8/23/2021 Double Pendulum

Math24
  Menu

Differential Equations
Systems of Equations

Double Pendulum

A double pendulum is undoubtedly an actual miracle of nature. The jump in complexity, which is
observed at the transition from a simple pendulum to a double pendulum is amazing. The
oscillations of a simple pendulum are regular. For small deviations from equilibrium, these
oscillations are harmonic and can be described by sine or cosine function. In the case of nonlinear
oscillations, the period depends on the amplitude, but the regularity of the motion holds. In other
words, in the case of a simple pendulum, the approximation of small oscillations fully reflects the
essential properties of the system.

https://www.math24.net/double-pendulum 1/16
8/23/2021 Double Pendulum

Double pendulum “behaves” quite differently. In the regime of small oscillations, the double
pendulum demonstrates the phenomenon of beats. The character of oscillations of the pendulums
changes radically with increasing energy − the oscillations become chaotic. Despite the fact that the
double pendulum can be described by a system of several ordinary differential equations, that is by
a completely deterministic model, the appearance of chaos looks very unusual. This situation is
reminiscent of the Lorenz system where a deterministic model of three equations also shows chaotic
behavior. Try to experiment with the application below and watch the movement of the double
pendulum at different mass ratios and initial angles.

Animation is available on desktop screens.

Next, we will build a mathematical model of the double pendulum in the form of a system of
nonlinear differential equations. Let’s start with the derivation of the Lagrange equations.

Lagrange Equations
In Lagrangian mechanics, evolution of a system is described in terms of the generalized
coordinates and generalized velocities. In our case, the deflection angles of the pendulums α1 , α2
 and the angular velocities α̇1 , α̇2  can be taken as the generalized variables. Using these variables,
we construct the Lagrangian for the double pendulum and write the Lagrange differential equations.

A simplified model of the double pendulum is shown in Figure 1.

Figure 1.

We assume that the rods are massless. Their lengths are l1  and l2 . The point masses (they are
represented by the balls of finite radius) are m1  and m2 . All pivots are assumed to be frictionless.

https://www.math24.net/double-pendulum 2/16
8/23/2021 Double Pendulum

We introduce the xy-coordinate system, the origin O of which coincides with suspension point of
the upper pendulum. The coordinates of the pendulums are defined by the following relationships:

x1 = l1 sin α1 , x2 = l1 sin α1 + l2 sin α2 , y1 = −l1 cos α1 , y2 = −l1 cos α1 − l2 cos α2 .

The kinetic and potential energy of the pendulums (respectively, T  and V ) are expressed by the
formulas

m1 v21 m2 v22 m1 (ẋ21 + ẏ21 ) m2 (ẋ22 + ẏ22 )

T= + = + , V = m1 gy1 + m2 gy2 .
2 2 2 2

Then the Lagrangian can be written as

m1 2 m
L = T − V = T1 + T2 − (V1 + V2 ) = (ẋ1 + ẏ21 ) + 2 (ẋ22 + ẏ22 ) − m1 gy1 − m2 gy2 .
2 2

Take into account that

ẋ1 = l1 cos α1 ⋅ α̇1 , ẋ2 = l1 cos α1 ⋅ α̇1 + l2 cos α2 ⋅ α̇2 ,

ẏ1 = l1 sin α1 ⋅ α̇1 , ẏ2 = l1 sin α1 ⋅ α̇1 + l2 sin α2 ⋅ α̇2 .

Consequently,

m1 2 m1 2 2 2 m1 2 2
T1 = (ẋ1 + ẏ21 ) = (l1 α̇1 cos α1 + l21 α̇21 sin2 α1 ) = l α̇ ,
2 2 2 1 1

m2 2 m2 2
T2 = (ẋ2 + ẏ22 ) = [(l1 α̇1 cos α1 + l2 α̇2 cos α2 )
2 2
2 m
+ (l1 α̇1 sin α1 + l2 α̇2 sin α2 ) ] = 2 [l21 α̇21 cos2 α1 + l22 α̇22 cos2 α2
2
+ 2l1 l2 α̇1 α̇2 cos α1 cos α2 + l21 α̇21 sin2 α1 + l22 α̇22 sin2 α2 + 2l1 l2 α̇1 α̇2 sin α1 sin α2 ]
m2 2 2 2 2
= [l α̇ + l α̇ + 2l1 l2 α̇1 α̇2 cos(α1 − α2 )],
2 1 1 2 2

V1 = m1 gy1 = −m1 gl1 cos α1 ,

V2 = m2 gy2 = −m2 g(l1 cos α1 + l2 cos α2 ).

As a result, the Lagrangian of the system takes the following form:

https://www.math24.net/double-pendulum 3/16
8/23/2021 Double Pendulum
m1 m m
L = T − V = T1 + T2 − (V1 + V2 ) = ( + 2 ) l21 α̇21 + 2 l22 α̇22
2 2 2
+ m2 l1 l2 α̇1 α̇2 cos(α1 − α2 ) + (m1 + m2 ) gl1 cos α1 + m2 gl2 cos α2 .

Now we can write the Lagrange equations (sometimes they are called as the Euler-Lagrange

equations):

d ∂L ∂L
− = 0, i = 1, 2.
dt ∂ α̇i ∂αi

The partial derivatives in these equations are expressed by the following formulas:

∂L
 = (m1 + m2 ) l21 α̇1 + m2 l1 l2 α̇2 cos(α1 − α2 ),
∂ α̇1

∂L
 = −m2 l1 l2 α̇1 α̇2 sin(α1 − α2 ) − (m1 + m2 ) gl1 sin α1 ,
∂α1

∂L
 = m2 l22 α̇2 + m2 l1 l2 α̇2 cos(α1 − α2 ),
∂ α̇2

∂L
 = m2 l1 l2 α̇1 α̇2 sin(α1 − α2 ) − m2 gl2 sin α2
∂α2

Hence, the first Lagrange equation can be written as

d
[(m1 + m2 ) l21 α̇1 + m2 l1 l2 α̇2 cos(α1 − α2 )] + m2 l1 l2 α̇1 α̇2 sin(α1 − α2 )
dt
+ (m1 + m2 ) gl1 sin α1 = 0,

⇒ (m1 + m2 ) l21 α̈1 + m2 l1 l2 α̈2 cos(α1 − α2 ) + m2 l1 l2 α̇22 sin(α1 − α2 )

+ (m1 + m2 ) gl1 sin α1 = 0.

Cancelling l1 ≠ 0, we obtain:

(m1 + m2 ) l1 α̈1 + m2 l2 α̈2 cos(α1 − α2 ) + m2 l2 α̇22 sin(α1 − α2 ) + (m1 + m2 ) g sin α1 = 0.

Similarly, we derive the second differential equation:

d
[m2 l22 α̇2 + m2 l1 l2 α̇1 cos(α1 − α2 )] − m2 l1 l2 α̇1 α̇2 sin(α1 − α2 ) + m2 gl2 sin α2 = 0,
dt

https://www.math24.net/double-pendulum 4/16
8/23/2021 Double Pendulum

⇒ m2 l22 α̈2 + m2 l1 l2 α̈1 cos(α1 − α2 ) − m2 l1 l2 α̇21 sin(α1 − α2 ) + m2 gl2 sin α2 = 0.

After canceling m2 l2 ≠ 0 the equation takes the form

l2 α̈2 + l1 α̈1 cos(α1 − α2 ) − l1 α̇21 sin(α1 − α2 ) + g sin α2 = 0.

Thus, the nonlinear system of two Lagrange differential equations can be written as

(m1 + m2 ) l1 α̈1 + m2 l2 α̈2 cos(α1 − α2 ) + m2 l2 α̇22 sin(α1 − α2 ) + (m1 + m2 ) g sin α1 = 0.

l2 α̈2 + l1 α̈1 cos(α1 − α2 ) − l1 α̇21 sin(α1 − α2 ) + g sin α2 = 0.

Small Oscillations of the Double Pendulum

Assuming that the angles α1 (t) , α2 (t) are small, the oscillations of the pendulums near the zero
equilibrium point can be described by a linear system of equations. To get such a system, let’s get
back to the original Lagrangian of the system:

m1 m m
L=T −V =( + 2 ) l21 α̇21 + 2 l22 α̇22 + m2 l1 l2 α̇1 α̇2 cos(α1 − α2 )
2 2 2
+ (m1 + m2 ) gl1 cos α1 + m2 gl2 cos α2 .

We write this Lagrangian in a simpler form, expanding it in a Maclaurin series and retaining the
linear and quadratic terms. The trigonometric functions can be replaced by the following
approximate expressions:

2
α21 α22 (α1 − α2 )
cos α1 ≈ 1 − , cos α2 ≈ 1 − , cos(α1 − α2 ) ≈ 1 − ≈ 1.
2 2 2

Here we have taken into account that the term with cos(α1 − α2 ) contains the product of small
quantities α̇1 α̇2  and has the second order of smallness. Therefore, we can leave only the linear term
in the cosine expansion.

Substituting this in the original Lagrangian and considering that the potential energy is defined up
to a constant, we obtain the quadratic Lagrangian for the double pendulum in the form:

m1 m2 2 2 m2 2 2 m1 m2
L=T −V =( + ) l1 α̇1 + l2 α̇2 + m2 l1 l2 α̇1 α̇2 − ( + ) gl1 α21
2 2 2 2 2
m2
+ gl2 α22 .
2

We derive the Lagrange differential equations for the given Lagrangian. They are written as

https://www.math24.net/double-pendulum 5/16
8/23/2021 Double Pendulum

d ∂L ∂L d ∂L ∂L
− = 0, − = 0.
dt ∂ α̇1 ∂α1 dt ∂ α̇2 ∂α2

Find the partial derivatives:

∂L
= (m1 + m2 ) l21 α̇1 + m2 l1 l2 α̇2 ,
∂ α̇1

∂L
= − (m1 + m2 ) gl1 α1 ,
∂α1

∂L
= m2 l22 α̇2 + m2 l1 l2 α̇1 ,
∂ α̇2

∂L
= −m2 gl2 α2 .
∂α2

We get the system of two differential equations

d
[(m1 + m2 ) l21 α̇1 + m2 l1 l2 α̇2 ] + (m1 + m2 ) gl1 α1 = 0,
dt

d
[m2 l22 α̇2 + m2 l1 l2 α̇1 ] + m2 gl2 α2 = 0.
dt

or

(m1 + m2 ) l21 α̈1 + m2 l1 l2 α̈2 + (m1 + m2 ) gl1 α1 = 0,

m2 l1 l2 α̈1 + m2 l22 α̈2 + m2 gl2 α2 = 0.

This system of equations can be written in a compact matrix form. We introduce the matrices

α1 (t) (m1 + m2 ) l21 m2 l 1 l 2

α (t) = [ ], M = [ ],
α2 (t) m2 l 1 l 2 m2 l22
(m1 + m2 ) gl1 0 0
K=[ ], 0 = [ ].
0 m2 gl2 0

Then the system of differential equations can be represented as

M α̈ + Kα = 0.

https://www.math24.net/double-pendulum 6/16
8/23/2021 Double Pendulum

In the case of one body, this equation describes the free undamped oscillations with a certain
frequency. In the case of the double pendulum, the solution (as you will see below) will contain
oscillations with two characteristic frequencies, which are called normal modes. The normal modes
are the real part of the complex-valued vector function

α1 (t) H
α (t) = [ ] = Re ([ 1 ] eiωt ) ,
α2 (t) H2

where H1 , H2  are the eigenvectors, ω is the real frequency. The values of the normal frequencies 
ω1,2  are determined by solving the auxiliary equation

det (K − ω2 M) = 0.

In the case of arbitrary masses m1 , m2  and lengths l1 , l2 , the auxiliary equation takes the form

(m1 + m2 ) g 2 − ω2 (m1 + m2 ) (l1 + l2 ) g + ω4 m1 l1 l2 = 0.

Thus, we have a biquadratic equation for the frequencies ω. The general solution for this equation is
somewhat cumbersome. Therefore we consider the case when the lengths of the rods of both
pendulums are equal: l1 = l2 = l. Then the normal frequencies will be determined by a compact
formula

g m
ω21,2  =  = [1 + μ ± √(1 + μ) μ] , where μ = 2 .
l m1

m2
As it can be seen, the eigenfrequencies ω1,2  depend only on the mass ratio μ = m1 . The
g
dependencies of the frequencies ω1 , ω2  on the parameter μ (provided  = 1) are shown
l
in Figure 2.

Figure 2.
https://www.math24.net/double-pendulum 7/16
8/23/2021 Double Pendulum

For equal masses m1 = m2 = m, that is when μ = 1, the frequencies are given by

g
ω1,2 = √ √2 ± √2.
l

Now, after the eigenfrequencies ω1,2  are known, we still have to determine the eigenvectors H1,2  to
describe the normal modes. They can be found by solving the vector-matrix equation

(K − ω2 M) H = 0.

T
Let the eigenvector H1 = (H11 , H21 )  (the superscript T  denotes transposition) be corresponded to
the normal frequency ω1 . Then we have the following equation for H1 :

g
(K − ω21 M) H1 = 0, where ω21 = [1 + μ + √(1 + μ) μ],  ⇒ 
l
(1+μ) (g − ω21 l) −ω21 μl H11
m1 l⋅[ ]⋅ [ ] = 0.
−ω21 μl μ (g − ω21 l) H21

The coordinates of the eigenvector H1  satisfy the equation

(1 + μ) (g − ω21 l) H11 − ω21 μlH21 = 0,

(1 + μ) [μ + √(1 + μ) μ] 1+μ
H
⇒ 21  = − = −√ .
H11 μ
μ [1 + μ + √(1 + μ) μ]

Thus, the eigenvector H1  is given by

1
H11
H1 = [ ]=[ 1+μ ] .
H21 −√ μ

T
Similarly, we find the coordinates of the second eigenvector H2 = (H12 , H22 ) :

H12 1
H2 = [ ] = [ 1+μ ] .
H22 √ μ

The general solution of the matrix equation can be written as

https://www.math24.net/double-pendulum 8/16
8/23/2021 Double Pendulum

α1 (t) H 1
α (t) = [ ] = Re ([ 1 ] eiωt ) = C1 [ 1+μ ] cos(ω1 t + φ1 )
α2 (t) H2 −√ μ
1
+ C2 [ 1+μ ] cos(ω2 t + φ2 ),
√ μ

where the constants C1 , C2 , φ1 , φ2  depend on the initial positions and velocities of the pendulums.

Consider the character of small oscillations for a specific set of initial data. Suppose, for example,
that the initial positions and velocities of the pendulums have the following values:

π
α1 (t = 0) = 0, α2 (t = 0) = , α̇1 (t = 0) = 0, α̇2 (t = 0) = 0.
6

In this case, the initial phases are zero: φ1 = φ2 = 0. Determine the constants C1  and C2 :

⎧ α1 (0) = C1 + C2 = 0 1+μ π
⎨ , ⇒ C1 = −C2 , ⇒ 2C2 √ = ,
⎩ α2 (0) = −C1 √ 1+μ + C2 √ 1+μ = π
μ 6
μ μ 6

π μ π μ
⇒ C2 = √ , ⇒ C1 = − √ .
12 1 + μ 12 1 + μ

Then the law of oscillations of the pendulums is expressed by the formulas

π μ π μ
α1 (t)  = − √ cos(ω1 t) + √ cos(ω2 t),
12 1 + μ 12 1 + μ

π π
α2 (t)  =  cos(ω1 t) + cos(ω2 t),
12 12

where the angular frequencies ω1,2  are given by

g
ω1,2 = √ ⋅√1 + μ ± √(1 + μ) μ.
l

Here the angles α1 (t) , α2 (t) are expressed in radians, and the time t in seconds. Figures 3 − 5
 show plots of small oscillations for three values of μ : μ1 = 0.2, μ2 = 1, μ3 = 5,provided 
m
l = l1 = l2 = 0.25 m, g = 9.8 2 . For convenience, the deflection angles of the pendulums are
s
given in degrees.

https://www.math24.net/double-pendulum 9/16
8/23/2021 Double Pendulum

Figure 3.

Figure 4.

Figure 5.

As can be seen from the graphs, the energy is cyclically transferred between the two bobs. When
one of the pendulums almost stops, the other swings with maximum amplitude. After some time,
the bobs “switch roles” and so on.

ω1 +ω2
The resulting oscillations are represented as oscillations at the higher frequency ω̄ = with a
2
ω1 −ω2
periodic amplitude modulation with frequency Δω = 2 , which is also known as the beat
frequency.

https://www.math24.net/double-pendulum 10/16
8/23/2021 Double Pendulum

Thus, the small oscillations of the double pendulum look as periodic changes and are described by
the sum of two harmonics with frequencies ω1 , ω2  depending on the system parameters.

Strictly speaking, the small oscillations of the double pendulum will be periodic if only the ratio of
the eigenfrequencies ω1 , ω2  is equal to a rational number. If the ratio of the frequencies is an
irrational number, the small oscillations cannot possibly be periodic. For more details see the
topic Sum of two periodic functions is periodic?

Legendre Transformation and Hamilton Equations

We now turn back to the original nonlinear system of equations and examine the character of
oscillations of arbitrary amplitude. This system of equations can not be solved analytically.
Therefore, we consider a numerical model of the double pendulum.

The Lagrange equations given above are second order differential equations. It is more
conveniently to convert them into the form of Hamilton’s canonical equations. As a result, instead
of the two second-order equations, we obtain a system of four differential equations of the first
order.

In Hamiltonian mechanics, the state of a system is determined by the generalized

coordinatesand generalized momenta. In our case, we can use again as in the Lagrange equations
the angles α1 , α2  as the generalized coordinates. Instead of the generalized velocities α̇1 , α̇2  (in the
Lagrangian), we now introduce the generalized momenta p1 , p2 , related to the velocities by the
formulas

∂L ∂L
p1 = , p2 =
∂ α̇1 ∂ α̇2

or in the brief description:

∂L
pi = , i = 1, 2.
∂ α̇i

The transition from the Lagrangian to the Hamiltonian form of the equations is performed using
the Legendre transformation, which is defined as follows.

Suppose that f (x) is a smooth convex downward function (Figure 6).

https://www.math24.net/double-pendulum 11/16
8/23/2021 Double Pendulum

Figure 6.

Consider the line y = px passing through the origin. The distance between the line y = pxand the
function y = f (x) along the y-axis depends on the coordinate x. This distance will be maximal at a
certain value of x. Clearly, it depends on the slope of the line, that is on the parameter p. Thus, we
introduce a new function g (p) :

g (p) = max (px − f (x)) .

x

Such transformation of the function f (x) into the conjugate function g (p) is called the Legendre
transformation. Note that the function g (p) reaches a maximum value with respect to the variable x
df
 when p = . Indeed,
dx

d df (x) df (x)
(px − f (x)) = p − = 0, ⇒ p = = p (x) .
dx dx dx

Knowing the dependence p (x) , we can find the inverse function x (p) . Then the Legendre
transform is expressed by the relationship

df
g (p) = px (p) − f (x (p)) , where p = .
dx

The Legendre transformation is easily generalized to the case of functions of several variables. In
the model of the double pendulum, the transition from the Lagrangian to the Hamiltonian is
described by the Legendre transformation of the form:

https://www.math24.net/double-pendulum 12/16
8/23/2021 Double Pendulum
2
H (α1 , α2 , p1 , p2 )  = ∑ α̇i pi − L (α1 , α2 , α̇1 , α̇2 ) = α̇1 p1 + α̇2 p2
i=1
∂L ∂L
− L (α1 , α2 , α̇1 , α̇2 ) , where p1 = , p2 = .
∂ α̇1 ∂ α̇2

In this expression, L is the Lagrangian, and the function H  is the Hamiltonian of the system, which
depends on the generalized coordinates α1 , α2  and the generalized momenta p1 , p2 .

As a result of this transformation, each Lagrange equation becomes a system of two Hamilton’s
canonical equations of the form:

d ∂L ∂L ⎧ ṗi = − ∂H
∂αi
= , ⇒⎨ .
dt ∂ α̇i ∂αi ⎩ α̇i = ∂H
∂p i

Omitting the formal (and cumbersome) algebraic transformations, we write the Hamilton’s
canonical equations for the double pendulum in the final form:

p1 l2 −p2 l1 cos(α1 −α2 )

⎧
⎪
α̇1 = 2
⎪

⎪
l1 l2 [m1 +m2 sin2 (α1 −α2 )]
⎪

⎪ p2 (m1 +m2 )l1 −p1 m2 l2 cos(α1 −α2 )

α̇2 =
⎨ m2 l1 l22 [m1 +m2 sin2 (α1 −α2 )]
⎪

⎪
ṗ1  = − (m1 + m2 ) gl1 sin α1 − A1 + A2
⎪

⎩
⎪
ṗ2  = −m2 gl2 sin α2 + A1 − A2

where

p1 p2 sin(α1 − α2 )
A1  =  ,
l1 l2 [m1 + m2 sin2 (α1 − α2 )]

1
A2  =  2
⋅[p21 m2 l22 − 2p1 p2 m2 l1 l2 cos(α1 − α2 )
2
2l21 l22 [m1 + m2 sin (α1 − α2 )]

+ p22 (m1 + m2 ) l21 ]⋅sin[2 (α1 − α2 )]

We can now proceed to the numerical analysis of the equations.

Numerical Simulation of Chaotic Oscillations

The most common method of numerical solution of differential equations is the 4th or 
5thorder Runge-Kutta method. Different variations of this method are used in most mathematical
packages (usually with an automatic error control and adaptive time-stepping).
https://www.math24.net/double-pendulum 13/16
8/23/2021 Double Pendulum

To model the motion of the double pendulum, we also use the classical 4th order Runge-Kutta
method (RK4) . We somewhat simplify the differential equations assuming that the lengths of the
pendulums are the same: l1 = l2 = l. By introducing the parameter μ equal to the mass ratio: 
m
μ = m2 , we can write the system in the following form:
1

p1 −p2 cos(α1 −α2 )

⎧
⎪
α̇1 =
⎪

⎪
m1 l2 [1+μ sin2 (α1 −α2 )]
⎪

⎪ p2 (1+μ)−p1 μ cos(α1 −α2 )

α̇ =
⎨ 2 m1 l2 [1+μ sin2 (α1 −α2 )]
⎪

⎪
ṗ1 = −m1 (1 + μ) gl sin α1 − A1 + A2
⎪

⎩
⎪
ṗ2 = −m1 μgl sin α2 + A1 − A2

where

p1 p2 sin(α1 − α2 )
A1  =  ,
2
l2
m1 [1 + μ sin (α1 − α2 )]

1
A2  =  ⋅[p21 μ − 2p1 p2 μ cos(α1 − α2 ) + p22 (1 + μ)]⋅
2
2m1 l2 [1 + μ sin2 (α1 − α2 )]
sin[2 (α1 − α2 )]

This system can be rewritten in vector form:

T T
Z′ = f (Z) , where Z = (α1 , α2 , p1 , p2 ) , f = (f1 , f2 , f3 , f4 ) .

The vector Z is composed of 4 canonical variables of the system, and components of the vector f
 correspond to the right hand sides of the differential equations.

The Runge-Kutta (RK4) method requires at each step sequential evaluation of the four intermediate
vectors:

1 1
Y1 = τf (Zn ) , Y2 = τf (Zn + Y1 ) , Y3 = τf (Zn + Y2 ) , Y4 = τf (Zn + Y3 ) .
2 2

The value of the vector Zn+1  in the next node is given by

1
Zn+1 = Zn + (Y1 + 2Y2 + 2Y3 + Y4 ) .
6

The total error of the algorithm on a finite interval has the order O (τ 4 ) , i.e. the computational
accuracy increases by 16 times while reducing the time step τ  twice.

https://www.math24.net/double-pendulum 14/16
8/23/2021 Double Pendulum

The described model is implemented in the animation shown at the beginning of the web page. For
simplicity, we assume that the initial deflection angles of the pendulums are equal: α1 = α2 = α.
 This application demonstrates the chaotic dynamics of the double pendulum for different values of 
μ and α. Interestingly, in some regimes, compact regions of attraction such as in Figure 7 appear in
the system. It seems that the nonlinear dynamics of double pendulum is not yet fully studied by
physicists and mathematicians and carries a lot of surprises.

Figure 7 (μ = 2.75, α = 171°).

Recommended Pages

Method of Matrix Exponential

Equilibrium Points of Linear Autonomous Systems

Method of Lyapunov Functions

Double Pendulum

Mass-Spring System

Properties of Matrices

https://www.math24.net/double-pendulum 15/16
8/23/2021 Double Pendulum

Legal
Copyright © 2021 math24.net
Alex Svirin, PhD 

Privacy Policy

Contact
info@math24.net

Tools

https://www.math24.net/double-pendulum 16/16