3 Lagrangian Dynamics

(Most of the material presented in this chapter is taken from Thornton and Marion, Chap. 7)

3.1 Introduction
Although Newtons equation F = p correctly describes the motion of a particle (or a system of particles), it is often the case that a problem will be too complicated to solve using this formalism. For example, a particle may be restricted in its motion such that it follows the contours of a given surface or is otherwise constrained, so that the forces of constraint are not easily expressible. Consider how complicated the first example of Chapter 1, with a mass constrained to move on a surface, was. It may not even be possible at times to find expressions for some forces of constraint. Such occurrences would render it impossible to treat the problem with the Newtonian formalism since this requires the knowledge of all forces acting on the particles. In this section we will study a different approach. The formalism that will be introduced is based on the so-called Hamiltons Principle, from which the equations of motion will be derived. These equations are called Lagranges equations. Although the method based on Hamiltons Principle does not constitute in itself a new physical theory, it is probably justified to say that it is more fundamental that Newtons equations. This is because Hamiltons Principle can be applied to a much wider range of physical phenomena than Newtons theory (e.g., quantum mechanics, quantum field theory, electromagnetism, relativity). However, as will be shown in the following sections, the Lagranges equation derived from this new formalism are equivalent to Newtons equations when restricted to problems of mechanics. Lagranges formalism doesnt use forces: there is no need for them! How can this be? The need for knowledge of the forces can often be avoided by using the powerful concept of energy, particularly the potential energy. Consider the force of gravity which (because it is conservative) can be represented by the gradient of a potential. The gravitational potential contains all the information on the forces but is simpler (it is just a scalar, not a vector). So for example, if I told you that a particle fell a height h in Earths gravity, you could calculate the speed at the end of its fall just using conservation of energy, you wouldnt actually have to calculate the force on the particle at all. The brilliance of Lagranges approach lies in its simplification of the equations we have to solve (though again, the end result has to be the same as Newtons). Note that Lagranges approach does not work well for forces that are not conservative (like friction and air resistance) so Newtons approach is still often useful. We should note that Hamiltons Principle was derived after Lagrange came up with his new formalism. Hamilton derived his Principle almost 50 years after Lagranges work and later pushed on to develop Hamiltons equations. These constitute a variation on Lagranges equations that often prove useful (eg. quantum mechanics) though we will concentrate mostly on Lagranges approach here.

56

3.2 Important Details on Notation

In this chapter, unless otherwise stated, the following notation conventions will be used: 1. Einsteins summation convention. Whenever an index appears twice (and only twice), then a summation over this index is implied. For example,

xi xi

xi xi =

xi 2 .

(3.1)

2. The index i is reserved for Cartesian coordinates. For example, xi , for i = 1, 2, 3 , represents either x, y , or z depending on the value of i . Similarly, pi can represent px , p y , or pz , momenta in the x,y,z directions. This does not mean that any other indices cannot be used for Cartesian coordinates, but that the index i will only be used for Cartesian coordinates. 3. When dealing with systems containing multiple particles, the index will be used to identify quantities associated with a given particle when using Cartesian coordinates. For example, if we are in the presence of n particles, the position vector for particle is given by r , and its kinetic energy T by

T =

1 m x ,i x ,i , 2

= 1, 2, ... , n and i = 1, 2,3.

(3.2)

Take note that, according to convention 1 above, there is an implied summation on the Cartesian velocity components (the index i is used), but not on the masses since the index appears more than twice. Correspondingly, the total kinetic energies is written as

T=

1 n 1 n m x ,i x ,i = m ( x 2 + y 2 + z 2 ). 2 =1 2 =1

(3.3)

3.3 Hamiltons Principle

Hamiltons Principle is concerned with the minimization of a quantity (i.e., the action). You have already encountered minimization problems in calculus, where you might want to find the value of a function at its minimum. The minimization problem posed by Hamiltons Principle is more complicated as you have to minimize the integral of a function, but it can be solved using the calculus of variations (described in Ch 6 of Thornton and Marion) if youre really interested. Hamiltons Principle can be stated as follows:

57

The motion of a system from time t1 to t2 is such that the line integral (called the action or the action integral),

where L = T U (with T , and U the kinetic and potential energies, respectively), has a stationary value for the actual path of the motion.

Note that a stationary value for equation (3.4) just means that it is an extremum (either a minimum or a maximum) for the action, not necessarily a minimum. But in almost all important applications in dynamics a minimum occurs. A more plain-English description is Of all the possible paths along which a dynamical system may move from one point to another within a specified time interval (consistent with any constraints), the actual path followed is that which minimizes the time integral of the difference between the kinetic and potential energies. (Thornton and Marion p 230) Ok, so why should we expect Hamiltons Principle to apply? Certainly it doesnt seem obvious at first glance. Hamiltons Principle can be derived rigorously from Newtons equations (a little like we derived conservation of momentum and energy earlier) but it is a more involved process which we wont go through here. For the purposes of this course well just assume that its true (see section at the end of this chapter for more if youre curious). We should note however that such mysterious minimization principles are one of the beauties of physics and crop up in different contexts. An example is the propagation of a light photon through a set of transparent objects (eg lenses). The speed of light in transparent matter is generally slower than in a vacuum and is related to its index of refraction. If you measure the amount of time it takes a particular photon to fly from point A to point B through a setup of lenses and such (lets call this time t0), you will find that (for its initial position and velocity) it took the path with the shortest possible flight time. That is, if at any point along its path it had moved a little to one side or the other, it would have taken longer than t0 to get to point B. This is true regardless of the setup of transparent objects in its path and seems to have nothing to do with it. But lets get back to Hamiltons Principle. Because of the dependency of the kinetic and potential energies on the coordinates xi , the velocities xi , and possibly the time t , it is found that
L = L ( xi , xi , t ) .

I=

t2 t1

L dt ,

(3.4)

(3.5)

58

(that is, we turn out not to need to worry about higher order derivatives), Hamiltons Principle can now be expressed mathematically by

t1

in the notation of the calculus of variations. Equation (3.6) can readily be solved (by techniques we wont get into here). The solution is
L d L = 0, xi dt xi


Equations (3.7) are called the Lagrange equations of motion, and the quantity L ( xi , xi , t ) is the Lagrangian. Lagranges equations essentially provide us with the equations of motion for a system based on its kinetic and potential energies. We still have to solve these equations (much as we would if we started from F=ma instead) but Lagranges equations often turn out to be easier to solve. And we dont have to compute forces! For example, if we apply Lagranges equation to the problem of the one-dimensional harmonic oscillator (without damping), we have
 

L = T U =

1 2 1 2 mx kx , 2 2

and L = kx x d L d = ( mx ) = mx. dt x dt
After substitution of equations (3.9) into equation (3.7) we find mx + kx = 0 k x+ x =0 m
        

for the equation of motion. This result (simple harmonic oscillator) is identical to what was obtained using Newtonian mechanics. This is, however, a simple problem that can easily (and probably more quickly) be solved directly from the Newtonian formalism.

59

t2

L ( xi , xi , t ) dt = 0 .

(3.6)

i = 1, 2, ... , n .

(3.7)

(3.8)

(3.9)

(3.10)

The benefits of using the Lagrangian approach become more obvious if we consider more complicated problems. For example, we try to determine the equations of motion of a particle of mass m constrained to move on the surface of a sphere in much the same way we did in the first example of the first chapter. I had actually hoped to do exactly the same problem here and show how easy it is in the Lagrange formalism, but there was one complication I forgot about that means its still a bit tricky to use the LF! Instead Ill have to confine myself to a slightly restricted version of the earlier problem, but if you go back and reanalyze the earlier one, I think youll agree that the LF involves much less algebra. So lets consider a mass constrained to move on the surface of a sphere (by a massless rod say) and subject to a conservative force F = F e , with F a constant (rather than a more general force F ( , ) as we used earlier. In this case we have

1 1 mv 2 + mv 2 2 2 (3.11) 1 1 2 2 2 2 2 = mR + mR sin ( ) 2 2 In this case, we dont have the potential energy, but we can deduce it fairly simply from the fact that F = U U 1 U 1 U U = e r + e + e (3.12) r r r sin Since only F is non-zero, 1 U U = e r U = F r + const = F R (3.13) T=

where we have chosen the constant in the potential energy such that U = 0 when = 0 . The Lagrangian is given by
" "

L = T U =

1 1 mR 2 2 + mR 2 sin 2 ( ) 2 + F R . 2 2

Upon inspection of the Lagrangian, we can see that there are only two of the three spherical coordinates that appear in the Lagrangian. This is because the constraint that r=R means there is no motion in the radial direction. This is built into the Lagrangian without the need to specify the radial force that holds the particle in place. The absence of r from the Lagrangian means that, even though we have three spherical coordinates there are only 2 degrees of freedom (see next section) for this problem, i.e., , and . We now need to calculate the different derivatives that compose the Lagrange equations

60

(3.14)

L = mR 2 2 sin ( ) cos ( ) + F R L =0
))

d L d mR 2 = mR 2 = dt dt
)

d L d mR 2 sin 2 ( ) = mR 2 2 sin ( ) cos ( ) + sin 2 ( ) , = dt dt

applying equation (3.7) for , and we find the equations of motion to be

F = mR 2 sin ( ) cos ( )
00 0 00

0 = mR 2 sin ( ) sin ( ) + 2 cos ( ) .

0 0

This problem was analyzed at the end of Chapter 1 on Newtonian Mechanics (problem 22, with F = 0 ), where Newtons equation was used to solve the problem. One can see how simpler the present treatment is. There was no need to calculate relatively complex equations like e r . One last note here: the derivation above pulls one trick, though a slightly subtle one. Equation (3.7) is expressed in Cartesian coordinates, but our derivation used spherical coordinates. In fact we will show below that this is OK, in fact you can construct your Lagrangian using any coordinates you want. In fact you might note that all three spherical coordinates r, and are treated the same when using the Lagrangian formalism, even though they all have different units (this is one of the strengths of this approach as well see).
11

3.4 Degrees of Freedom and Generalized Coordinates

If a system is made up of n particles, we can specify the positions of all particles with 3n coordinates. On the other hand, if there are m equations of constraints (for example, if some particles were connected to form rigid bodies), then the 3n coordinates are not all independent. There will be only 3n m independent coordinates, and the system is said to possess 3n m degrees of freedom. For example, for the particle confined to a sphere above, even though it can move in three dimensions, they are not all independent. In spherical coordinates, this means that r=R where R is a constant. Even if we had expressed the problem in Cartesian coordinates, the positions would be related through x 2 + y 2 + z 2 = R 2 . So the particle is restricted to a 2dimensional surface in 3-dimensional space. Thus constraints reduce the number of degrees of freedom (in some sense the dimension) of the problem.

61

))

(3.15)

&

&

( )

# % ' # % '

(3.16)

Furthermore, the coordinates and degrees of freedom do not have to be all given in the same coordinate system. In fact, we can choose to have different types of coordinate systems for different coordinates. Also, the degrees of freedom do not even need to share the same unit dimensions. For example, a problem with a mixture of Cartesian and spherical coordinates will have lengths and angles as units. Because of the latitude available in selecting the different degrees in freedom, the name of generalized coordinates is given to any set of quantities that completely specifies the state of the system. The generalized coordinates are usually written as q j . It is important to realize that there is not one unique way of setting up the generalized coordinates, there are indeed many different ways of doing this. Unfortunately, there are no clear rules for selecting the best set of generalized coordinates. The ultimate test is whether or not the choice made leads to a simple solution for the problem at hand. However, there are some systems which are better for particular problems than others. In the case of the particle confined to a sphere, we could have expressed the Lagrangian in Cartesian x,y,z coordinates, but would have gotten a much more complicated expression. The correct choice of a coordinate system is often a key to solving a problem, and is where a physicists intuition (developed by practice) is most valuable. In addition to the generalized coordinates q j , a corresponding set of generalized

velocities q j is defined. In general, the relationships linking the Cartesian and generalized coordinates and velocities can be expressed as
x ,i = x ,i ( q j , t )
3 3 3

x ,i = x ,i ( q j , q j , t ) ,

In other words, the Cartesian positions are only an explicit function of the generalized coordinates for that particle and (possibly) the time, while the Cartesian velocity of a particle is only an explicit function of its own generalized coordinates and generalized velocities (and possibly the time). Basically, a particles Cartesian position and velocity is only a function of its own generalized coordinates and velocities, not those of other particles. Alternatively, we can flip this around and say that the generalized coordinates of a particular particle are only a function of its Cartesian position (and possibly the time).
q j = q j ( x ,i , t )
4 4 4

We must also include the equations of constraints f k ( x ,i , t ) = f k ( q j , t ) = 0. (3.19)

= 1, ... , n,

i = 1, 2,3,

j = 1, ... , 3n m,

(3.17)

q j = q j ( x ,i , x ,i , t ) .

(3.18)

62

This allows us to make the technical point that Hamiltons Principle can now be expressed in term of the generalized coordinates and velocities (rather than the Cartesian coordinates used in (3.6))

t1

with Lagranges equations given by

L d L = 0, q j dt q j
7

Examples
1) The simple pendulum. Lets solve the problem of the simple pendulum (of mass m and length ) by first using the Cartesian coordinates to express the Lagrangian, and then transform into a system of cylindrical coordinates.
8

Solution. In Cartesian coordinates the kinetic and potential energies, and the Lagrangian are

1 2 1 2 mx + my 2 2 U = mgy 1 1 L = T U = mx 2 + my 2 mgy. 2 2
9 9 9 9

T=

63

Figure 3-1 A simple pendulum of mass m and length

t2

L ( q j , q j , t ) dt = 0 ,

(3.20)

j = 1, 2, ... ,3n m .

(3.21)

(3.22)

We could now proceed with Lagranges equations except we still have a constraint to consider, that is that x 2 + y 2 = l 2 and which we have not yet included in the calculation. We note that we start with two coordinates and have one constraint so we have one degree of freedom. This means that the two coordinates are not independent (they are linked by x 2 + y 2 = l 2 and we should eliminate either x or y from the Lagrangian (doesnt matter which, well recover the solution for both). To do this we would use x = l 2 y2 x=
@

yy

l 2 y2

where the second expression is just d/dt of the first. However, it is easier to express the constraint in polar coordinates and well want to change to that coordinate system anyway to solve the problem. This is a point which it is difficult to overemphasize. Choosing the right coordinate system can make all the difference between a set of equations which (while correct) are difficult to solve and a set of equations which are easy to solve. Imagine trying to solve the motion of a freely moving particle in polar coordinates with an oddly chosen origin. Though you could get the equations of motion, they would be quite complex because the motion (a straight line) is not easily expressed in the chosen coordinate system. Choosing the coordinate system is the single most important step of the problem! You might ask, what is the best coordinate system? Hamiltonian mechanics essentially asks (and answers) that question but determining the best coordinate system is as hard as solving the problem: in essence you do have to solve the problem first. However for our purposes here, the obvious choice is almost always the correct one. Choosing which coordinate system to use at each step is a skill acquired with practice. Even in this simple example here we chose to express the Lagrangian first in Cartesians (often a good choice because T and U are easily expressed in those coordinates) and then we transform to polar coordinates before determining the equations of motion. The initial use of Cartesians is probably not needed for this problem but is often a good starting point. Ok back to the problem. We now transform the coordinates with the following relations. Note that we plug the constraint in directly (that is, that r = l)

x = sin ( )
A A

y = cos ( ) .

Taking the time derivatives, we find

64

(3.23)

x = cos ( ) y = sin ( )
C B C B C

L=

1 m 2 2 cos 2 ( ) + 2 2 sin 2 ( ) + mg cos ( ) 2 1 = m 2 2 + mg cos ( ) . 2

C B C

We can now see that there is only one generalized coordinates for this problem, i.e., the angle . We can use equation (3.21) to find the equation of motion L = mg sin ( ) d L d = m 2 = m 2 , dt dt and finally
QQ P Q P P G E I Q D

This is the equation of motion for a pendulum, though weve looked at it before in the limit of small oscillations (that is sin ) which yields the harmonic oscillator equation just as we saw in Chapter 2. g + = 0. (3.27)
U TT

So Lagranges equations give us identical results to using Newtons equations in this case. At this point, Lagranges equations really dont seem to have done much. We get the same equation of motion as we did using Newtons method. Its value will become clearer in more complicated situations. However we should note that even with Lagranges method, it doesnt mean that the solutions will always be easy to find, even with the right coordinate choice! 2) The pendulum on a rotating rim. A simple pendulum of length b and mass m moves on a mass-less rim of radius a rotating with constant angular velocity (see Figure 3-2). Get the equation of motion for the mass.
Solution. If we choose the center of the rim as the origin of the coordinate system, we calculate

RR

B B

(3.24)

(3.25)

+ sin ( ) = 0.

(3.26)

65

x = a cos ( t ) + b sin ( )

y = a sin (t ) b cos ( ) ,

(3.28)

and

x = a sin (t ) + b cos ( ) y = a cos ( t ) + b sin ( ) .

V V V V

(3.29)

Figure 3-2 A simple pendulum attached to a rotating rim.

The kinetic and potential energies, and the Lagrangian are

a a a

T=

L = T U 1 = m a 2 2 + b 2 2 + 2ab sin ( t ) mg ( a sin ( t ) b cos ( ) ) . 2

We now calculate the derivatives for the Lagrange equation using as the sole generalized coordinate
a

) )

66

` X

1 m a 2 2 + b 2 2 + 2ab sin ( ) cos (t ) sin (t ) cos ( ) 2 1 = m a 2 2 + b 2 2 + 2ab sin ( t ) 2 U = mg ( a sin ( t ) b cos ( ) )

Y W a

)
(3.30)

L = mab cos ( t ) mgb sin ( ) d L = mb 2 + mab cos ( t ) . dt Finally, the equation of motion is
b bb b

(3.31)

a g sin ( t ) = 0. (3.32) b b The question only asks for the equation of motion, so we are done. However, this equation is perhaps not easily solved, but try solving it with Newtons equations. This is much harder because the mass is accelerated by the tension in the rod which is rather harder to calculate than it is to express the constraint it provides (simply that its length is fixed at b).

2 cos ( t ) +
cc

3) The sliding bead. A bead slides along a smooth wire that has the shape of a parabola z = cr 2 (see Figure 3-3). At equilibrium, the bead rotates in a circle of radius R when the wire is rotating about its vertical symmetry axis with angular velocity . Find the value of c .

Figure 3-3 A slides along a smooth wire that rotates about the z -axis .

Solution. We choose the cylindrical coordinates r, , and z as generalized coordinates. The kinetic and potential energies are

67

1 m r 2 + r 2 2 + z 2 2 U = mgz.
d d d

T=

(3.33)

We have in this case some equations of constraints that we must take into account, namely
z = cr 2 z = 2crr ,

(3.34)

and

= t = .
f

Inserting equations (3.34) and (3.35) in equation (3.33), we can calculate the Lagrangian for the problem
L = T U 1 = m ( r 2 + 4c 2 r 2 r 2 + r 2 2 ) mgcr 2 . 2
g g

It is important to note that the inclusion of the equations of constraints in the Lagrangian has reduced the number of degrees of freedom to only one, i.e., r . We now calculate the equation of motion using Lagranges equation L = m ( 4c 2 rr 2 + r 2 2 gcr ) r d L = m ( r + 4c 2 r 2 r + 8c 2 rr 2 ) , dt r and
h hh h hh h

r (1 + 4c 2 r 2 ) + r 2 ( 4c 2 r ) + r ( 2 gc 2 ) = 0.
i ii

This equation of motion doesnt look particulary easy to solve. However, we also have the fact that the bead is in equilibrium. When the bead is in equilibrium, we have r = R and r = r = 0 , and equation (3.38) reduces to
pp p

R ( 2 gc 2 ) = 0,
or
68

(3.35)

(3.36)

(3.37)

(3.38)

(3.39)

c=

2
2g

(3.40)

At no point do we have to calculate the force that the wire exerts on the bead, its all included in the constraints.

3.5 Constraints
A system that is subjected to constraints that can be expressed in the form f ( x ,i , t ) = f ( q j , t ) = 0 , (basically, that are simply constraints on the position, called
holonomic constraints) will always allow the selection of a proper set of generalized coordinates for which the equations of motion will be free of the constraints themselves.

What other constraints are possible? The most common kind of constraint after a position constraint is a velocity constraint. Velocity and other constraints are more difficult to handle. However, constraints that are functions of the velocities, and which can be written in a differential form and integrated to yield relations amongst the coordinates are also holonomic. For example, an equation of the form
Ai dxi +B=0 dt

(3.41)

(note: we are using the Einstein summation convention!) cannot, in general, be integrated to give an equation of the form f ( xi , t ) = 0 . Such equations of constraints are nonholonomic. We will not consider this type of constraints any further. However, if the constants Ai , and B are such that equation (3.41) can be expressed as f xi f + = 0, xi t t or more simply
df = 0, dt

(3.42)

(3.43)

then it can be integrated to give

f ( xi , t ) C = 0

(3.44)

where C is a constant. Using generalized coordinates, and by slightly changing the form of equation (3.43), we conclude that, as was stated above, constraints that can be written in the form of a differential

69

df =

f f dq j + dt = 0 q j t

(3.45)

are similar to the constraints considered at the beginning of this section, that is f ( q j , t ) C = 0. (3.46)

Basically, some velocity constraints are holonomic and can be handled, if they can be expressed in the form of (3.45). These are the only kind we will consider here.

3.6 Lagranges Equations with Undetermined Multipliers

In some cases, we may be interested in knowing the forces of constraint. For example, in designing a roller coaster, the car is constrained to the track, but only as long as the forces on the track dont exceed their strength. We can solve for the forces of constraints in problems involving constraints of the holonomic kind in exactly the same manner as was done earlier to get Lagranges equations. This is done using Hamiltons Principle and by introducing the so-called Lagrange undetermined multipliers, but we wont get into the details here. When this is done, we find that the following form for the Lagrange equations
L d L f + k ( t ) k = 0 q j dt q j q j where the index j = 1, 2, ... , 3n m, and k = 1, 2, ... , m convention for the last term on the left).
q

(3.47)

(note Einstein summation

Although the Lagrangian formalism does not require the insertion of the forces of constraints involved in a given problem, these forces are closely related to the Lagrange undetermined multipliers. The corresponding generalized forces of constraints can be expressed as (summation convention again!) Q j = k ( t ) f k . q j (3.48)

Examples
1) The rolling disk on an inclined plane. We now solve the problem of a disk of mass m and of radius R rolling down an inclined plane (see Figure 3-4).
Solution. Separating the kinetic energy into a translational rotational part, we can write

70

where I = mR 2 2 is the moment of inertia of the disk about its axis of rotation (which we look up from a table). The potential energy and the Lagrangian are given by

U = mg ( l y ) sin ( ) L = T U 1 1 = my 2 + mR 2 2 mg ( l y ) sin ( ) , 2 4
s s

where l , and are the length and the angle of the inclined plane, respectively. The equation of constraint given by f = y R = 0. (3.51)

This problem presents itself with two generalized coordinates ( y and ) and one equation of constraints, which leaves us with one degree of freedom. We would normally now use the constraint to eliminate one of the coordinates (either or y) but we dont when we want to solve for the forces of constraint. Rather we now apply the Lagrange

equations as defined with equation (3.47)

Figure 3-4 - A disk rolling on an incline plane without slipping.

L d L f + = mg sin ( ) my + = 0 y dt y y 1 L d L f + = mR 2 R = 0. 2 dt
From the last equation we have
tt tt t t

71

r r

T=

1 2 1 2 my + I 2 2 1 1 = my 2 + mR 2 2 , 2 4

(3.49)

(3.50)

(3.52)

 = mR ,
which using the equation of constraint (3.51) may be written as
uu

1 2

(3.53)

= my.
Inserting this last expression in the first of equation (3.52) we find
y= 2 g sin ( ) , 3
ww vv

1 2

(3.54)

(3.55)

and

= mg sin ( ) .
In a similar fashion we also find that

1 3

(3.56)

=
xx

2 g sin ( ) . 3R

(3.57)

Equations (3.55) and (3.57) can easily be integrated, and the forces of constraints that keep the disk from sliding can be evaluated from equation (3.48)
f 1 = = mg sin ( ) y 3 f 1 = R = mRg sin ( ) . Q = 3 Qy =

(3.58)

Take note that Qy and Q are a force and a torque, respectively. This justifies the appellation of generalized forces (they dont always have units of force).

3.7 Equivalence of Lagranges and Newtons Equations

In this section we will prove the equivalence of the Lagrangian and the Newtonian formalisms of mechanics. We consider the simple case where the generalized coordinates are the Cartesian coordinates, and we concentrate on the dynamics of a single particle not subjected to forces of constraints. The Lagrange equation for this problem is

72

L d L = 0, xi dt xi
y

i = 1, 2, 3.

(3.59)

Since L = T U and T = T ( xi ) , and U = U ( xi ) for a conservative system (e.g., for a particle falling vertically in a gravitational field we have T = my 2 2, and U = mgy ), Lagranges equation becomes since T U = = 0. xi xi For a conservative system we also have Fi = and
d T d 1 d = mxk xk = ( mxi ) = pi , dt xi dt xi 2 dt where pi is component i of the momentum. From equations (3.62) and (3.63) we finally obtain Fi = pi , which are, of course, the Newtonian equations of motion.

U d T = , xi dt xi

(3.60)

(3.61)

U , xi

(3.62)

(3.63)

(3.64)

3.8 Conservation Theorems

Before deriving the usual conservation theorem using the Lagrangian formalism, we must first consider how we can express the kinetic energy as a function of the generalized coordinates and velocities.

The Kinetic Energy

In a Cartesian coordinates system the kinetic energy of a system of particles is expressed as

73

1 n T= m x ,i x ,i , 2 =1

(3.65)

where a summation over i is implied. In order to derive the corresponding relation using generalized coordinates and velocities, we go back to the first of equations (3.17), which relates the two systems of coordinates

x ,i = x ,i ( q j , t ) ,

j = 1, 2, ... ,3n m.

(3.66)

Taking the time derivative of this equation we have

x ,i =

x ,i x q j + ,i , q j t

and squaring it (and summing over i )

x ,i x ,i =

x ,i x ,i x x x x q j qk + 2 ,i ,i q j + ,i ,i . q j qk q j t t t

An important case occurs when a system is scleronomic, i.e., there is no explicit dependency on time in the coordinate transformation, we then have
x ,i t = 0,

and the kinetic energy can be written in the form

T = a jk q j qk

with

a jk =

x x 1 n m ,i ,i , 2 =1 q j qk

where a summation on i is still implied. Just as was the case for Cartesian coordinates, we see that the kinetic energy is a quadratic function of the (generalized) velocities. If we next differentiate equation (3.69) with respect to ql , and then multiply it by ql (and summing), we get
ql T = 2a jk q j qk = 2T ql

74

(3.67)

(3.68)

(3.69)

(3.70)

(3.71)

since a jk is not a function of the generalized velocities, and it is symmetric in the exchange of the j and k indices.

Conservation of Energy
Consider a general Lagrangian, which will be a function of the generalized coordinates and velocities and may also depend explicitly on time (this dependence may arise from time variation of external potentials, or from time-dependent constraints). Then the total time derivative of L is
dL L L L = qj + qj + . dt q j q j t But from Lagranges equations, L d L = , q j dt q j and equation (3.72) can be written as
dL d L L L = qj + qj + dt dt q j q j t d L L = qj + . dt q j t It therefore follows that d L L dH L = + =0 qj L + t dt q j dt t
or
dH L = , dt t
p p r s s o o q n m i i k l l h h j ll l g l i i k l h h j f g f e dd d d

(3.72)

(3.73)

(3.74)

(3.75)

(3.76)

where we have introduced a new function

In cases where the Lagrangian is not explicitly dependent on time we find that

75

H = qj
t

L L. q j

(3.77)

If we are in presence of a scleronomic system, where there is also no explicit time x ,i = x ,i ( q j ) ), then dependence in the coordinate transformation (i.e.,

U = U ( q j ) and U q j = 0 and
v

L (T U ) T = = . q j q j q j

Equation (3.78) can be written as

H = qj
x

T L q j
x

= 2T L = T + U = E = cste, where we have used the result obtained in equation (3.71) for the second line.

The function H is called the Hamiltonian of the system and it is equal to the total energy only if the following conditions are met: 1. The equations of the transformation connecting the Cartesian and generalized coordinates must be independent of time (the kinetic energy is then a quadratic function of the generalized velocities). 2. The potential energy must be velocity independent. It is important to realize that these conditions may not always be realized. For example, if the total energy is conserved in a system, but that the transformation from Cartesian to generalized coordinates involve time (i.e., a moving generalized coordinate system), then equations (3.69) and (3.71) dont apply and the Hamiltonian expressed in the moving system does not equal the energy. We are in a presence of a case where the total energy is conserved, but the Hamiltonian is not.

3.9 DAlemberts Principle and Lagranges Equations

(This section is optional and will not be subject to examination. It will not be found in Thornton and Marion. The treatment presented below closely follows that of Goldstein, Poole and Safko, pp. 16-21.)

It is important to realize that Lagranges equations were not originally derived using Hamiltons Principle. We used Hamiltons Principle above (well, ok we didnt really use it much) because it is more straightforward. However, Lagrange himself placed the subject on a sound mathematical foundation by using the concept of virtual work along

76

H = qj
u

L L = constant. q j

(3.78)

(3.79)

(3.80)

with DAlemberts Principle which is closely tied to Newtons equations. This section describes this approach in some detail.

Virtual Work and DAlemberts Principle

A virtual displacement is the result of any infinitesimal change of the coordinates r that define a particular system, and which is consistent with the different forces and constraints imposed on the system at a given instant t . The term virtual is used to distinguish these types of displacement with actual displacement occurring in a time interval dt , during which the forces could be changing. Its a mathematical concept rather than an actualized physical one. Now, suppose that a system is in equilibrium. In that case the total force F on each particle that compose the system must vanish, i.e., F = 0 . If we define the virtual work done on a particle as F r (note that we are using Cartesian coordinates) then we have
{ z y

F r = 0.

(3.81)

( Lets now decompose the force F as the sum of the applied force Fa ) and the force of constraint f such that ( F = F ) + f ,
a

(3.82)

then equation (3.81) becomes

} }

In what follows, we will restrict ourselves to systems where the net virtual work of the forces of constraints is zero, that is f r = 0,
 ~

and
( F ) r = 0.

This condition will hold for many types of constraints. For example, if a particle is forced to move on a surface, the force of constraint if perpendicular to the surface while the virtual displacement is tangential. It is, however, not the case for sliding friction forces since they are directed against the direction of motion; we must exclude them from our analysis. But for systems where the force of constraints are consistent with equation (3.84), then equation (3.85) is valid and is referred to as the principle of virtual work.

77

( F ) r +

f r = 0.

(3.83)

(3.84)

(3.85)

Now, lets consider the equation of motion F = p , which can be written as

F p = 0.

(3.86)

Inserting this last equation in equation (3.81) we get

( F p ) r

= 0,

and upon using equations (3.82) and (3.84) we find

( F( ) p ) r

Equation (3.88) is often called DAlemberts Principle.

Lagranges Equations
We now go back to our usual coordinate transformation that relates the Cartesian and generalized coordinates

x ,i = x ,i ( q j , t ) , from which we get dx ,i x ,i x = q j + ,i . dt q j t

Similarly, the components x ,i of the virtual displacements vectors can be written as

x ,i =

x ,i qj. q j

Note that no time variation t is involved in equation (3.91) since, by definition, a virtual displacement is defined as happening at a given instant t , and not within a time interval t . Inserting equation (3.91) in the first term of equation (3.88), we have
a

F( ,i ) r ,i =

= Q j q j ,

where summations on i, and j are implied, and the quantity

78

(3.87)

= 0.

(3.88)

(3.89)

(3.90)

(3.91)

F( ,i )
a

x ,i q j

qj

(3.92)

are the components of the generalized forces. Concentrating now on the second term of equation (3.88) we write
x ,i qj. q j

m x ,i

This last equation can be rewritten as m x ,i

We can modify the last term since

x d x ,i = ,i . dt q j q j Furthermore, we can verify from equation (3.90) that

x ,i qk

x ,i qk

Substituting equations (3.96) and (3.97) into (3.95) leads to

Combining this last result with equations (3.93) and (3.94), we can express (3.88) as

where we have introduced T the kinetic energy of the system, such that

79

,i

,i

,i

( F ( ) p ) x

= Qj

d T T dt q j q j

q j = 0,

d 1 m x ,i x ,i dt q j 2

1 m x ,i x ,i q j 2

q j

m x ,i

qj =

x ,i

x x d m x ,i ,i m x ,i ,i q j q j q j dt

q j

x ,i

qj =

x d d x ,i m x ,i ,i m x ,i dt q j dt q j

p ,i x ,i =

Qj =

F( ,ai )

x ,i q j

(3.93)

(3.94)

qj.

(3.95)

(3.96)

(3.97)

(3.98)

q j.

(3.99)

Since the set of virtual displacements q j are independent, the only way for equation (3.99) to hold is that
d T T = Qj. dt q j q j for each j. If we now limit ourselves to conservative systems, we must have
F( ,ai ) = and similarly,
Q j = F( ,i )
a

U , x ,i

x ,i q j

U x ,i x ,i q j

U = , q j

with U = U ( x ,i , t ) = U ( q j , t ) the potential energy. We can, therefore, rewrite equation (3.101) as (T U ) d (T U ) = 0, q j q j dt since U q j = 0 . If we now define the Lagrangian function for the system as

L = T U ,
we finally recover Lagranges equations

L d L =0. dt q j q j

80

T=

1 2

m x ,i x ,i .

(3.100)

(3.101)

(3.102)

(3.103)

(3.104)

(3.105)

(3.106)

Dissipative Forces and Rayleighs Dissipative Function

So far, we have only dealt with system where there is no dissipation of energy. Lagranges equations can, however, be made to accommodate some of these situations. To see how this can be done, we will work our way backward from Lagranges equation d L L = 0. dt q j q j

(3.107)

If we allow for the generalized forces on the system Q j to be expressible in the following manner

then equation (3.107) can be written as d T T = Qj. dt q j q j

We now allow for some frictional forces, which cannot be derived from a potential such as expressed in equation (3.108), but for example, are expressed as follows
f j = k j q j ,

where no summation on the repeated index is implied. Expanding our definition of generalized forces to include the friction forces
Q j Qj + f j ,

Equation (3.109) becomes

d T T U d U = + + fj, dt q j q j q j dt q j or, alternatively d L L = f j. dt q j q j

Dissipative forces of the type shown in equation (3.110) can be derived in term of a function R , known as Rayleighs dissipation function, and defined as

81

Qj =

U d U + , q j dt q j

(3.108)

(3.109)

(3.110)

(3.111)

(3.112)

(3.113)

From this definition it is clear that

R , q j and the Lagrange equations with dissipation becomes fj = L R d L + = 0, dt q j q j q,

so that two scalar functions, L and R , must be specified to obtain the equations of motion.

Velocity-dependent Potentials
Although we exclusively studied potentials that have no dependency on the velocities, the Lagrangian formalism is well suited to handle some systems where such potentials arise. This is the case, for example, when the generalized forces can be expressed with equation (3.108). That is, Qj =

U d U . + q j dt q j
  

This equation applies to a very important type of force field, namely, the electromagnetic forces on moving charges. Consider an electric charge, q , of mass m moving at velocity v in a region subjected to an electric field E and a magnetic field B , which may both depend on time and position. As is known from electromagnetism theory, the charge will experience the so-called Lorentz force

F = q E + ( v B) .
   

Both the electric and the magnetic fields are derivable from a scalar potential and a vector potential A by
E = A , t

and

82

R=

1 2

k j q j2.

(3.114)

(3.115)

(3.116)

(3.117)

(3.118)

(3.119)

B = A.

(3.120)

The Lorentz force on the charge can be obtained if the velocity-dependent potential energy U is expressed

U = q qA v, so that the Lagrangian is

L = T U
 

(3.121)

1 2 mv q + qA v. 2

(3.122)

83