constant. This yields  2 1 dW mω 2 2 + q = −dt V. the ordinary differential equation s   dW 2 2 2γ 2 = mω −q (5. 2 mω mω 2 2|γ| The sum V (0) + W (0) is actual only a single independent integration constant. Hamiltonian mechanics 131 Based on experience. It is the constant shift of F2 . Likewise. so-called parabolic partial differential equations with no mixed terms. required to solve the Hamilton-Jacobi equation. with initial condition V (0). and therefore plays no role in the following and can be set to zero. yielding r s !! q 2γ γ mω 2 W (q) = mω − q2 + sin−1 q + W (0). it can be shown that this is not the case. a good starting point is a so-called separation ansatz F2 = W (q) + V (t). it is identified with the generalized momentum P .63) dq mω 2 can also be solved by a (non-trivial) direct integration. γ = α. this equation can only hold if both sides are equal to some constant γ. However. 15 It is possible to worry whether making this ansatz excludes possible solutions. that is it is assumed that the dependence on the coordinates and on the time are additive15 . . This leaves the new constant γ. 2 mω 2 mω 2 2|γ| This yields the total function F2 r s !! q 2γ γ mω 2 F2 = mω 2 − q2 + sin−1 q − γt + V (0) + W (0). as discussed in the previous subsection. due to the separation ansatz. As required by the solution strategy. for partial differential equations of this type. The ensuing ordinary differential equation for V ∂t V = −γ is solved by direct integration V (t) = −γt + V (0).

Though this is not necessarily the result of the Hamilton-Jacobi strategy. The generalized momentum is obtained by r ∂F2 dW 2α √ q= = = mω − q 2 = 2αm cos (ω(t + β)) . yielding Z Z Z r ! ∂ ∂F2 ∂2W 1 1 1 1 −1 mω 2 β= dq = dq −t= dq q = sin q − t.63) with respect to α and then to integrate with respect to q.132 5. up to α and β. Finally. The value of α is just the total energy. Thus. this requires to solve for α and β for the initial conditions. Choosing q(0) = q0 and p(0) = 0 yields mω 2 q02 α = =E 2 π β = .6. the fact that energy and time are canonical conjugate variables holds actually for quite a variety of systems. while β has dimension of time. ∂α ∂q ∂q∂α ω ω 2α − q2 ω 2α mω 2 This can be solved straightforwardly for q yielding r 2α q= sin (ω(t + β)) . Hamilton-Jacobi theory The next step is to solve ∂F2 Q= =β ∂α This is actually simpler by first deriving (5. The final step is to insert this into the solutions of q and p. Thus. the new canonical variables are energy and (a constant) time. as in section 2. yielding r 2E q(t) = cos(ωt) mω 2 √ p(t) = − 2Em sin(ωt). energy and time are canonical conjugate variables. This result has also an interesting implication. which still need to be determined by the initial conditions. . 2ω in which E is the total energy.13. mω 2 which already has the form of the final solution. ∂q dq mω 2 where the last step follows from a trigonometrical identity.

∂qi where the αi are again the initial conditions for the Hamilton-Jacobi equation. . assume that ∂t H = 0. . ∂qj In this context W is called Hamilton’s characteristic function. + ∂t F2 = 0. Hamiltonian mechanics 133 The ansatz of a separation of variables made the solution of the problem in the previous subsection possible. Make the ansatz F2 = W (qi ) − Et. Time plays a quite different role than space. The actual reason why this approach worked out rather well was the fact that the time-dependence could be separated. the only case a time-derivative appears is in the second term. By reducing to a spatial problem. As this type of ansatz is of great value far beyond the realm of mechanics. This emphasizes that E. In general. many things simplify. only appears as    ∂W ∂W H qi6=1 . f1 q1 . = E. but often it is still so. is not an inert quantity. but in practice this is often a fundamental advantage16 . this may not be true for generalized coordinates. Note the constant E will in general depend on all the initial conditions still required to solve the equation for W . This implies that the canonical transformation satisfies ∂W ∂E Qi = − t ∂αi ∂αi ∂W pi = . because the Hamilton function did not depend on the time explicitly. ∂qj Again. with E for now some constant. although being a constant.Chapter 5. The Hamilton-Jacobi equation takes then the form   ∂F2 H qi . The only difference is that the Hamilton- Jacobi equation for W has one variable less. but plays an important role in the final solution. Of course. it is worthwhile to understand it better. this is actually less surprising. − E = 0. say q1 . In the case of skleronom constraints E will usually turn out to be the total energy. This may seem like not a big gain. This removes the time dependence from the Hamilton-Jacobi equation and yields   ∂W H qi . ∂qi6=1 ∂q1 16 This will become even more apparent once treating the quantum-mechanical version of the problem. the latter being Euclidean. This procedure can be continued. From here on the next steps are as before. if one of the variables. Considering special relativity.

The constants αi will then take the role of the new generalized momenta and integration constants. separable.8 for all but the one non-cyclic variable. the ansatz W (qi ) = W ′ (qi6=1 ) + W1 (q1 ) with   dW1 f1 q1 . If this is the case. almost by definition. In the ideal case. depends on all the integration constants. Thus. if also for some of the other variables such a structure emerges. just as E. a suitable ansatz is X W = W1 (q1 ) + αi qi . c1 . Of course. the equation for W1 becomes an ordinary differential equation. at least in principle. If only the generalized constants starting from some k should be cyclic. c1 − E = 0 ∂qi6=1 is possible. If all but one generalized coordinate are cyclic. this final form of the problem is always separable: Since the .. and is therefore equivalent to a set of ordinary differential equations. This reduces the complexity further. as an ordinary differential equation is much simpler to solve. this implies that in this case X W = W ′ (q1 . as the Hamilton-Jacobi equation itself is then already an ordinary differential equation. qk−1 ) + αi qi i≥k is a suitable ansatz. this procedure can be repeated.13. since any mechanical system can. = c1 dq1   ∂W H qi6=1 . At the same time. since all variables are independent. the Hamilton-Jacobi equation is. In this case. i6=1 which induces the identity transformation of section 5. the Hamilton-Jacobi equation is said to be separable. Whether the Hamilton-Jacobi equation is separable or not is actually not a question of the system. and this will maintain the cyclicity of those generalized coordinates which are already cyclic. with the solution involving one of the integration constants. the problem can be entirely reduced in this way. but merely of the choice of variables. .. . be reduced to a constant Hamilton function by usage of the Hamilton-Jacobi strategy..134 5. Hamilton-Jacobi theory Just as before. Evidently. and one variable less in a partial differential equation is also often a substantial simplification.

this can be solved by a separation of variables. than Hamilton-Jacobi theory to find ideal coordinates.Chapter 5. Consider the central potential problem of section 2. and especially no more constructive way. The second new variable is then Z ∂W ∂W αφ Q2 = = = φ − dr q = β2 . Since φ is cyclic. yields as new coordinate Z ∂W ∂W m Q1 = = = dr q = β1 + t. the problem is separable. due to the importance of it far beyond mechanics. there is no better. This yields the Hamilton-Jacobi equation  2  2 !   1 ∂W 1 ∂W 1 2 αφ2 + 2 + V (r) = (dr W1 ) + 2 + V (r) = E. It is useful to investigate a few more examples with this strategy. the functions fi are all trivial. ∂α2 ∂αφ r 2 2m(E − V (r)) − αφ r2 17 Unfortunately. E = α1 . This emphasizes again how important a good choice of generalized coordinates is17 . r and thus Z r αφ W = dr 2m(E − V (r)) − 2 + αφ φ.7. . Hamiltonian mechanics 135 Hamilton function is constant and does not depend on any of the variables. and the ansatz for W is W = W1 (r) + αφ φ. This implies that often experience and intuition is invaluable to simplify a mechanical problem. yielding r αφ dr W1 = 2m(E − V (r)) − 2 . r Choosing E as the second integration constant. 2m r r 2 where it has already been used that angular momentum conservation enforces a movement in the plane with the radius variable r and the angle φ together with their respective generalized momenta. 2m ∂r r ∂φ 2m r As an ordinary differential equation. Its Hamilton function is   1 2 p2φ H= p + + V (r). ∂α1 ∂E 2m(E − V (r)) − r2 αφ where the last equality follows by construction.

7.18).20-2. This yields the Hamilton-Jacobi equation 1  (dz W1 )2 + αx2 + αy2 + mgz = E. the new variables are then ∂W 1 q Q1 = t + β1 = =− 2m(E − mgz) − αx2 − αy2 ∂E mg ∂W αx q Q2 = β2 = =x+ 2 2m(E − mgz) − αx2 − αy2 ∂αx mg ∂W αy q Q3 = β3 = =y+ 2 2m(E − mgz) − αx2 − αy2 . Z q 1 3 W1 = dz 2m(E − mgz) − αx2 − αy2 = c − 2 2m(E − mgz) − αx2 − αy2 2 . as a cyclic coordinate. This solves the problem. 2m i where the two coordinates x and y are cyclic.13. An adequate ansatz for the characteristic function is therefore W = W1 (z) + αx x + αy y. which can again be dropped. somewhat simpler than in section 2. Hamilton-Jacobi theory which is. the usage of the Hamilton-Jacobi equation produced in a straightforward way. 3m g where c is an integration constant.7. ∂αy mg . 2m As before. consider the case of a particle in a homogeneous gravity field. In this case. the usage of the approach may also make problems more involved. which can be solved by separation of variables. if the potential is exchanged for the effective potential of (2. the result. As an example. this is an ordinary differential equation. identify β2 = φ0 1 x = r αφ = l yielding Z 1 φ = φ0 − dx q  2m 1 l2 E−V x − x2 in agreement with (2.21). This integration constant is associated with the initial condition for the time. time-independent.136 5. To compare with the results of section 2. Following the program. However. The Hamilton function is then 1 2 H= p + mgz.

due to an intimate connection to the action formalism. Hamiltonian mechanics 137 This solves the problem completely. The solution to the Hamilton-Jacobi equation F2 is a generating function for a canonical transformation. choose ~x(0) = ~0 and p~(0) = p0~ex . The final result is thus p0 x = t m y = 0 gt2 z = − . constant coordinates βi and αi are exchanged for the initial conditions. This becomes more evident when the generalized. leading to αx x = β2 + (t + β1 ) (5.3 Connection to the action In subsection 5.4). it seems to be a rather involved way of expressing the movement of a particle under a constant. The former and (5. 2 which already looks much more like (2. The latter and since F2 depends on the old coordinates and new momenta implies ∂W t=0 px = ∂x = αx = p0 ∂W t=0 py = ∂y = αy = 0 q q ∂W t=0 ! pz = ∂z = 2m(E − mgz) − αx2 − αy2 = 2mE − p20 = 0. as the name suggests. This is. and the last equation requires E = p20 /(2m). (5. (5. as expected and identifying the constant E indeed as the energy at t = 0.18).13. Transforming back to the original variables can be done by solving the above expressions for them.65) m 1 2 2mE − (αx2 + αy2 ) z = − g(t + β1 ) + . external force.66) imply βi = 0.13.64) m αy y = β3 + (t + β2 ) (5.66) 2 2m2 g While this results coincides (necessarily) with (2.Chapter 5.67) .64-5. As such. For convenience. as will be shown now. 5. it follows that ∂F2 ∂F2 dt F2 = ∂t qi + ∂t Pi + ∂t F2 ∂qi ∂Pi = pi ∂t qi − H = L.1 the generating function was also refereed to as Hamilton’s action.

as long as the initial conditions have not been entered. Thus. the Lagrange function for the harmonic oscillator with the used initial conditions is p2 mω 2 2 L= T −V = − q = E(sin2 (ωt) − cos2 (ωt)) = 2E sin2 (ωt) − E. and thus F2 equals almost the usual action S of section 5. ∂qi ∂Pi as W is just the time-independent part of F2 . from (5. W is identical to the indefinite action A. 5. In fact. much can be learned about a periodic system by studying the involved frequencies. The only difference is that F2 is the undetermined integral of the Lagrange function.67) follows dW dq dt F2 = − α = p − α = 2E sin2 (ωt) − E. 2m 2 On the other hand. A very similar consideration also applies to the characteristic function W .13. Integrating yields Z Z W = pi dt qi dt = pi dqi .5.2. as to actually performing the inte- gration requires to know the time-dependence of the generalized coordinates and momenta. It is useful to see this in the case of the example of the harmonic oscillator of section 5.2. the total time-derivative of F2 is just the Lagrange function.138 5. connecting it to the action for the principle of least action of section 5.13. Hamilton-Jacobi theory Thus.13. In fact. Unfortunately. and therefore providing information on all possible trajectories in phase space. F2 can be considered to create all possible actions for any initial conditions. this . dq dt as expected. rather than the positions of the involved particles. since ∂W ∂W dt W = dt qi + dt Pi = pi dt qi .4 Angular and action variables A particular important special case of Hamilton-Jacobi theory are periodic systems. and therefore the action of Maupertius’ principle. this does not help in computing F2 .1. Performing an inte- gration over time yields Z Z F2 = dtdt F2 = dtL. As has been seen in section 2. On the one hand.6.

An example is given by a pendulum with q being the angle and p its conjugated momenta. the system is also said to be periodic with period τ . However. If. in section 5. In the context of periodic motion it is now useful to define so-called action variables I Ji = pi dqi . the trajectory does not closed.Chapter 5. and the motion is called of limited periodicity. a rotation. Therefore. it is important to further classify by what is meant by periodicity. If each pair is proper periodic. τ . In the context of theoretical mechanics. If the system is described by more than a single set of canonical conjugated momenta. This case is called a libration. Is is therefore practically useful to see how these features emerge in general. even to much more abstract concepts. called the period.7 it has been seen that the concepts of position and momenta are exchangeable. the system is said to be periodic and its trajectory in phase space is closed. this is a libration. If. Hamiltonian mechanics 139 observation generalizes to many systems. with q a generalized coordinate and p its canonically conjugated momentum. then for each pair a period τi can be defined. If it rotates. then the total trajectory in phase space is closed. a system is considered periodic if a particle comes back to a given position after some fixed period τ . it becomes therefore doubtful to define periodicity just in terms of coordinates. on the other hand q(t + nτ ) = q(t) + nq0 p(t + τ ) = p(t) holds. Given a set of canonical conjugated variables q and p. even is a subset of particles come back to their original position multiple times during this period. It is therefore. especially in quantum physics. for all t some quantity τ . If the periods τi are rational multiples of each other. If the pendulum does not rotate. In the usual case. it is replaced by the following definition in the case of a single coordinate. the angle is enlarged by q0 = 2π every period. the system is only periodic if all the particles come back to a given position after some time τ . exists such that q(t + τ ) = q(t) p(t + τ ) = q(t). but the periods have non-rational ratios. but this is called a rotation. as the value of the angle is between some fixed values. If there a multiple particles involved. This case is called proper periodic. Before embarking on this endeavor. but merely swings. as the name suggests.

explaining the name. So far. It is therefore possible to write the Hamilton function as a function of the action increments per period. If the system is fully separable. However.13. according to section 5. the difference vanishes. The conjugate variables to the Ji are the new variables given by ∂W ωi = . any canonical variable has to move exactly the same distance around zero to have a periodic motion.3. This explains the name of the variables. Hamilton-Jacobi theory where no summation is performed on the i and the integral symbol indicates that the integral should be taken over a full period. Thus. as indicated. this yielded little more than an application of the Hamilton-Jacobi theory to the case of a periodic motion. Jj ) as the set of variables of the characteristic function. which is just the behavior of an ordinary angular variable for conventional periodic move- ments. and therefore it is possible to select W (qi . when averaging over many periods. ∂qj ∂qj ∂Ji ∂Ji ∂qj ∂Ji τj τj τj τj This can be understood in the following way: Barring all offsets. Since in the separable case all the variable pairs are independent. Their equations of motion are ∂h d t ωi = = νi (Jj ). the action variable takes the form I dWi Ji = dqi = Wi |t+τ t = ∆Wi (αi ). ∂Ji and these are the constant frequencies. This value will depend on the αi .140 5. due to this set of canonically conjugated variables per period of this pair. this can be solved for the Ji . dqi τ where ∆Wi is thus the change of the action. yielding ωi = νi t + βi . if it is not exactly the same . consider the change of any given angular variable i during the period of some other pair J for the separable case.13. I I I I ∂ωi ∂2W ∂ ∂W ∂ ∆j ωi = dωi = dqj = dqj = dqj = Jj = δij . which is even kept beyond the separable case. showing again the fundamental importance of the action. ∂Ji which are called angular variables. Since in the separable case all coordinates are cyclic this implies that the Hamilton function only depends on the Ji .

the real power of the method arises when realizing that it makes it unnecessary to obtain the full solutions to obtain the frequencies only. Thus νi = τi−1 are the frequencies of the corresponding periodic movements. this information would be redundant and therefore eliminated by the constraints. This also implies ∆i ωi = 1 = τi νi . and thus only a contribution of the same variable remains. It separated Hamilton-Jacobi equation reads (dq W )2 mω 2 q 2 + = E. with no summation implied. and it therefore does not appear. mω 2 It is now not necessary to solve this equation. mω 2 The action variable has thus the value I Zq+ r r 2 !!. and therefore the result is equivalent to the average over many periods of the variable j. which have thus the values r 2E q± = ± . not an average.6. It merely suffices to note that for a periodic movement the rate of change has to vanish at the inflection points. The speed of movement does not matter. This can already be seen when considering the simple harmonic oscillator without external force of section 2. Hamiltonian mechanics 141 period. This is. If there would be a second variable with the same period. As noted above.Chapter 5. Though this seems to be trivial enough at first glance. this already delivers a lot of information on a system. of course. 2m 2 yielding the ordinary differential equation r 2E dq W = ±mω − q 2 = p. But an arbitrary period.

q+ .

q 2E 2 E −1 mω .

2πE J = pdq = 2 pdq = 2mω − q + sin q .

= . 2 mω 2 mω 2 2E .

ω q− q− This equals. the Hamilton function ωJ h=E= 2π and the relevant frequency is ∂H ω ν= = . ∂J 2π . by construction.

This can be used to reduce the number of degrees of freedom to S − M. This postulate was a de- ceisive step in the understanding of the line spectrum of hydrogen. Hamilton-Jacobi theory Of course. it is called completely degenerate. Thus. since ∂h ν̄i = ∂ J¯i this implies that the new Hamilton function only depends on the non-degenerate action variables. Planck’s constant. If M is maximum. all the M degenerate frequencies are now zero.6. In case of S = 1 the system is considered to be also completely degenerate.13. Chose M X S X ¯ = F2 (ω. S − 1 if there are S frequencies. was that all eigenaction variables can only take values which are integer multiplies of the quantum of action. The action variables associated with non-zero. degenerate motion does not contain independent information. J) J¯m nm i ωi + J¯m ωm . i. m=1 m=M +1 where the second term is the identity transformation for the independent frequencies. Modern quantum physics supersedes this postulate with more fundamental ones. Thus. but a (modified) version of the Bohr-Sommerfeld postulate can then be derived for periodic systems. where m counts the condition and the nmi are rational numbers. Note that by suitable multiplications the rational numbers nmi can always be replaced by integer numbers. the so-called Bohr-Sommerfeld postulate. non-degenerate frequencies are also called eigenaction variables. For the M degenerate frequencies the new frequencies are ν̄m = nm i νi = 0. Furthermore. this coincides with the result of section 2. With this postulate quantization was build into any (periodic) mechanical system. . One of the basic postulates in the early formulation of quantum physics.142 5. If M of the frequencies are integer multiple of each other then the system is called M-fold degenerate. If there is M-fold degeneracy then there are M relations nm i νi = 0. e. but here it was not necessary to find φ(t) to determine the period and frequency in terms of the parameters of the Hamilton function.

where it was used that φ is cyclic.3. 2m r r sin θ r Studying this equation. Making the ansatz W = Wr (r) + Wθ (θ) + αφ φ. it is seen that θ has only a single explicit appearance. These are somewhat involved problems and concepts. this will be sufficient to fully separate the system. 2m 2m sin2 θ 143 . They should therefore demonstrate the possibilities of what has been achieved up to this point. consider its Hamilton-Jacobi equation   2 2 (∂θ W )2 (∂φ W )2 k (∂r W ) + 2 + 2 2 − = E. In this case. The Hamilton function reads   1 2 pθ p2φ k H= pr + 2 + 2 2 − . However. and require a full use of the developed techniques. often considered to be the epitome of theoretical mechanics. after some rearrangements. to r2 1  αφ  (dr Wr )2 − kr − Er 2 = (dθ Wθ )2 + . 6.13 this does not seem to be fully separable.7.Chapter 6 Selected special topics In this chapter a few selected problems and aspects of theoretical mechanics are discussed using the apparatus developed in the previous chapters.1 The Kepler problem revisited To give a challenging and non-trivial application of the Hamilton-Jacobi theory. since only the coordinate φ is cyclic. consider once more the Kepler problem of section 2. leads. 2m r r sin θ r In contrast to the examples in section 5.

They occur whenever the argument of the root vanishes. At the same time. yielding lz sin θ± = . Consider them separately. all three relevant constants have immediate physical implications. as the coordinate is cyclic.1. as both sides have to be independently equal to a constant. it is useful to determine the action variables next. The ensuing integral is far from trivial.4. (6. As in section 5. To determine I I r lz2 Jθ = dθ Wθ dθ = l2 − dθ (6. Thus. The Kepler problem revisited This equation can be separated. it was shown that αφ = lz . The situation for Jθ is a bit more involved. l To determine the initial and final points requires to ensure that the root stays real. The case of Jφ is straightforward. As the motion is periodic. Using the expression for the angular momentum in spherical coordinates identifies αθ2 = ~l2 = l2 . Thus I I Jφ = dφ Wφ = αφ dφ = 2πlz . q Zθ+ 1− lz2 sin2 θ l2 Jθ = 2ilz2 dθ sin θ θ−  q !. and the result is only cited here1 . to be called αθ . the points of inflection can be obtained from the vanishing of derivatives.2) sin2 θ τθ τθ requires to determine the points of inflection in θ.144 6.13.1) τφ τφ This is the rotation of the whole plane in which the movement takes place at a constant angular velocity.


θ+ lz2 2 r 1 1− l2 sin θ + cos θ lz lz l2 .

= 2ilz2 − ln q + ln cos θ + 1 − z2 sin2 θ .


= 2π(l − lz ). 2 lz2 l l l 1− l2 sin2 θ − cos θ .

θ− even though the final result is deceptively simple. . This is a consequence of the rotational 1 Rather than this brute-force approach. a solution can also be obtained by employing the constancy of each component of angular momentum separately. An interesting observation so far is that the value of k did not yet play a role. The angular motion is completely independent of it.

Chapter 6. Of course. 6. it is useful to consider a particular example. this feature will be reencountered as the Wigner-Eckhart theorem. 2 In a more general context.7. which reads r r 2m 2m Jr = −2πl + πk = −(Jθ + Jφ ) + πk . the bounded motion are periodic. 3 An analytic evaluation is not directly possible. but requires function theory for a solution with rea- sonable effort. in the end this has to play a role.3.2 Spinning top The treatment of rotating bodies with the Lagrangian and Hamiltonian formalism is often needed. From these investigations Kepler’s third law arises as a(n almost) trivial extension.3) −E −E The appearance of the negative of the energy E should not come as a surprise: Only bounded motions are periodic. . The integrand is actually even more cumbersome than before3 . That they are added emphasizes that these cannot be considered as the components of a vector. and it resurfaces now in the equation for Jr .5) and the (not explicitly stated) value of r2 it follows that r r m r23 m τ = πk = 2π − −2E 3 k and thus τ 2 ∼ r22 . To discuss its features. The three frequencies are ∂h 4π 2 mk 2 1 νi = = 3 ≡ν= (6. Because of (6.4) (Jr + Jθ + Jφ )2 which only depends on the action variables. Selected special topics 145 symmetry of the problem2 . and thus only the result will be quoted.3-6. r 4π 2 r 2 r 2mr 2 τr τr As is already known from section 2. I s   I s   k (Jφ + Jθ )2 k l2 Jr = 2m E + − dr = 2m E + − dr. the new Hamilton function is obtained as 2π 2 mk 2 h=− . (6. With this. (6. and bounded motions have negative energies for a potential energy written as k/r. the so-called spinning top. and the values r1 and r2 between it oscillates can be obtained in the same way as before. the system is completely degenerate.5) ∂Ji (Jr + Jθ + Jφ ) τ and coincide.

The movement of the center of mass has nothing to do with the body itself. and that it should be a continuous function of time. as was already used in section 3. Euler’s angles provide exactly this feature. It has therefore only six degrees of freedom. This leaves the rotation of the body. and a vector in the new system is given by ~r′ = J~r. Furthermore. its determinant needs to stay one for all times.1. As an orthogonal transformation. The time-dependent position of the vector ~r′ in terms of the original position are given by making the angles time-dependent.146 6.1 Euler’s angles A spinning top is a rigid body. Very useful in this context are Euler’s angles.2. and is rather following the trajectory of a mass point in which all of the mass is concentrated. It is useful to require that the fixed coordinate system coincides with the body-fixed system at t = 0. rotate by an angle ψ around the original z axis in the coordinate system xzy. The first step is the choice of suitable generalized coordinates. i. it holds that ~r = J T ~r′ . three describing the movement of the center of mass.5). e. rotation axis: There is always a fixed rotation axis for any rotation. But for any real eigenvalue there is a real eigenvector with J(t)~ω (t) = ~ω (t). They are derived from the fact that any rotation in three dimensions can be described by a sequence of three rotations. Given that its determinant is one4 at t = 0.2. First. This gives a new coordinate system ξηz. the system is rotated around the ζ axis by an angle φ. This gives a new system ξη ′ ζ. A number of interesting results can be deduced from the fact that J is an orthogonal rotation matrix. and three describing possible rotations of the body. Spinning top 6. In terms of the matrices (3. Finally. The reason for this is that suitable generalized coordinates need to be independent of each other.2. as otherwise not all constraints have been eliminated. and thus ω(t) does not change under J. As will be seen. J(t = 0) = 1. At every instant in time. Euler’s angles are therefore introduced in the following way: Consider a fixed coordinate system. In this coordinate system. the relative orientation of the body is then fixed by three consecutive rotations. rotate around the ξ axis by an angle θ.8. a complete rotation reads J = J1 (φ)J3 (θ)J1 (ψ). . possibly time-dependent. It is thus the. it has at least one real eigenvalue of value +1. which is known as Euler’s 4 It could also be −1. which happens if the two coordinate systems are mirrored with respect to each other. as discussed in section 2.

.2 The equations of rotation Start with the Lagrange function. dφ dψ cos θ dt + dt The derivatives of the potential with respect to Euler’s angles will in general not be a component of the torque. and therefore do not define a physical eigenvector.11.2.8) dt dω2 I2 − ω1 ω3 (I3 − I1 ) = M2 . However. (6.8-6.6) 2 2 i where the angles are still Euler’s angles. Euler’s equations can actually be solved if the torque vanishes in their full generality in terms of so-called elliptic functions.9) together are known as Euler’s equations. (6. the choice of z-axis was arbitrary. Thus its Euler-Lagrange equations reads dω3 d ∂T ∂T V I3 − ω1 ω2 (I1 − I2 ) = − =− = M3 . this could be done by exchanging the axes. they correspond to the flipped rotation axis. The exception is ψ. Note that the angle actually no longer appears explicitly in the equation of motion. The other two eigenvalues can only be real. The connection of the rotational speed to Euler’s angle is given by   sin θ sin ψ dφ dt + cos ψ dθ dt ~ω = sin θ cos ψ dφ   dt − sin ψ dθ dt  . to maintain the value of the determinant at time zero. the tensor of inertia. neither of this is fully transparent.9) are not the Euler-Lagrange equations of the other Euler angles. it is not necessary to resort to them.9) dt (6. Otherwise. and it is better to study some special cases. or both are −1. only the generalized speeds. Assume that a coordinate system has been chosen in which the tensor of inertia has been brought to its main axis form.7) dt dt ∂dt ψ ∂ψ ∂ψ The other two equations would have quite a different form.Chapter 6. At the same time. If they are −1. and (6. (6. and are therefore nothing new. Selected special topics 147 theorem. ψ). The kinetic energy for a rotation involves. Then the Lagrange function reads ! 1 1 X 2 L= = Ii ωi − V (θ. 6. as this is the rotation associated with the fixed z-axis. leading to dω1 I1 − ω2 ω3 (I2 − I3 ) = M1 (6. Still. even the top’s qualitative behavior can be deduced without solving them. Therefore. In fact. if J(t) = 1. However. they are valid equations of motion. φ. as discussed in section 2. they are complex. to identify the most pertinent features.7-6.

3 Precession The first interesting case occurs for a symmetric object with I1 = I2 . Likewise. is   A sin(Ωt)   ~ω = A cos(Ωt) ωz I1 − I3 Ω = ωz .2.8) once more with respect to time. This is the behavior of a (spinning) top in a gravitational field.2. ∂L pψ = = Iz (dt ψ + cos θdt φ) = I3 ωz = Ix a ∂dt ψ dL pφ = = (Ix sin2 θ + Iz cos2 θ)dt φ + Iz cos θdt ψ = Ix b. This behavior is called pre- cession. Spinning top 6.9).4 Top in a gravitational field Consider as a more complicated case again an object with a symmetry axis. In this case the Lagrange function (6. but now under the influence of a constant force.6) reduces to Ix  Iz L= (dt θ)2 + sin2 θ(dt φ)2 + ((dt ψ) + cos θ(dt φ))2 − Mgl cos θ. 6. ∂dt φ . this is achieved by deriving (6.148 6.8-6.9) implies that ω3 is constant in time. a similar equation arises for ω2 . dt dt I1 This is the equation of a harmonic oscillator. 2 2 where M is the mass of the top and l is the distance between the point where the top touches the ground and its center of mass on its symmetry axis. for suitable initial conditions. This gives two integrals of motions. It is immediately seen that both ψ and φ are cyclic. I1 The rotation axis therefore rotates itself around the z-axis.2. The solution. yielding d 2 ω1 dω2 (I1 − I3 )2 2 I1 2 = ω3 (I1 − I3 ) = − ω3 ω1 . This allows to eliminate either ω1 or ω2 when combining (6. which in the form of a gyroscope is one of the most important practical applications of this problem. and therefore torque. Choosing ω1 . Then (6. which are two components of the angular momentum.

Fortunately. (6. (6. similar to the case of the central potential in section 2.7. the three angles as a function of time. Selected special topics 149 where the recasting in terms of new constants a and b will simplify calculations below. yielding b − a cos θ dt φ = .11). Fortunately. leading to Zu(t) du t − t0 = p . and therefore only differential equations of first order need to be solved.10). also the energy is constant. (6. the final result is once more not very instructive. reading (dt u)2 = (1 − u2 )(α − βu) − (b − au)2 . The first step is to single out dt φ from the combination of equations. reading (dt θ)2 sin θ = sin2 θ(α − β cos θ) − (b − a cos θ)2  2  2 E − Iz2ωz α = Ix 2Mgl β = . it is possible to use the three integrals of motion for that. (6. to be determined. . it is possible to read off the qualitative behavior already without performing the integral explicitly.13) (1 − u2 )(α − βu) − (b − au)2 u(t0 ) Thus. the problem is fully solved by the equations (6.13). Similarly. and (6.10) sin2 θ and thus φ can be obtained from θ. For this. Since the system is conservative. Ix  Iz E =T +V = (dt θ)2 + sin2 θ(dt φ)2 + ωz2 + Mgl cos θ. it is possible to combine the three integrals of motion such that an equation involving only involving θ is obtained.12) is performed.13). it is useful to consider the differential form of (6.11) Iz sin2 θ and thus also ψ is known once θ is.Chapter 6. Ix a b − a cos θ dt ψ = − cos θ. While this integral can be solved using elliptic functions. Ix This equation can be solved by direct integration if the substitution u = cos θ (6. 2 2 This leaves only three quantities. However.

described by a mass density ρ(~r). 6.12). that of many (coupled) oscillators. this actually corresponds to the fact that θ = π/2. e. this implies that the right-hand-side goes to ±∞ for u → ±∞.3. and the rotation axis performs loops due to the nutation. In this case. where the loop degenerates to a point. Furthermore. giving an edge in the path of the rotation axis. as it leads to closed expression for the frequencies of the movements as a function of the gravitational field. only the interval −1 ≤ u ≤ 1 is relevant. and therefore the top’s symmetry axis cannot fall to the floor. This is the purview of continuum mechanics. the right-hand-side is necessarily negative or zero. as already hinted at in section 2.13). a second movement overlays with this. 6. This is the precession of section 6. and therefore plays no role. it is very helpful to first consider a very special system with many degrees of freedom. Finally.150 6. it can move not only forwards. as are the constants α and β.3. and thus the total effect is called precession-nutation. One is that the symmetry axis always moves forward in θ. In many of these cases it becomes suitable to deal with these particles rather in terms of density than of the individual particles. but also backwards in θ. there are three possibilities. Alternatively. This is called a nutation. However. This leads beyond the scope of this lecture. this implies that one of the roots has to be at a location u > 1. in various limits. This can be made more quantitative. Continuum mechanics The right-hand-side is a polynomial of third order.3 Continuum mechanics Many systems in mechanics contain a very large number of point particles. This is particularly important for practical application. Considering the curves which the symmetry axis exhibits. . and rotates around the direction of gravity. Because the left-hand-side is positive. This implies that there are necessarily three real roots. if the spin has much more or much less kinetic energy than potential energy.8.2.1 Systems of many oscillators Before actually treating matter as a continuum system. g. there is a limiting case between both other options. as there the rate of change of u vanishes. Since the roots are the points of reflection. without actually solving (6. and therefore can also give information on geographical position. this implies that u can move in general such that −1 < cos θ < 1. matter is considered as a continuum. If both are negative. since the other angles also moves. One is that the symmetry axis oscillates up and down in some range. and the top is upright.3. If both are zeros. If u = ±1. because of (6.

Selected special topics 151 Such systems in themselves are relevant. Considering only conservative systems with the kinetic energy a quadratic function in the generalized speeds. Moreover. any system close to a stable equilibrium is also described by a system of many oscillators. if there is a minimum of the potential in terms of the generalized coordinates. Also in generalized coordinates stable equilibriums are encountered. as many physical systems can be described by them. taking the limit of an infinite number of these oscillators already provides a good description of many continuum systems. the total Lagrange function reads dηi dηj L = mij − Vij ηi ηj dt . such system.Chapter 6. For this to happen. if the system should remain close to equilibrium also the kinetic energy needs to be close to zero. especially in solid state physics. and it is therefore technically useful to first study a finite. It is then also possible to expand the potential in a Taylor series in the generalized coordinates. Moreover. Furthermore. Taking for N generalized coordinates qi the values qi0 to be the equilibrium position. it is not necessary that the system is described in position space coordinates.9. but large. according to section 2. it is then useful to introduce the fluctuation coordinates ηi = qi − qi0 .

dt ∂ 2 V .


∂qi ∂qj . Vij = .

Because each individual equation has the same form as an harmonic oscillator. Intentionally here only the same frequency is used for the whole set of oscillators. an ansatz of type  ηi = ℜ ai e−iωt appears as a reasonable starting point.qj =qj where the constant term in the potential has been dropped and it was permitted that the kinetic term has also off-diagonal elements in the generalized coordinates. 0 0 qi =qi . The different . Fortunately. this type of coupled differential equations can be solved in closed form. (6. it appears possible that there are common oscillations of all oscillators.6 that an harmonic oscillator under an external influence starts to behave like the external influence.14) dt and therefore form a set of N coupled differential equations. Thus. This is based on the insight from section 2. The Euler- Lagrange equations are d2 ηj mij 2 + Vij ηj = 0.

Especially. mij = mi δij . it is possible to reinterpret the problem as a problem in a space which is not Euclidean. which implies that one of the (non-vanishing) amplitudes will be arbitrarily selectable. 5 The latter will not appear. where λ = ω 2 are the eigenvalues. as any symmetric matrix can be diagonalized. The vectors of amplitudes are then the eigenvectors of the matrix V . which thus yields 2N solutions.14) into a set of linear equations for the amplitude factors ai . V ~a = λ~a. producing all the solutions. Vij aj − mij ω 2 aj = 0. and the problem takes the form of an eigenvalue problem. If the matrix m is not diagonal.3. the eigenspaces to a given eigenvector are not one- dimensional. This gives an Nth order polynomial for the frequency ω. but rather has a metric m. (6. ultimately leading to equivalent results. If there are degenerate eigenvalues. This implies that all eigenvalues are real. which is necessarily also a canonical transformation.16) vanishes.152 6. These amplitudes are also called eigenoscillations of the system. Continuum mechanics possible solutions will eventually be found to be different values for ω. Note that this also implies that there exists an orthogonal transformation.15) Such a system of equations has only a solution if the determinant of the matrix Bij = Vij − mij ω 2 (6. and the remaining follow from the solutions of the linear system. In these cases it is necessary to construct a basis within these eigenspaces. also the complex solutions are admissible. Furthermore. This turns the system of differential equations (6. as these will be exponentially damped or enhanced (unstable) oscillations5 . Newton’s third law requires that the V are symmetric. the result are again 2N eigenvalues ±ωk and N eigenvectors ~a with arbitrary norm in this metric. In this case. which decouples all oscillators. which then can be superposed to find arbitrary solutions. and form a complete basis and are orthogonal with respect to each other. it is possible to construct a very similar line of reasoning. . An interesting reinterpretation is possible if the mass matrix is diagonal. with no summation implied. which follows from the general theory of linear systems. and thus the so-called eigenfrequen- cies ω 2 are either purely real or purely imaginary. the masses drop out. In the end. if the initial assumption of a system in equilibrium is not violated. The amplitudes are the solutions to this homogeneous system. every time ±ω. Rescaling the ηi by mi .

and define the normal coordinates ζ such that η~ = Aζ~ ζ~ = A−1 ~η. . It is very important to make this distinction carefully.14) can be constructed in the same way6 . 6 Note that there can be no more than 2N solutions. In the equilibrium case it is therefore a positive semi-definite matrix. Define the matrix Aij = aji from the normalized eigenvectors ~ak of B. where the matrix λ has the eigenvalues λk = ω 2 as entries. Selected special topics 153 In the end. either in the form of some combination of real and imaginary parts of Ck± or as amplitude factors fk and phase shifts δk . They belong to every of the N oscillations. as guaranteed by the mathematical theorem on ordinary differential equations. the normal coordinates are again a set of generalized coordinates. one can be sure that all solutions have been captured. The general solution will therefore be X  X i ηi = ℜ aki Ck+ eiωk t + Ck− e−iωt = ak fk cos(ωt + δk ). Therefore having constructed all 2N solutions. if also the generalized speeds are transformed accordingly. For each of these oscillations the eigenvectors ~a determine the relative amplitudes of the oscillators. if this is not the case. as any rotation in configuration space is a canonical transformation. Therefore. If this would have been rather a partial differential equation. and this also implies that the inverse exist and that A−1 = AT . It should be noted that the eigenvectors determine the relative ratio of the amplitudes of the oscillators. and are determined by the initial conditions. and therefore also in this case all solutions to (6. It is useful to introduce normal coordinates. especially a ro- tation.Chapter 6. this matrix also diagonalizes V to its diagonal form. but not the absolute value. the total dimensionality does not change. or may be only of limited periodicity. Furthermore. k k where 2N undetermined constants remain. The eigenvectors ~ak of a symmetric matrix form an orthonormal base. any orthogonal matrix defines a coordinate transformation. The matrix A is therefore orthogonal. a similar statement is in general not possible. Note that the motion may be proper periodic if all the frequencies have rational ratios. λ = AV AT . and are not determined by any initial conditions. These are a property of the system. By construction.

This is actually also possible if the matrix M is not diagonal and scaled out. there would be no need to consider equilibrium positions. Important here is. every oscillating according to the corresponding eigenfrequency ωk2 . In these coordi- nates every oscillations is also called normal oscillation. Comparison to section 5. though orthogonality and rotations have then to be defined with respect to a metric M. any number of these normal oscillations can be excited. the potential energy is k k V = (x2 − x1 − a)2 + (x3 − x2 − a)2 . g.154 6. ~ 2 dt dt 2 2 dt dt 2 where a rescaling of the fluctuations η by the mass has been performed. This picture is very helpful when introducing continuous systems. 2 2 As the equilibrium position7 will be that the springs are at the normal length a. as the potential is already of harmonic oscillator form. the normal oscillations. the difference between the equilibrium positions is also a. electromagnetism or other field theories. Take e. but only movements are possible which can be decomposed into normal oscillations.13. 2 2 7 In the present case. The particles at the ends should have a mass m and the one in the center a mass M. Geometrically. it is possible to consider the physical picture: The system is able to perform a number of movements. As the full movement is now given by a composition of normal oscillations as dictated by the initial conditions. The system is only able to respond in a certain. in normal coordinates all oscillations of the ζ decouple. Depending on the initial conditions. Thus. and also quantum systems.3. and the system is described by N decoupled oscillators. this transformation rotated into the coordinate system created by the eigenvectors. three particles.4 shows that this is also the form which would be obtained after applying Hamilton-Jacobi theory to the system. all confined to a linear motion in one dimension. . connected by two springs with the same strength k. and in terms of the fluctuations ηi the potential energy takes the form k k V = (η2 − η1 )2 + (η3 − η2 ). It is useful to consider an explicit example. however. For the sake of sticking with the developed procedures this will be done nonetheless. Continuum mechanics This implies that it is possible to rewrite the Lagrange function as  T !T 1 ~η d~η 1 T 1 ζ~ ζ~ 1 L= M − ~η V ~η = − ζ~T λζ. If the length of the springs is a. only that it is always possible to recast the Lagrange function in this form. strictly speaking. quasi-quantized form.

say. e. 0 −k k − ω2m This form .15) to find the relative sizes. Selected special topics 155 The corresponding kinetic energy is m  M T = (dt η1 )2 + (dt η3 )2 + (dt η2 )2 . m M The first frequency appears somewhat odd. which yields the (normalized) solutions   1     q 1 1  2m(1+ 2m M ) 1 1  −2  ~a1 = √ ~a2 = √ ~a3 =      1  0 . this is a consistent solution.Chapter 6. q M  2m + M 2m   1 −1 q 1 2m(1+ M ) 2m . Thus.elements on the diagonal and otherwise only elements on directly adjacent subdiagonal. i. Since the constraints do not forbid it.16) is then   k − ω2m −k 0   B =  −k 2k − ω 2 M −k  . the behavior obtained in the following is qualitatively already rather characteristic of such systems. This is an ordinary system of linear equations. there is a comparatively simple explanation for it. Such systems play a certain role in many physical contexts. is characteristic for a one-dimensional system of springs. leaving only non-vanishing frequencies. especially in solid state physics and optical lattices. However. This is a uniform motion of the whole system in either direction. The next step is to solve the equation 0 = det B = ω 2 (k − ω 2m)(k(M + 2m) − ω 2 Mm) for ω 2 . ω1 = 0 r k ω2 = sm   k 2m ω3 = 1+ . There are three solutions. where only nearest neighbors are connected. solve B~ak = 0. the center-of- mass coordinate needs to be fixed. The next step is to solve (6. as this would be no oscillation at all. This would have reduced the degrees of freedom from three to two. 2 2 The matrix B from (6.  2M (2+ m )  . This could have been excluded by adding a constraint that.

i. as then the symmetry transformation needs to leave the motion unchanged. i 2 2 In principle.3.6. the so-called module of Young Y .156 6. in general it will become more involved to determine them. as a short spring has a larger spring constant. but which should all have the same mass.3. The Lagrange function is then given by X m k  2 2 L= (dt ηi ) + (ηi+1 − ηi ) . As a next step consider again the linear chain. The non-trivial. If now the equilibrium distance between any two particles is equal and of size a. The second is an oscillation of the endpoints with respect to a fixed center. but rather by their position.2 Continuous systems of oscillators After this preliminary investigation. η(x) is the displacement of the particle with equilibrium position x against its equilibrium . just like m/a. However. The quantity ka has no direct interpretation. which leaves also the center of mass fixed. there need to be some additional terms to take care of boundary terms. one lesson is that always the motion of the center of mass will appear. The third is a relative oscillation of the center mass with respect to the two other masses. The first case is indeed a collective motion in one direction. especially as later the limit N → ∞ will be taken. non-degenerate eigenfrequencies then fully characterize the system. which is in general not possible in closed form. Further degeneracies will occur if the system shows a symmetry. Here. and will therefore not be done here. Now label the particles not by a discrete index. but now admit an (arbitrarily large) number N of particles. it will be required by definition that ka = Y . there will always be 3N − 3 non-trivial eigenfrequencies. This leads to no new conceptual insights. is held constant for any value of a. if not constrained. e. 6. If the connection is indeed in the form of springs. It is possible to extent the present treatment to generalize the full case with dissipation and external driving forces. this can be rewritten as  2 ! X X m ka ηi+1 − ηi L= a (dt ηi )2 + = aLi i 2a 2 a i The quantity m/a = µ is the mass density of the system. they will not play a role. as in section 2. it is now possible to go to continuous systems. Thus. Continuum mechanics where the normalization is chosen for later convenience. Generalizing to masses connected by effective springs in three dimensions. it turns out that for usual springs this product is constant. However. as it requires to solve an 3N − 3 order polynomial.

it would also be possible to define vector fields or fields of even higher tensors. Z Z S = dtL = dtd3~rL. This already suggests that an extension to special relativity is rather straight-forward.Chapter 6. and assigns in such a context the Lagrange density a more fundamental role than the Lagrange function. a a dx i. the equation of motion turns from md2t ηi − k(ηi+1 − 2ηi + ηi−1 ) = 0 to µd2t η(x) − Y d2x η = 0 (6.3. t). ∂t η. For the Lagrange function to be scalar. ~r.17) which is now an equation for the function η(x. At the current time.3 Continuous systems The previous section outlined everything necessary to define the Lagrange formulation for continuous system. the Lagrange density needs to be a density. Z L = d3~rL(∂i η. At the same time the sum becomes a Riemann sum. where L is called the Lagrangian density. or rate of change. sometimes called just Lagrangian. Then ηi+1 − ηi η(x + a) − η(x) a→0 dη = == . . for simplicity here only scalar fields will be considered. the difference in fluctuations becomes in the limit of smaller and smaller distance a of the particles the derivative. leading to Z Z 2 2  L = dx µ(dtη) − Y (dx η) = dxL(x). e. the Lagrange function is given by a volume integral over the Lagrange density L. rather than for the finite number of quantities ηi (t). and indeed this will already in classical electromagnetism be necessary. and thus a space-time integral over the Lagrange density. t) which can now depend explicitly on both time and the space coordinate. However. in the displacement η. η is itself also a scalar under rotations. Given a field η(~r. Likewise. The action is still a time-integral over the Lagrange function. However. Selected special topics 157 position. t) of two variables. 6.

∂∂i η ∂∂i η dri ∂∂i η Some comment is in order concerning the appearance of total and partial derivatives. however.3. The field η does only depend explicitly on the coordinates and the time. Integration and differentiation do only invert each other for full derivatives. but only something with performing partial derivatives.19) To rewrite all contributions in terms of δη note that the variations at the boundaries vanish.158 6. but the coordinates and time do no longer depend also on any other quantities. and can therefore not be directly integrated. Thus. At the same time it is now necessary to consider also the changes in direction. Thus. x)η(x)|21 = dx (a(η(x). these have been applied to the coordinate at fixed time. the Lagrange density can depend on them explicitly. This has nothing to do with physics. Thus.18) ∂η ∂∂t η ∂∂i η In very much the same manner as in section 5. Continuum mechanics However. (6. Z2 Z2   d da dη a(η(x).1 it is now possible to derive Euler-Lagrange equations by requiring δS = 0. an important difference now occurs when considering virtual displacements. which leads to the requirement to perform a total derivative. Then Z Z Z ∂L ∂L d ∂L dt δ∂t η = dt ∂t δη = − dtδη ∂∂t η ∂∂t η dt ∂∂t η Z Z Z ∂L ∂L d ∂L dt δ∂i η = dt ∂i δη = − dtδη . of the field. derivatives can be moved around freely. 8 For simplicity now ∂t = ∂/∂t and ∂i = ∂/∂xi will be used as abbreviations. x)η(x)) = dx η+a . not partial derivatives. at fixed time and coordinates. which implies Z dtd3~rδL = 0.2. . (6. virtual displacements modify the field. a variation of the Lagrange density under a virtual displacement reads8 ∂L ∂L ∂L δL = δη + δ∂t η + δ∂i η. not only in time. Now. total and partial derivatives coincide in this case. the elemen- tary degree of freedom is the field rather than the coordinates. x)η(x)) = dx η+a dx dx dx 1 1 Z2   Z2 Z2   ∂a ∂a ∂η ∂η ∂ ∂a ∂η = dx η+ η+a 6= dx (a(η(x). Thus. η → η + δη. So far. It is thus necessary to take this into account when performing a partial integration. However. ∂x ∂η ∂x ∂x ∂x ∂x ∂x 1 1 1 The last term is not a total differential.1. as in section 5.

This can be generalized to a functional calculus. Selected special topics 159 Returning to (6. except for making expressions more compact. but of more limited use. In this case. in classical mechanics. both space and time appear on equal footing. (6. one for each field.18-6. It is sometimes useful to define a functional derivative δF [η.1.21) δη ∂η dri ∂∂i η where Z F = d3~rf (η. ~r) implying that the Euler-Lagrange equations can be written as d δL δL − dt δ∂t η δη where use has been made of the fact that L does not depend on ∂t ∂i η. where only a set of functions of a single variable needed to be solved. dxµ ∂∂µ η ∂η This is one of the reasons the Lagrangian formulation is particularly useful for relativistic problems. Note that this form is already Lorentz covariant.Chapter 6.2 the continuum Euler-Lagrange equations can now be read off d ∂L d ∂L ∂L + − = 0.20) dt ∂∂t η dri ∂∂i η ∂η which are the dynamical equations governing continuum mechanics. provided both η and L are Lorentz scalars. It is therefore substantially more complicated as the point-particle case. this is a functional differential equation. ∂η dt ∂∂t η dri ∂∂i η Just as in section 5. yielding a partial differential equation for η in its four variables. and can then be written as d ∂L ∂L − = 0. this generalizes to a set of such partial differential equations. As η is now a function. If there is more than one field involved. . which is of eminent importance in quantum field theory.19) then leads to Z   3 ∂L d ∂L d ∂L dtd ~r − − δη = 0. ~r] ∂f d ∂f = − (6.

2 where µ is again the density. this only motivates why this particular case could be of . This is a one-dimensional wave equation with the solutions   i vt−x −i vt−x η = ℜ c1 e x0 + c2 e x0 s Y v = . Its Lagrangian density was 1  L= µ(∂t η)2 − Y (∂x η)2 . In Principle. Here. on which it reacts with some movement in the three spatial directions.160 6. µ which describe forward and backward propagating waves.17). 2 The corresponding derivatives are ∂L = µ∂t η ∂∂t η ∂L = −Y ∂x η ∂∂x η ∂L = 0. To study a more complex problem consider a fluid or gas. is given by a vector field η~ . which gives at every point in space (and time) the displacement of the medium. is given by ~η )2 − 2P ∂~ V = γP (∂~ ~η where P is the equilibrium pressure and γ characterizes how the medium’s pressure reacts to changes of volume. Here. such that the mass density can be taken to be approximately constant.3. Without proof.22) which coincides with (6. (6. The kinetic energy is therefore µ T = (dt ~η)2 .3. the case of only small distortions will be considered.4 Applications of the Lagrange formulation Consider again the problem of section 6. and thus how the other particles oppose the movement of a particle. Consider now it to be slightly disturbed. this density would need to be also a field. ∂η Its Euler-Lagrange equation is therefore µ∂t2 η − Y ∂x2 η = 0. This movement field. the potential energy.2. what is perceived as sound.3. which is essentially sustained by the presence of the medium. Continuum mechanics 6. however.

Selected special topics 161 practical interest.23) This is not yet a simply solvable system. not where it comes from.22). (6. The rate of change of the movement of the gas therefore forms waves. it make sense that there will be only one relevant direction. This is a generic statement: Terms linear in derivatives of the fields do not alter the movement. with a speed depending on the properties of the gas and its pressure.. But the focus is how to treat this system. L=T −V = (∂t ~η )2 − γP (∂~ 2 To obtain the equations of motions requires the following quantities ∂(∂t ~η )2 = 2∂t ηi ∂∂t ηi ~η ∂ ∂~ = δij ∂∂i ηj ~ η )2 ∂(∂~ = ~η 2δij ∂~ ∂∂i ηj ∂L = 0. this is a wave equation. However. The equations of motion are thus   µ∂t2 ~η − γP ∂~ ∂~ ~ η = 0.23). This is as expected: Sound is an oscillatory wave. g. (6. To rewrite the problem in this form. The Lagrangian density is therefore µ ~η )2 + 2P ∂~ ~ η.Chapter 6. ∂ηi ~ η does not contribute to the Interestingly. and therefore a single scalar field should actually be sufficient to solve the problem. σ = −∂~ which is the directed rate of change. it cannot be ignored. yielding µ 2 ∂2σ − ∂ σ = 0. . define the scalar quantity ~η. γP t Just as before. e. To make progress. this result implies that the term linear in ∂~ movement. act with ∂~ on the equation of motion (6. if the energy stored in the system should be determined. This is the purview of a thermodynamic lecture. As the system is essentially isotropic.

3.25) δπ ∂π ∂H ∂L = . or just Hamiltonian.4.162 6.3. Especially in the context of special relativity the Hamilton formulation is not manifestly covariant. . This is.5 Hamilton formulation The Lagrange formulation for continuous systems treats space and time on very much equal footing. the Hamilton function is a volume integral of the Hamilton density Z H = d3~r (πdt η − L) . though of course it is still correct. of course. which are defined using a time-derivative.26) ∂t ∂t In comparison to (6.6 define ∂L δL π= = . also a field. which always uses time as a particular quantity? Can it be useful? To reach the Hamilton formulation from the Lagrange formulation of section 6.3 requires to define the canonical momenta for a field. Continuum mechanics 6. if there a multiple fields there will be for each field an own momentum density.3. Only more tedious in most cases. ∂∂t η δ∂t η which makes only reference to the time-derivative.4 us- ing fields and also derivatives with respect to the coordinates to find the corresponding equations of motion. In analogy to section 4. a Hamilton density. it is sufficient to trace out the steps of section 5.24) δη ∂η dxi ∂∂i η δH ∂H ∂t η = = (6. and therefore breaks the equal treatment of the Lagrange formulation.20) the difference between the Hamilton and the Lagrange formulation becomes now manifest: Derivatives with respect to the momentum density. Just as with the Lagrange function. as in section 5. What is then about the Hamilton formulation. (6. Just as with the Lagrange case. Skipping these details leads to the continuum Hamilton equations of motion. the so-called momentum density π. Define then. Of course. differ in the absence of spatial derivatives from those of derivatives with respect to the fields themselves. δH ∂H d ∂H −∂t π = = − (6. H = πdt η − L.

Z G = d3~rG.26) and the definition of the functional derivative (6. consider the gas or fluid of section 6. of course. Given two trajectories at some . 2µ 2 where the irrelevant linear term has been skipped. H} + dt δη δπ δπ δη ∂t ∂t which defines also the continuum version of the Poisson brackets. Comparing this expression with (6. and lead to the same result.Chapter 6.21) permits to rewrite this as Z   dG 3 δG δH δG δH ∂G ∂G = d ~r − + = {G. As an important example consider the continuum version of the evolution equation (5.4 Chaos A particular subclass of physical systems are those which exhibit a so-called chaotic be- havior.3. It takes the form Z   dG 3 ∂G ∂G ∂G ∂G ∂G = d ~r ∂t η + ∂i ∂t η + ∂t π + ∂t ∂i π + dt ∂η ∂∂i η ∂π ∂∂i π ∂t Z      3 ∂G d ∂G ∂G d ∂G ∂G = d ~r − ∂t η + − ∂t π + .43). consider its total time derivative. Chaotic behavior is best defined in phase space. 6. The momentum density is ∂L πi = = µ∂t ηi . which are. Most of the formal developments of chapter 5 can be repeated for the continuous case essentially completely analogous. equivalent to the Lagrange case. ∂∂t ηi The Hamilton density is ~π 2 P γ ~ 2 H = ~π~η − L =+ (∂~η ) . Given some function G. Selected special topics 163 As an example.24- 6.4. The corresponding equations of motion are ~π ∂t ~η = µ   ~ ~ ∂t~π = ∂ P γ ∂~η . which is a volume integral of a density G depending on the field and momentum density. ∂η dxi ∂∂i η ∂π dxi ∂∂i π ∂t where partial integrations have been performed. even though it would be needed to interpret the Hamilton density as the energy density of the system.

This is still classical mechanics. It is actually not necessary that a system shows the behavior (6. The reason why such a behavior is said to be chaotic is that any error in the determination of the trajectory will therefore be exponentially amplified over time. initial conditions. constant. 9 There exists actually a mathematically more precise definition of what a chaotic system is. the attractor is called strange. can show chaotic behavior. as in any practical (experimental) situation there are always errors. for explicitly time- independent systems. If they do so without forming any kind of periodicity. the attractor can become a point in phase space. Chaos is therefore a feature of the unavoidable measurement errors rather than a problem coming from theory. Thus a prediction becomes essentially impossible in practice. which leads too far beyond the scope of this lecture. It can well be that the system is chaotic for some cases of initial conditions. especially those of many degrees of freedom. and therefore completely deterministic. The simplest such system is actually a double pendulum. As many practically relevant systems. This is quite different for quantum chaos. attractors. It is also possible that after some time of exponentially diverging trajectories come again close to each other. g. . though the volume may become exponentially deformed. Chaos point in phase space at some time t. where quantum effects can introduce addi- tional chaotic behavior. and therefore the phase space density can change. nor for all times.12 still holds for collections of trajectories. this subject is of great importance. i. it is possible to define an Euclidean distance between two trajectories as d(t) = (qi1 − qi2 )2 + (p1i − p2i )2 . e.. its detailed study warrants an own lecture. These are localized regions in phase space at which multiple trajectories. the density in phase space remains. e. Thus. but not all. but for the present purpose this is sufficient. A system is said to show chaotic behavior if the distance increases exponential (or faster) as a function of time9 . coalesce. but as soon as the kinetic energy exceeds a critical value it becomes chaotic.164 6. Classifying such features of a system is the purview of chaos theory. which is called a fixed point. For small oscillations it shows no chaotic behavior. and it will not be continued here. Note that also Liouville’s theorem of section 5. Interesting features of chaotic systems are. If the system is dissipative. d(t) ∼ eat .27) where a is some constant characteristic of the system. even from completely different initial conditions. (6.4. and stay there.27) for all pairs of possible trajectories. Due to its large number of phenomena and features.

29) −∂qi H . λ∗ and 1/λ∗ . Selected special topics 165 6. in phase space. Note that phase space is always even-dimensional. and reappear in one way or another in all fundamental theories known. which will not be needed here. and thus Hamiltonian mechan- ics. which warrants some thoughts. and the two vectors ! qi ~v = pi ! ∂qi ∂~ = . It can be shown that for the submatrices of the matrix A ! a b A= c d holds that (aT c)T = aT c (bT d)T = bT d aT d − cT b = 1. Consider now a not explicitly time-dependent Hamilton function H. Also. each of dimension n. Note for further use that J −1 = J T = −J and J 2 = −1. Sp(2n). quite general statements about the multiplicities of eigenvalues can be made. This is especially the case as geometric structures are prevalent in physics. if λ 6= 0 is an eigenvalue of A. ∂pi where the subvectors are n-dimensional.5 The geometry of phase space Classical mechanics has actually an interesting geometric interpretation. so is 1/λ. Construct the vector ! ~ = ∂qi H J ∂H . with dimension 2n in the following. These are the matrices A with the property AT JA = J (6. This will become most evident when using phase space. (6.Chapter 6. Furthermore.28) ! 0 1 J = . where the last two can coincide with the first two if λ is real. In such spaces there exist a particular group of matrices. −1 0 where the elements of J are either zero matrices or (negative) unit matrices. the so-called symplectic matrices.

symplectic velocity fields J ∂G (infinitesimal) canonical transformations in phase space.12) take the compact form of a vector equation ~ ∂t~v = J ∂H. where the function G is the deviation of the generating function F2 from the unity trans- formation. The geometry of phase space which contains Hamilton’s equations.29) describes the trajectory in phase space.12. creates by Hamil- ton’s equation the movement. consider the total time derivative of H. Hamilton’s equation (5. In addition. As has been noted there. ∂Qj ∂Pj . Thus. Now. ~ T ∂t~v = (∂H) dt H = (∂H) ~ T J ∂H ~ = 0. which was actually a canonical transformation. the gradient of the Hamilton function in phase space. There it was shown that movement along a trajectory is created by the Hamilton function. Thus. any canonical transformation is given similarly by ~ dv = J ∂G. and the conservation of the Hamilton function along the trajectories is a geometric consequence of this. ∂H dqi = ∂pi dt ∂H dpi = − ∂qi dt.166 6. As a first example how this formulation could be advantageous. Thus. Especially. combine these insights with those of section 5. time evolution was anyhow only ~ create all a special case of such a transformation.there it is a gradient in ordinary space which creates by Newton’s equations the movement of the system. supplemented by the matrix J. It is even possible to state more. It is a statement about the trajectories generated by the function H.5. the expression (6.11-5. rather than any other derivatives. This is quite similar in spirit to what the potential does .12. Here. the time-independence of the Hamilton function becomes now a geometric feature of phase space. Consider the matrix of derivatives of the canonical transformations ! ∂qi ∂qi ∂Qj ∂Pj M= ∂pi ∂pi . as this hinged only on the features of the matrix J. as defined in 5. This implies that a trajectory develops in phase space as ~ dv = J ∂Hdt.

6. To give this a further twist.52). Because such a scalar product is invariant under canonical transformations. Especially. as noted above. wj } = −Jij .6 The relativistic string In many areas of physics string-like objects appear. A space endowed with such a metric is called a symplectic space or symplectic manifold. it follows that ∂Q ! i ∂Pi −1 ∂qj ∂pj M = ∂Qi ∂Pi . with v and w two vectors in phase space is invariant under any canonical transformations. the Poisson bracket acts as directional derivative with respect to the flow ∂F . (6.46) in fact define the metric of phase space. it is interesting to consider a rela- tivistic string.28). Thus. it is a scalar product in this symplectic space. as this is equivalent to (6.44-5. But this just defines a symplectic transformation. Also the Poisson brackets can be integrated into this formulation.Chapter 6. G} = −(∂F )T J(∂G). Selected special topics 167 Because the canonical transformations are invertible. obtained in section 5. Therefore.10. the invariance of Poisson brackets under canonical transformations. the scalar product vw = v T Jw. {F. It is therefore useful to understand their classical behavior. . It then follows by explicit evaluation that −JM = (JM −1 )T holds.50-5. Since in (6. As a consequence. J = M T JM. follows here trivially from the geometry of phase space.30) the expression J∂G defines a flow along a canonical transformation.30) Thus. a finite canonical transformation can be characterized by symplectic matrix. ∂qj ∂pj Add into this the relations (5.11 a geometric meaning. {vi . the fundamental Poisson brackets (5. This gives also the identification of the Poisson bracket as a differential operator in section 5.

Xµ = Xµ (σ. This extension can be infinite or finite.6. The functions Xµ describing the position of the points of the world sheet in space-time are therefore functions of both parameters. as a generalization of hτ τ = Ẋ µ Ẋµ .1.19) √ Geometrically. at any fixed eigentime τ the string has an extension. 2πα 2πα′ again the direct generalization of the point-particle action.6... . its ends are connected. . Both parameters together describe any point on the world sheet.2.. Derivatives with respect to the two parameters will be counted by Latin indices a. = ∂τ . This Regge slope can shown to be associated with the string tension T as T = 1/(2πα′)... which had been used to define the action (5.b.168 6. τ ). which can be written down for this system.31) such that the position of the world sheet is not depending on the parametrization. ∂σ ∂0 = ∂τ ∂1 = ∂σ . or open. as for the point particle of section 5. analogously to the problem of minimizing the surface of a rotating object in section 5. τ ′ (σ. Furthermore.2. the string can be closed. The constant α′ is the so-called Regge slope. the world line becomes a world sheet. In the latter case. τ )) (6... i. these functions should be reparametrization invariant X µ (σ. is the Nambu-Goto action Z SN G = dτ dσLNG M in which M is the world-sheet of the string and LNG is the Nambu-Goto Lagrangian 1 p 1 p LNG = − ′ − det hab = − ∂τ Xµ ∂σ X µ ∂σ Xρ ∂τ X ρ − ∂τ Xµ ∂τ X µ ∂σ X ρ ∂σ Xρ . It is then possible to define the so-called induced metric on the world sheet as hab = ∂a X µ ∂b Xµ . τ ) = X µ (σ ′ (σ. e. the minimum area of the world sheet minimizes the action. The relativistic string For strings. Here a finite string of length L will be considered. τ ) . − det hab dτ dσ is an infinitesimal element of the world sheet area. The simplest possible Poincare-invariant action. Analogous to the eigentime then an eigenlength σ can be introduced. As a consequence. In particular. ∂a.

As in section 5. To obtain the relation of the Polyakov action to the Nambu-Goto action it is again necessary to obtain its equation of motion. As in case of the point particle. µ Thus. This metric is taken to have a Lorentz signature for some chosen coordinate system ! + 0 γab = . the metric is still traceless and has negative determinant. Selected special topics 169 The Nambu-Goto action has two symmetries. this metric is traceless. it is useful to introduce a world-sheet metric γab (τ.2 it is rather cumbersome to use an action involving a square root. 0 − Thus. With it the new action. σ). depending on the index. and hence also the action as well as the Lagrangian and any other quantity constructed from it is.6.Chapter 6. as in the case of the point particle in section 5. σ) . To construct a simpler action. This is most conveniently obtained using the variational principle. ∂ω ′c ∂ω ′d ′ γ (τ ′ . This guarantees that for all invertible reparameterizations. This can be seen directly. except that now the Jacobian appears. One is invariance under reparametriza- tion. which are continuous deformations of the identity transformation. the induced metric is Poincare invariant. However. the functions Xµ transform as X ′µ = Λµν X ν + aµ ν δγ z }| { ∂a Λµν X ν ∂b Λµγ Xγ = Λ ν Λµγ ∂a X ν ∂b Xγ = ∂a X µ ∂b Xµ . the world-sheet metric γab has to have a non-trivial transformation property under reparameterizations. the general relation for determinants of metrics δγ = γγ ab δγab = −γγab δγ ab is quite useful.6. which leaves the world-sheet parameters τ and σ invariant. ∂ω a ∂ω b cd where the variables ω denote either σ and τ . and has a determinant smaller zero. The second invariance is Poincare invariance. the Brink-Di Vecchia-Howe-Deser-Zumino or Polyakov action. Z 1 1 2 γ ab h SP = − ′ dτ dσ (−γ) ab (6. For this. where γ denotes det γab . σ ′ ) = γab (τ.2.32) 4πα M is constructed. .

6. Then only Z   1 1 1 cd δSP = − dτ dσ (−γ) hab − γab γcd h 2 δγ ab 4πα′ 2 is left. 2 4πα′ 2 The second term is again the Polyakov Lagrangian. canceling the zero-order term. 4πα′ Expanding the term with indices cd up to first order in the variation leads to Z     1 1 1 cd ab ab  δSP = − dτ dσ LP − (−γ) 2 1 − γ δγcd γ + δγ hab 4πα′ 2 Z     1 1 ab ab 1 cd ab = − dτ dσ LP − (−γ) γ + δγ − γ γ δγcd hab .170 6.33) 2 Division of each side by its determinant finally yields  hab 1 γab γcd hcd 1 = (−h) 2 2 det − 1 γ γ hcd  12 2 ab cd  1 γab γcdhcd =  12 2 1 cd 2 γ h 2 cd det −γab γab = 1 (−γ) 2 In the second line it has been used that γcdhcd is a scalar. The condition that this expression should vanish yields the equation of motion for the world-sheet metric as 1 hab = γab γcd hcd (6. where the fact that the quantity γ ab γab is two. The relativistic string Abbreviating the Polyakov Lagrangian by LP and performing a variation with respect to γ yields Z  1 1 ab ab   δSP = − dτ dσ LP − (−γ − δγ) 2 γ + δγ hab 4πα′ Z  1 cd  12 ab ab   = − dτ dσ LP − −γ + γγ δγcd γ + δγ hab 4πα′ Z  1 1 cd  21 ab ab   = − dτ dσ LP − (−γ) 2 1 − γ δγ cd γ + δγ hab . has been used. permitting it to pull it out of the determinant. The result implies that h and γ are essentially proportional. Inserting this result in the Polyakov action yields Z Z 1 ab 1 1 1 SP = − dτ dσγ γ ab (−h) 2 = − dτ dσ (−h) 2 = S NG 4πα′ 2πα′ showing that it is indeed equivalent to the Nambu-Goto action. . due to the Lorentz signature of γ.

(6. σ) h′ab = hab ′ γab = e2ω(τ. this implies that γ ′ab = e−2ω γ ab . The diffeomorphism invariance follows directly from the transformation properties of the world-sheet metric. df = 0. (−γ ′ ) 2 (−γ ′ ) 2 (−γe4ω ) 2 (−γ) 2 Also the action is invariant. thus γ ab′ = Λγ ab = γ ab . σ). it should remain a constraint over time. since it is proportional to the Poincare-invariant induced metric. Thus. The expression of γ in terms of the induced metric h is invariant under this transformation. In particular. but considerably more lengthy since track of both variables has to be kept. 6. given by X ′µ (τ. The redundancy introduced with the additional degree of freedom γ grants a further symmetry. The Poincare invariance follows since γ is Poincare invariant. The origin of this symmetry comes from the unfixed proportionality of induced metric and the world-sheet metric. given a constraint f = 0. as noted before.σ) γab . Consider the not explicitly time-dependent situation. the expression appearing in the action transforms as 1 1 1 (−γ ′ ) 2 γ ′ab = −γe4ω 2 γ ab e−2ω = (−γ) 2 γ ab . As a consequence. Since γab γ ab has to be a constant. To see this note that γab is indeed a metric.7 First-order and second-order constraints It is not always possible to formulate constraints quite as concise as wished for. for arbitrary functions ω (τ. This is the so-called Weyl symmetry.Chapter 6. σ) = X µ (τ. in total analogy with the point-particle case. the Weyl invariance is indeed a symmetry. Selected special topics 171 The Polyakov action thus retains the Poincare and reparametrization invariance of the Nambu-Goto action. ′ γab γab e2ω γab e2ω γab 1 = 1 = 1 = 1 .34) dt .

The advantage is that with respect to the Dirac bracket all constraints are always exactly . this may entail another constraint from {g. By construction the Dirac brackets of all constraints of the first kind still vanish. but cannot be written as the Poisson bracket of the Hamilton function with some other quantity are called primary constraints. Given the total sets of constraints φi . as otherwise (6. χj }D = {χi . and therefore everything remains form-invariant. This implies there exists functions Cij with {χi . Further- more {χi . If f should be a constraint this requires g to be zero. All constraints which are not of the first kind are called of second kind. Constraints for which the Poisson bracket with the Hamilton function identically vanish. and all constraints act on equal footing. H}. χi }(C −1 )ij {χj . χj } − Cik (C −1 )kl Clj = Cij − Cij = 0. H} = g. All other constraints. Without proof. are called sec- ondary constraints. constraints of the first kind γi are the subsets of constraints which fulfill {γi .172 6. Of course. If the matrix C is invertible. χj } = Cij which form an antisymmetric matrix C. for all other constraints. and thus {f. Note that secondary constraints can only be defined once the Hamilton function is known. First-order and second-order constraints The time evolution is governed by (5. G}. G}D = {F. g = 0. Because of (6. thus {φi } = {γi } ∪ {χi }.7.35) the Dirac brackets of all constraints with the Hamilton function also vanish. provided these are not linearly dependent on the primary constraints. is can be shown that now all appearances of the Poisson brackets can be replaced by Dirac brackets.35) with g some function. G} − {F.43). such as g.34) is violated. It is possible to reclassify the set of primary and secondary constraints. Therefore now all brackets between constraints vanish. φi } = 0. All of them have to hold. as this can be taken to be a postulate. Therefore the time-evolution formulated in terms of the Dirac brackets still respect the constraints. define the Dirac brackets as {F. (6. χi .

In fact. and makes it something conceptually new. To see how brackets can play a role for quantum physics. it is useful to have a lit- tle preview to see how quantum physics could have something to do with the Hamilton function. It is the necessity of new postulates which sets quantum mechanics apart. and quantum physics. though introducing physically new concepts. as it is not easy to find an adequate. as anyhow only solutions to the equations of motions are searched for. all knowledge on mechanics is encoded in the presented formalism. However. Except for specifying the Hamilton or Lagrange functions for a given system. it is still a quite interesting question of how classical physics can be considered . For many systems the arising equations of motions are so involved that an exact solution is not possible. 6. it may be far from trivial. Furthermore. a whole new set of postulates will be necessary to do so. especially for complex systems with many degrees of freedom.8 A preview of quantum physics This lecture fully reviews classical mechanics. This by no means guarantees that all mechanical problems are solved. Leaving aside all the involved technical problems. and ultimately classical thermodynamics. these are from a mechanical point of view nothing conceptually new.Chapter 6. In classical physics. In fact. say. The problem of fluids is a famous and notorious example. constraints of the second kind are only fulfilled if evaluated on the solutions of the equations of motions. classical mechanical systems are therefore fully formulated. Nonetheless. Selected special topics 173 fulfilled. potential for a given observed system. It is therefore even possible that even the existence of a solution is a non-trivial question. In the context of quantum physics this difference becomes relevant: In quantum physics also trajectories which do not fulfill the equations of motions play a role. In terms of the Poisson brackets. A conceptually new situation arises only when introducing quantization. It will be the contents of the corresponding lectures to deal with these postulates in detail. in contrast to the case of ordinary differential equations the mathematical foundation of partial differential equations is much less developed. and there is no non-trivial time evolution which needs to be fulfilled. In the limit of very large numbers of degrees of freedom. this leads to classical statistical mechanics. and therefore primary and secondary constraints would make a difference. In this way all constraints are satisfied identically. Even approximate or numerical solutions can pose serious challenges. to write down a consistent Hamilton or Lagrange function. this is irrelevant. Therefore. in the context of constrained quantum physics a clean formulation uses the Dirac brackets rather than the Poisson brackets.

what is the rate of change of the action S = F2 when moving during some infinitesimally time to some point. this has an interesting implication. the system moves like waves over constant action surfaces. The function W only depends on the variables q as the P are constant. The condition that W (q. The notation ∂f . to the surfaces of constant action.8. Following section 5. e. this furthermore implies for ordinary coordinates ~ ~p = ∂W 10 ~ is a vector with components ∂f /∂xi . P. Since W is part of a canonical transformation function F2 .13. and it is therefore important to understand what this means. using the ordinary space coordinates for the description. The function F2 . t) = W (q. where E is the total energy. Start with a conservative system with the Hamilton function H = T + V = E.174 6. according to section 5. changes with time. the function W is related to the action. by construction. P ) − Et such that h = 0. ∂xi where the wave speed is defined as ~u = d~r/dt. the requirement that F2 should be constant defines how the system moves. This is given by the differential of10 S.3. If the system should stay at constant action this requires ~ = E. it is then possible to find a canonical transformation F2 (q. i. A preview of quantum physics as being part of something larger. Thus. P and Q in these new coordinates are constant. i. P ) should be constant therefore defines surfaces of constant action in the configuration space.13. This is quite abstract. Thus. It also gives some more insights into various concepts in classical mechanics. In a sense. a solution to the Hamilton-Jacobi equation. Since. which gives a preview of what comes. but in a linear fashion. This is best understood by first simplifying the problem to a single particle. ∂S dS = ∂t Sdt + dxi = −Edt + d~r∂~i W = −Edt + ~u∂W ~ dt. ~u∂W Since ∂~i W is perpendicular. by construction. Consider first the speed of the action waves. as a function of time crossing equal-action surfaces in configuration space. e. however. This will be done here. and forms action waves.

the particle moves also perpendicular to the surfaces of constant action.36) by a new equation. it is useful to rewrite the equations of mechanics a little bit more to make the duality more apparent. the motion of the particle is in the same direction as the action waves: Particles essentially ride action waves. |E| |E| u = =p p 2m(E − V ) 1   E = ~ )2 + V. It turns out that in the end a reformulation of (6. . for the free particle |E| |E| |E| |~u| = = = ~ | |∂W |~p| m|~v| Since the ratio |E|/m is fixed this implies that the relative direction of ~u and ~v are also fixed and in fact they are inverse proportional.3. Thus.36) using the Hamilton function is better suited for the postulates necessary. e. One major difference will be. quantized. of classical mechanics. which in the long-distance limit reduces to (6. it will come back to (6. though it is not obvious that it is true in general: Movement of particles can be equivalently described by either the movement as a trajectory in some suitable space or as action waves. at long distances. that there is not yet any quantum in the description: Both energy and momentum can take any arbitrary.13. Especially. (∂W 2m Since furthermore ∂F ~ 2=W ~ and since F2 = S the action.36). This has an interesting interpretation. For this. and often only discrete. compare the Hamilton-Jacobi equation with the magnitude of the wave speed. More important is that the postulates will require to also replace the concept of trajectories and measurements. which plays an important role in quantum physics.36). Selected special topics 175 Thus. the Schrödinger equation. While classical systems adhere to it. i. To further the transition. and therefore the mathematical foundation. according to section 5.36) u which can be considered as the wave-equation for classical mechanics. by replacing (6. this leads to  2 ~ 1 ∂S = 2 (∂t S)2 (6. however. but this will actually be a minor detail. continuous value. The major change induced by quantum physics will be that this is no longer possible. values will be possible. This already hints at a kind of duality between waves and particles. But still.Chapter 6. It is here where the postulates of quantum physics set in. this equation is no longer valid at quantum scales.