Chapter 3

Lagrangian Mechanics

3.1 The Lagrange generalized coordinates

We saw how the constraints are taken care of in the equation of motion.
Solving the system of differential equations, Lagrange multiplier and a con-
straint is usually a very complicated problem. In many situations we are
not interested in the forces of constraints and would like to eliminate them.
Lagrange (Mécanique analytique 17881) has worked out an elegant method
where by a special transformation to a new set of coordinates, now known as
Lagrangian coordinates, the forces of constraints are eliminated and the re-
sulting equations can be solved without any reference to the constraints. The
Lagrangian coordinates is a set of generalized coordinates usually denoted by
q1 , q2 , ..qn where n is the number of degrees of freedom of the system, tak-
ing into account the constraints on the motion of the system. The method
presented here works only for holonomic constraints.

3.1.1 Example: a particle on a cylinder

Consider a particle moving on the inner surface of a cylinder of radius R
as shown in fig. (3.1) The Newton’s equations of motion are written in an
inertial system say the (x,y) . There are two equations and one constraint.
The equation of motion are:
1
The book Mécanique analytique by Lagrange (1736-1813) summarizes all the me-
chanics till that time and includes Lagrange’s own numerous contributions to mechanics.
The impact of this work cannot be over estimated and in effect, it represents a revolution
in mechanics.

53
54 CHAPTER 3. LAGRANGIAN MECHANICS

g Fc
β θ

Figure 3.1: A particle moving on the inner surface of a cylinder.

mẍ = Fc cos β
mÿ = Fc sin β − mg
x2 + y 2 = R2 . (3.1)

Here Fc is the force of constraint and β is the angle between the force of
constraint and the x axis. Thus, in spite of the fact that the system has only
one degree of freedom there are two equations and one condition to solve.
The basic idea is to use the constraint and reduce the number of equations to
the number of degrees of freedom. How to perform this reduction? There are
many possibilities. However, the simplest one is just to take the angle θ as
the first Lagrangian coordinate. What about the second coordinate? If the
particle were free to move it would have moved all over the x, y plane. Define
therefore, a coordinate q2 as equal to the constraint namely, q2 = x2 + y 2.
Define now the following transformation from (x, y) to (q1 , q2 )

q1 = θ where θ + β = π/2 (3.2)

q2 = x2 + y 2 (3.3)

so that the constraints is simply

δq2 = 0. (3.4)
3.1. THE LAGRANGE GENERALIZED COORDINATES 55

R
θ

φ y

Figure 3.2: A particle moving without friction on a rotating hook .

The transformation from (q1 , q2 ) to (x, y) is simply

√
x = q2 cos q1 = R cos θ
√
y = q2 sin q1 = R sin θ. (3.5)
What has been accomplished by this transformation? First we have only one
differential equation to solve, the one for q1 . Secondly, the force of constraint
does not appear anymore in the equation. The system of two differential
equations and one algebraic condition has been reduced to one differential
equation with no additional conditions.

3.1.2 Example: a bead on a rotating wire hoop

Consider a bead forced to move on a rotating hoop of radius R (fig. 3.2). We
assume the motion to be friction less and the rotation speed to be constant
ω.
The first constraint is R2 = x2 + y 2 + z 2 and we assume q1 = R. We
choose as generalized coordinates θ and φ and the transformation is:
x(θ, t) = R sin θ cos(φ(t))
56 CHAPTER 3. LAGRANGIAN MECHANICS

y(θ, t) = R sin θ sin(φ(t))

z(θ, t) = −R cos θ (3.6)
The second constraint is φ(t) = φ0 + ωt where φ0 is the initial value of φ
which we take as zero. We assume q2 = φ. It is obvious that with the two
constraints the system has only one degree of freedom and the most sensible
Lagrange generalised coordinate is q3 = θ. Since the constraint is a moving
one of the generalised coordinates q2 is a known function of time. After
extensive algebra one gets:
! "
g
q̈3 + sin q3 − ω 2 cos q3 = 0. (3.7)
q1

3.1.3 Lagrangian coordinates

Assume that we have a system which has 3N degrees of freedom and on
which J holonomic constraints are imposed in an inertial system. Assume
further that the constraints can be written in the following form:
fi = fi (x1 , x2 , . . . x3N , t) f or i = 1, 2, . . . J (3.8)
The masses of the particles are denoted respectively, namely the mass mi is
given by the mass of particle j where j = int(i/3) + 1. Define the transfor-
mation
qi = fi f or i = 1, 2, . . . J, (3.9)
namely, we define the first J Lagrangian coordinates as equal to the J con-
straints. The rest k = 3N − J Lagrangian coordinates are chosen in an
appropriate way to the physical problem. For the transformation to be sin-
gle valued the Jacobian between the new and the old coordinates must be
different from zero namely,
∂(f1 , f2 , . . . f3N )
#= 0. (3.10)
∂(x1 , x2 , . . . x3N )
The non vanishing of the Jacobian guarantees that there is a one to one
correspondence between xi and qi . All transformations discussed here are
assumed to satisfy condition ( 3.10).
The equations for the constraints in the new system of coordinates are
simple. For the first J coordinates we have
qi = αi = Const f or i = 1, 2, . . . J. (3.11)
3.1. THE LAGRANGE GENERALIZED COORDINATES 57

The constants αi are absolute constants and do not depend on the initial
conditions of the problems. When the qi coordinates are fixed there are
no longer any constraints in the problem namely, in the formulation of the
problem in the new system of coordinate there are no equations for the
constraints because these are included in the system of coordinates. What is
the price for this success? The new system is not necessarily an inertial one
and hence Newton’s law may not always be valid in the simple form (we may
have to add complicated inertial forces). However the advantage of getting
rid of the cumbersome constraints overweights by far this loss.
What happens to the virtual displacements in the new system? The
differential of the old variable is given by
3N
# ∂xi ∂xi
dxi = dqs + dt f or i = 1, 2, . . . 3N, (3.12)
s=j+1 ∂qs ∂t

where the summation is only from s = J +1 because dqi = 0 for i ≤ J accord-

ing to the way we constructed the transformation. A virtual displacement
(δx1 , δx2 , . . . δx3N ) in the old system satisfies
3N
# ∂fk
δxi = 0 f or k = 1, 2, . . . J. (3.13)
i=1 ∂xi

Note that here the condition contains all the coordinates. The virtual dis-
placements in the new system are (δq1 , δq2 , . . . δq3N ). However, the first J
δqi displacements vanish identically because of condition (3.44) and we are
left with δqJ+1 , δqJ+2 , . . . δq3N at our free choice, in other words, the displace-
ments δqJ+1 , δqJ+2 , . . . are not constrained.
Let us demonstrate that condition (3.13)) is satisfied identically for any
δrs which is given by
3N
# ∂xs
δxs = δqk f or s = 1, 2, . . . 3N. (3.14)
k=J+1
∂qk

where δqJ+1 , δqJ+2 , . . . δq3N are completely arbitrary. To see this we substitute
the expression for δxs as obtained from equation (3.12) into the condition
(2.133) which is the condition for the virtual displacement and apply the
assumption of virtual displacements (so that the time is frozen). We find
3N 3N 3N 3N
! 3N "
# ∂fk # ∂fk # ∂xs # # ∂fk ∂xs
δxs = δqi = δqi
s=1 ∂xs s=1 ∂xs i=J+1 ∂qi i=J+1 s=1 ∂xs ∂qi
58 CHAPTER 3. LAGRANGIAN MECHANICS

3N
#
= δqi δki (3.15)
i=J+1

where δki is the Kronecker delta function namely, it is equal to unity if the
two indices are equal and i vanishes if they are different. But k ≤ J( there
are only J constraints ) and i ≥ J + 1 and hence the RHS always vanishes
irrespective of the value of δqi . The LHS in equation (3.44) is the condition
on the virtual displacements in the old system. We expressed this condition
in terms of the virtual displacements in the new system of coordinates and
we found that it contains only those δqi with i ≥ J + 1.But the RHS vanished
identically. Said in other words, a virtual displacement is any δqi with i ≥
J + 1. There is no more a condition on the virtual displacements δqi with
i ≥ J + 1. These displacements are completely arbitrary. The last result is
a very important one because till now when we encountered as expression
of the form i () δxi = 0 we could not claim that δxi is arbitrary and hence
$

each bracket must vanish separately because the displacements had to satisfy
condition (2.133). Now when we obtain an expression like i≥J+1 () δqi = 0
$

each δqi is arbitrary because according to the result (3.44) the LHS will
vanish irrespective of what we choose for δqi . Consequently in this case each
bracket vanishes separately.
The method to transform from the inertial frame of reference to the La-
grangian system applies only to holonomic constraints. Non holonomic con-
straints cannot be used as new coordinates since they are not integrable.

3.2 The transformation to Lagrange gener-

alised coordinates
The holonomic case
The equations of motions or Newton’s equation of motion were formulated in
an inertial frame of reference. Once we perform the transformation to a new
system of coordinates the equation of motion cannot be written in the same
way as before. Let us derive the equations of motion in the new Lagrange
coordinates. Let (q1 , q2 ., ....q3N , t) be the Lagrange coordinates for a system
with 3N degrees of freedom. Furthermore, let the transformation from the
new coordinates to the old ones be
xi = xi (q1 , q2 , . . . q3N , t). (3.16)
3.2. THE TRANSFORMATION TO LAGRANGE GENERALISED COORDINATES59

Since it is a general transformation the new system may not be an inertial

one. By simple differentiation of this expression we find
3N
# ∂xi ∂xi
ẋi = q̇j + . (3.17)
j=1 ∂qj ∂t

Two small results are required in order to proceed. The first one is
∂ ẋi ∂xi
= , (3.18)
∂ q̇j ∂qj
and it is derived by differentiating eq. (3.17) with respect to q̇j . Recall that
the new velocities do not appear in the expression for xi and that all the q " s
are independent of each other.
The second result is ! "
∂ ẋi d ∂xi
= (3.19)
∂ql dt ∂ql
(As if the time derivative applies only to the denumerator). This result is
derived in the following way. Return to equation (3.17) and differentiate with
respect to ql . We find
3N 3N
∂ ẋi # ∂ 2 xi # ∂xi ∂ q̇j ∂ 2 xi
= q̇j + +
∂ql j=1 ∂qj ∂ql j=1 ∂qj ∂ql ∂ql ∂t
3N
∂ 2 xi ∂ 2 xi
! "
# d ∂xi
= q̇j + = (3.20)
j=1 ∂qj ∂ql ∂ql ∂t dt ∂ql

The partial derivative ∂ q̇j /∂ql vanishes because qi and qj are independent
of each other or else the transformation would not be unique. The desired
result is obtained after changing the order of differentiation.
We can now perform the transformation to the new coordinates. The
kinetic energy is written as
3N
1#
T = mi ẋ2i . (3.21)
2 i=1

This form for the kinetic energy is correct only in inertial systems. Substitute
now the expression for ẋi in terms of q̇j into the kinetic energy eq. (3.21) to
get
Ekin = T0 + T1 + T2 , (3.22)
60 CHAPTER 3. LAGRANGIAN MECHANICS

where
3N
1 #
T2 = aij q̇i q̇j (3.23)
2 i,j=1
is a quadratic and homogeneous form of the second order in the new velocities
and aij are functions of the new coordinates. Except for this term, which
is the analog term to the form of the kinetic energy in inertial systems, we
have also the terms T1 and T0 which are
3N
#
T1 = ai q̇j , (3.24)
i=1

where T1 is a linear and homogeneous function in the new velocities, and

finally we have
T0 = f (q1 , q2 , . . . q3N , t), (3.25)
which is a function of the new coordinates. The new form for the kinetic
energy does not change the fact that the kinetic energy must be positive
definite. If the system can reach very large velocities then clearly T2 becomes
the dominant term and hence T2 must be also positive definite form. T1 can
be positive or negative. Finally, T0 is a non-negative function since it is given
by "2
3N
!
# ∂xi
T0 = mi . (3.26)
i=1 ∂t
We return to the equations of motion in the form of the d’Alembert principle
namely
3N
#
(mr ẍr − Fr ) · δxr = 0. (3.27)
r=1

Since this is a scalar condition, the transformation to the new coordinate

system is carried out by just substituting the corresponding values in the
new system. At this stage the virtual displacements are constrained and we
cannot claim that each bracket vanishes independently (it is actually equal to
the unknown forces of the constraints). The idea is to write this expression
in terms of the Lagrangian coordinates and reach a form in which each term
in the summation will vanish independently. To that goal we substitute the
expression
3N
# ∂xs
δxs = δq k f or s = 1, 2, . . . n. (3.28)
k=J+1 ∂qk
3.2. THE TRANSFORMATION TO LAGRANGE GENERALISED COORDINATES61

for the virtual displacement δxs in terms of the new coordinates q " s. Pay
attention to the fact that the summation is only for k = J + 1 to n because
the first J coordinates which were chosen to be equal to the constraints
equations have vanishing virtual displacements. The substitution of (3.28)
into (3.27) yields:
3N 3N
% &
# # ∂xr
(mr ẍr − Fr ) · δq k = 0. (3.29)
r=1 k=J+1 ∂qk

Consider the term ẍr ∂xr /∂qs and write it in the following form
! " ! "
∂xr d ∂xr d ∂xr
ẍr = ẋr − ẋr
∂qs dt ∂qs dt ∂qs
! "
d ∂ ẋr ∂ ẋr
= ẋr − ẋr , (3.30)
dt ∂ q̇s ∂qs

where eqs. (3.18) and (3.19) are used. Define now a generalized force Qs by
3N
# ∂xr
Qs = Fr · . (3.31)
r=1 ∂qs

The generalized force is the form the force takes in the new system of coor-
dinates. The exact physical meaning of Qs depends on the nature of the new
system of coordinates. Consider the coefficient of δqs in equation (3.29). We
find
3N
% ' ! " (&
# d ∂ ẋr ∂ ẋr
coef f of δqs = mr ẋr − ẋr − Qs
r=1 dt ∂ q̇s ∂qs
3N
∂ ẋ2r ∂ ẋ2
% ' ! " (&
1 d
− r
#
= mr − Qs
2 r=1 dt ∂ q̇s ∂qs
' (
d ∂T ∂T
= − − Qs (3.32)
dt ∂ q̇s ∂qs

where we introduced the summation under the derivative sign. Putting the
last result back into eq. (3.29) we find
3N
% ! " &
# d ∂T ∂T
− − Qs · δqs = 0 (3.33)
s=J+1 dt ∂ q̇s ∂qs
62 CHAPTER 3. LAGRANGIAN MECHANICS

The index r disappeared into the kinetic energy T and the generalised force
Qs . Note that QJ+1 δqJ+1 +. . . Q3N δq3N is the work that the generalized force
generates in a virtual displacements δqJ+1 , δqJ+2 . . . δq3N . Equation (3.34)
contains only the external forces (as generalized forces) and there is ’no men-
tion’ of the forces of constraints. The last equations are valid for any virtual
displacements and the δqs for s = J + 1, . . . n are completely independent of
each other. Our particular choice for the Lagrange coordinates has liberated
us from the equations of the constraints which in turn imposed conditions
on the virtual displacements δxr . We assume here that all the constraints
imposed on the system under discussion are holonomic. Since every δqs for
s = J + 1, . . . n is independent of the others, the coefficient of every δqs must
vanish. This is a property we did not have in the old system of coordinates.
We equate now each term to zero and get
! "
d ∂T ∂T
− = Qs . (3.34)
dt ∂ q̇s ∂qs

Assume further that the system is a conservative one and that all the external
forces are derived from a potential. Thus the force Fr is obtained as a gradient
of a function V , namely F = −∇V . Equivalently we can write
3N
#
δV = − Fr · δxr . (3.35)
r=1

Substitute now in (3.35) the expression for the virtual displacement in the
new system of coordinates to obtain
3N 3N #
3N 3N
# # ∂xr #
δV = − Fr · δxr = − Fr δqs = − Qs δqs , (3.36)
r=1 r=1 s=1 ∂qs s=J+1

in other words:
∂V
− = Qs . (3.37)
∂qs
The property of conservative forces that the force is given by the gradient
of the potential (in the relevant system of coordinates and with the reduced
degrees of freedom) does not depend on the system. This is not a classical
system transformation since we start from a system of N degrees of freedom
and J constraints, where Fi = ∂V /∂xi for i = 1, 2 . . . n and end with a system
3.2. THE TRANSFORMATION TO LAGRANGE GENERALISED COORDINATES63

of N − J degrees of freedom and Ql = ∂V /∂ql for l = J + 1, J + 2, . . . N. The

Lagrange equations become
3N
% ! " &
# d ∂T ∂T ∂V
− + · δqs = 0. (3.38)
s=J+1 dt ∂ q̇s ∂qs ∂qs

Always the generalized force times the differential of the generalized coor-
dinates give the work. Now comes the moment where we can exploit the
non-dependence of the δq " s and let the coefficient of each dq vanish. We find
! "
d ∂T ∂T ∂V
− + = 0. (3.39)
dt ∂ q̇s ∂qs ∂qs
Assume further that the potential function does not depend on the general-
ized velocities. Under such conditions we can define the Lagrange function
L = T − V and obtain the known form of the Lagrange equations for holo-
nomic and conservative systems, namely
! "
d ∂L ∂L
− . s = J + 1, J + 2 . . . N (3.40)
dt ∂ q̇s ∂qs
Note that the Largange equation is for the ’free’ coordinates qJ+1 . . . n and
not for those for which there are constraints. This is the reason why the
index runs from J + 1 to n and not over all coordinates.
When the generalized forces cannot be derived from a potential, then
eq.(3.37) does not apply. The Lagrange equation assume then the form:
! "
d ∂L ∂L
− = Qi . (3.41)
dt ∂ q̇i ∂qi
When the system contains only forces derivable from potentials, then the
knowledge of the Lagrangian is sufficient to determine the behavior of the
system. Additional information is required when this is not the case, for
example, the existence of forces which are not given as space gradients of a
potential function.
The Lagrange function plays a dominant role in physics that exceeds the
image you may get here that L is introduced as a mathematical trick to over-
come the problem of constraints etc. We started the mechanics by defining
the force along with Newton’s equation of motion. We derived the Lagrange
equations from Newton’s law. We can proceed in the opposite way and start
64 CHAPTER 3. LAGRANGIAN MECHANICS

by defining the Lagrange function of a system and derive from it all the prop-
erties of the system as well as the equations of motion. When starting with
Newton’s equations we have to guess the force, substitute it into the equa-
tions of motion, solve the equations and finally compare with observation. If
the predictions agree with the facts found in the laboratory we say that we
found the force on the system and we understand the behavior of the sys-
tem. If we fail we try another form for the force and repeat the process. The
same procedure is correct for the Lagrange function. We start by guessing a
function L. We solve the Lagrange equations and compare with the results
found in the laboratory. If the predictions are verified in the laboratory we
say we know the Lagrangian of the system and we understand its behav-
ior. Again, if the predictions fail we repeat the process. The assumptions of
holonomic constraints and conservative systems are almost always correct in
microscopic physics. The Lagrange equations can be derived directly from
the principle of Hamilton.

3.2.1 The non-holonomic case

Suppose now that the constraints are non holonomic or that few of the con-
straints are not holonomic. Let the form of the non-holonomic constraints
be
3N
#
Ajs dqs + Aj dt = 0 j = 1, 2, . . . J. (3.42)
s=1

In this example all constraint are non holonomic. The procedure is the same
as before except for the fact that now the δqs are not independent but they
satisfy the conditions of the virtual displacements which are in this case:
3N
#
Ajs dqs = 0 j = 1, 2, . . . J. (3.43)
s=1

The ingenious trick of defining the constraints as the new coordinates does
not help here. We have to find a way to to make the δqs independent. To
that goal introduce J new unknown variables λj and multiply each equation
(2.133) by λj and sum all the equations to get

J #
# 3N 3N
# J
#
Ajs dqs = δqs λj Ajs = 0 (3.44)
j=1 s=1 s=1 j=1
3.3. DERIVATION OF THE LAGRANGE EQUATIONS FROM HAMILTON’S PRINCIPLE65

Now add equation (3.44) to equation (3.34) and obtain:

 
3N  J
! "
# d ∂T ∂T # 
− − Qs + λj Ajs · δqs = 0 (3.45)
 dt ∂ q̇s ∂qs 
s=1 j=1

The λj are so far arbitrary and may be a function of time. The δqs on the
other hand, are bound by condition (3.44). There are altogether J arbitrary
λj added to the system and there are J conditions between the δqs to satisfy.
We make now a trade off. Instead of having the J conditions on the δqs we
impose on the λj the conditions
 
J
! "
 d ∂T ∂T # 
− − Qs + λj (t)Ajs ·δqs = 0 f or j = 1, 2, . . . J. (3.46)
 dt ∂ q̇s ∂qs 
j=1

Thus the first J brackets in expression (3.45) vanish. Hence, the value of
the first J δqs does not matter anymore. What about the remaining n − J
brackets? Since the first J δqs can take any value required by the constraint,
the n − J δqj for j = 1, . . . J are completely free to take any value. Hence,
the remaining brackets vanish as well. Once the δqs are independent of each
other we equate to zero each bracket in equation (2-3-28) and obtain:
 
J
! "
 d ∂T ∂T # 
− − Qs + λj (t)Ajs · δqs = 0 f or s = J + 1 . . . n.
 dt ∂ q̇s ∂qs 
j=1
(3.47)
We have J non holonomic conditions and n degrees of freedom. We end with
n − J Lagrange equations J unknowns and J conditions (3.44) to determine
the new unknowns. The λj are known as Lagrange multipliers. The Lagrange
multipliers express the force of constraint acting on the system. They are
therefore the generalized force in the case of non-holonomic constraints.

3.3 Derivation of the Lagrange equations from

Hamilton’s principle
Consider a holonomic system with a Lagrangian function L = L(qi , q̇i , t).
To simplify the notation, we assume that the independent coordinates are
q1 . . . qn and there are n degrees of freedom. Assume further that the sys-
tem’s path in the n dimension space of generalized coordinates is qi (t). The
66 CHAPTER 3. LAGRANGIAN MECHANICS

Hamilton’s principle states that:

/te
δ L(qj , q̇j , t)dt = 0, (3.48)
ts

where the variation is with repect to all virtual paths qj + δqj which satisfy
δqts = δqte = 0. (3.49)
Namely, all virtual paths start and end at the same points in time and
coordinates. By the way the coordinate system qi was constructed, all δqis
are independent. The Hamilton’s principle states that the path integral of
the Lagrange function along the path of the system is extremal. We dont
know (and do not care) whether the integral is maximum or minimum. We
consider the coordinates and the velocities as free independent variables. The
virtual displacements which change the location of the particle change also
the velocities of the praticle at the same point in time and space. Since we
consider here holonomic systems only, the virtual path of the system is also a
possible path of the system and hence we can restate the principle as follows:
of all possible paths which the system can choose, it selects the one along
which the integral of L is extremal. Let’s show how the Lagrange equations
follow from this principle. The change in the integral induced by the change
in the path along which the integration takes place is given by
/te /te #
 
n n
∂L # ∂L 
δ L(qj , q̇j , t)dt = δqj + δ q̇j dt. (3.50)
j=1 ∂qj j=1 ∂ q̇j
 
ts ts

Since we consider only virtual paths the time is frozen and consequetly we do
not differentiate the Lagrangian with respect to time even if it does depend
on time explicitely. We use now the identity
d
δqj δ q̇j = (3.51)
dt
which is a consequence of the fact that the time is frozen. We have now:
/te #
n /te #n
∂L ∂L d
δ q̇j dt = δqj dt
ts j=1 ∂ q̇j j=1 ∂ q̇j dt
ts
0ts
/te
0
0 n ! "
∂L 00 d ∂L
0
0#
= 0 δqj − δqj dt (3.52)
0
0j=1 ∂ q̇j 00 dt ∂ q̇j
ts ts
3.3. DERIVATION OF THE LAGRANGE EQUATIONS FROM HAMILTON’S PRINCIPLE67

Here we integrated by parts to obtain the second line. The integrated part
vanishes as the virtual displacements vanish at the end points. We are left
therefore with the result
/ts /te #
n
% ! " &
d ∂L ∂L
δ Ldt = − δqj dt (3.53)
j=1 dt ∂ q̇j ∂qj
ts ts

For the Hamilton’s principle we have to demand that the integral vanish for
every possible virtual displacement from the actual path of the system. Since
we are using Lagrangian coordinates and we have no constraints the virtual
dicplacements δqj are independent of each other. Hence for the principle to
be correct each bracket must vanish independently, in other words
! "
d ∂L ∂L
− =0 (3.54)
dt ∂ q̇j ∂qj

which are the Lagrange equations.

3.3.1 Examples for the use of Lagrange formalism

The Atwood machine
The simplest and classical example for the Lagrange equations is the Atwood
machine shown in fig. (3.3). Two non equal masses are hanged on a non
flexible cable. The total length of the cable is l. The simplest coordinates
are the distance of the masses from the horizontal line. The constraint is

x1 + x2 = l. (3.55)

The kinetic energy is given by

1 1
T = m1 ẋ21 + m2 ẋ22 (3.56)
2 2
and the potential energy is given by

V = −m1 gx1 − m2 gx2 . (3.57)

We chose the Lagrangian coordinates in the following way:

q1 = x1 + x2 ; q2 = x2 , (3.58)
68 CHAPTER 3. LAGRANGIAN MECHANICS

x1

x2

m1

m2

Figure 3.3: A typical Atwood machine.

so that the generalized velocities are

q̇1 = ẋ1 + ẋ2 ; q̇2 = ẋ2 . (3.59)
Due to the way the coordinates were selected we have
q̇1 = 0
= ẋ1 + q̇2 . (3.60)
The Lagrange function is therefore
1 1
T = m1 q̇22 + m2 q̇22
2 2
V = −m1 g (q1 + l − x2 ) = m2 gq2
= −m1 g (q1 + l − q2 ) = m2 gq2
1
L = T − V = (m1 + m2 )q̇22 + (m2 − m1 )gq2 + m1 g(q1 + l) (3.61)
2
The last term in L is constant and hence can be ignored. Constants in L
have no effect on the equations of motion (like a constant in the potential
energy). The Lagrange equations are therefore:
∂L
= (m2 − m1 )g
∂q2
3.3. DERIVATION OF THE LAGRANGE EQUATIONS FROM HAMILTON’S PRINCIPLE69

z z’

y’ y

x’

Figure 3.4: A rotating system of coordinates.

∂L
= (m1 + m2 )q̇2
∂ q̇2
! "
d ∂L
= (m1 + m2 )q̈2 . (3.62)
dt ∂ q̇2
The single equation of motion is:
(m1 + m2 )q̈2 = (m2 − m1 )g. (3.63)
The equation of motion was derived without any refence to the constrain-
ing tension in the cable. The force of constraint (the tension in the cable)
disappeard from the problem.

Rotating system of reference

Let a cartesian system of coordinates x, y, z be at rest and a second cartesian
system x" , y " , z " rotating with respect to the first one. The angular velocity
velocity is ω and the rotation axis coincides with the z axis (cf. fig. 3.4).
The problem is to transfer the Langrange equation for a free particle from the
system at rest to the rotating one. The Lagrange coordinates in the rotating
system will be:
x" = q1
70 CHAPTER 3. LAGRANGIAN MECHANICS

y " = q2
z " = q3 . (3.64)

The transformation to the new coordinates is:

x = x" cos ωt − y " sin ωt

y = x" sin ωt + y " cos ωt
z " = q3 (3.65)

Note that instead of using the usual notation q1 , q2 , q3 we use x" , y ", z " . There
is no potential energy and the kinetic energy is given by:
1
T = m(ẋ2 + ẏ 2 + ż 2 ). (3.66)
2
The substitution of the equation of the transformation into the expression
for the kinetic energy in the system at rest yields:
1 1
T = m(ẋ"2 + ẏ "2 + ẋ"2 ) + mω(x" ẏ " − ẋ" y ") + mω 2 (x"2 + y "2 ) (3.67)
2 2
The derivatives of the Lagrangian are:
∂L ∂T
= = mω ẏ " + mω 2 x"
∂x" ∂x"
∂L ∂T
"
= "
= −mω ẋ" + mω 2 y "
∂y ∂y
∂L ∂T
"
= = 0. (3.68)
∂z ∂z "
The equation of motion for the x" coordinate is then:
! "
d ∂L ∂L
− =0 (3.69)
dt ∂ ẋ" ∂x"
and hence: ! "
d ∂L
= mẋ" − mωy " (3.70)
dt ∂ ẋ"
Similarly the y " equation is:
! "
d ∂L
= mẏ " − mωx" (3.71)
dt ∂ ẏ "
3.3. DERIVATION OF THE LAGRANGE EQUATIONS FROM HAMILTON’S PRINCIPLE71

Carrying out the time derivative yields the final two Lagrange equations:

mẍ" − mω ẏ " = mω ẏ " + mω 2 x"

mÿ " − mω ẋ" = −mω ẋ" + mω 2 y " (3.72)

The equation for z " is identical to that for z. Collecting terms we find: :

mẍ" = 2mω ẏ " + mω 2 x"

mÿ " = −2mω ẋ" + mω 2 y " . (3.73)

Define now the rotation vector ω = (0, 0, ω), namely the rotation vector is in
the direction of the rotation axis and its absolute value equals to the angular
velocity.
The last two equations can be written in vector notation as

mr̈ = 2mṙ × ω + mω × (ω × r) (3.74)

The first term on the RHS is the Coriolis force while the second term is the
centrifugal one. We know that when we change the system of coordinates
in Newtonian mechanics we have to include inertial forces in the equations
of motion. This is not the case with the Lagrange equations. The inertial
forces enter the equations of motion in a natural way. Just choose a system
of coordinates and evaluate the Lagrange equations in the new system of
coordinates. The inertial forces are taken care of automatically. This is a
major advantage of the Lagrange equations. We can ask the following ques-
tion: a particle moves on a certain track. What are the accelerations of the
particle? The solution is simple in the Lagrangian formalism. Find a system
of coordinates where the constraint, namely the track, disappear. This is
done by taking the equation of the track as a generalized coordinate. The
remaining Lagrange equations will provide the required accelerations (and
inertial forces.) This fact implies that the Lagrange equations are invariant
under any transformation and not only under inertial system of coordinates.
Clearly, this is a unique advantage of the Lagrange equations.