int. j. math. educ. sci. technol.

, 2003
vol. 34, no. 1, 6579

Coupled spring equations

TEMPLE H. FAY*
Technikon Pretoria and Mathematics, University of Southern Mississippi, Box 5045,
Hattiesburg, MS 39406-5045, USA
E-mail: thfay@hotmail.com

and SARAH DUNCAN GRAHAM

University of Southern Mississippi
(Received 12 September 2002)
Coupled spring equations for modelling the motion of two springs with
weights attached, hung in series from the ceiling are described. For the linear
model using Hookes Law, the motion of each weight is described by a fourthorder linear dierential equation. A nonlinear model is also described and
damping and external forcing are considered. The model has many features that
permit the meaningful introduction of many concepts including: accuracy of
numerical algorithms, dependence on parameters and initial conditions, phase
and synchronization, periodicity, beats, linear and nonlinear resonance, limit
cycles, harmonic and subharmonic solutions. These solutions produce a wide
variety of interesting motions and the model is suitable for study as a computer
laboratory project in a beginning course on dierential equations or as an
individual or a small-group undergraduate research project.

1. Introduction
The classical syllabus for beginning dierential equations is rapidly changing
from emphasizing solution techniques for a variety of types of dierential
equations to emphasizing systems and more qualitative aspects of the theory of
ordinary dierential equations. In particular, there is an emphasis on nonlinear
equations due largely to the wide availability of high powered numerical algorithms and almost eortless graphics capabilities that come with computer algebra
systems such as Mathematica and Maple.
In this article, we investigate an old problem that appears now to be relegated
to the exercises in texts, if it appears at all (see for example [1, pp. 220221]). This
is the problem of two springs and two weights attached in series, hanging from the
ceiling. Under the assumption that the restoring forces behave according to
Hookes Law, this two degrees of freedom problem is modelled by a pair of
coupled, second-order, linear dierential equations. By dierentiating and substituting one equation into the other, the motion of each weight can be shown to be
determined by a linear, fourth-order dierential equation. We like this example for
this very reason, most models in elementary texts are only of second order.
Moreover, the questions about phase now have a very nice physical interpretation;
* The author to whom correspondence should be addressed.
International Journal of Mathematical Education in Science and Technology
ISSN 0020739X print/ISSN 14645211 online # 2003 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/0020739021000029258

66

T. H. Fay and S. D. Graham

we can investigate when the motions of the two weights are synchronized (in
phase) or opposing each other (1808 out of phase). By tinkering with the spring
constants, oscillatory motions can be produced that are much more interesting
than those obtained from the classical single spring model. Moreover, there are
other phenomena that can be investigated.
We also demonstrate, through examples, that interesting motions can arise
when a slight nonlinearity is introduced in an attempt to make the restoring force
more physically meaningful. In this situation, periodicity of the solutions becomes
a more delicate matter. If forcing is introduced, subharmonic solutions of long
periods can be found which again exhibit more interesting motions than for the
classical linear case.
There is the opportunity here for many numerical and graphical investigations
to be made by students. Periodicity, amplitude, phase, sensitivity to inital conditions, and many more concepts can be investigated by modifying the parameters
in the model. And of course, it would not be dicult for students to derive a model
for three springs and three weights (or more), and to investigate the various
motions that could arise from both linear and nonlinear restoring forces.

2. The coupled spring model

The model consists of two springs and two weights. One spring, having spring
constant k1 , is attached to the ceiling and a weight of mass m1 is attached to the
lower end of this spring. To this weight, a second spring is attached having spring
constant k2 . To the bottom of this second spring, a weight of mass m2 is attached
and the entire system appears as illustrated in gure 1.
Allowing the system to come to rest in equilibrium, we measure the displacement of the centre of mass of each weight from equilibrium, as a function of time,
and denote these measurements by x1 t and x2 t respectively.
2.1. Assuming Hookes Law
Under the assumption of small oscillations, the restoring forces are of the
form k1 l1 and k2 l2 where l1 and l2 are the elongations (or compressions) of

Figure 1.

The coupled springs.

Coupled spring equations

67

the two springs. Since the upper mass is attached to both springs, there are
two restoring forces acting upon it: an upward restoring force k1 x1 exerted by
the elongation (or compression) x1 of the rst spring; an upward force k2 x2  x1
from the second springs resistance to being elongated (or compressed) by
the amount x2  x1 . The second mass only feels the restoring force from
the elongation (or compression) of the second spring. If we assume there are no
damping forces present, then Newtons Law implies that the two equations
representing the motions of the two weights are
m1 x1 k1 x1  k2 x1  x2
2:1
m2 x2 k2 x2  x1
Thus we have a pair of coupled second-order linear dierential equations.
In order to nd an equation for x1 that doesnt involve x2 , we solve the rst
equation for x2 , obtaining
x2

m1 x1 k1 k2

x1
k2
k2

2:2

Substituting for x2 in the second dierential equation (2.1), and simplifying, we

obtain
4

m1 m2 x1 m2 k1 k2 m1 m2
x1 k1 k2 x1 0

2:3

Hence the motion of the rst weight is determined by this fourth-order linear
dierential equation.
Now to nd an equation that only involves x2 , we solve the second equation
(2.1) for x1 :
x1

m2
x2 x2
k2

2:4

and substitute into the rst equation (2.1), producing the equation
4

x2 k1 k2 x2 0
m1 m2 x2 m2 k1 k2 m1 m2

2:5

This is exactly the same fourth-order equation as that for the motion of the rst
weight. Thus the motions of each weight obey the same dierential equation, and
it is only the initial displacements and initial velocities that are needed in order to
completely determine any specic case.
Typically, in a model of this sort, we would be given the initial displacements
x1 0 and x2 0, and the initial velocities x_ 1 0 and x_ 2 0. But in order to solve
equations (2.3) and (2.5), we need to know the values of x1 0, EEE
x1 0, x2 0 and
EEE
x2 0. The values of the second derivatives are determined by evaluating equations
(2.2) and (2.4) at time t 0. The values of the third derivatives are obtained by
evaluating the derivatives of equations (2.2) and (2.4) at time t 0. Thus the
motions for any set of initial conditions are determined by solving two fourthorder initial value problems.
Alternatively, we can turn this pair of coupled second-order equations into a
system of four rst-order equations by setting x_ 1 u and x_ 2 v, so we have

68

T. H. Fay and S. D. Graham

x_ 1 u
u_ 

k1
k2
x1 
x1  x2
m1
m1

2:6

x_ 2 v
v_ 

k2
x2  x1
m2

and we need only consider the four initial conditions x1 0, u0, x2 0, and v0.
2.2. Some examples with identical weights
Let us consider the model having the two weights of the same mass. This
model may be normalized by setting m1 m2 1. In the case of no damping and
no external forcing, the characteristic equation of the dierential equations (2.3)
and (2.5) is
m4 k1 2k2 m2 k1 k2 0

2:7

which has the roots



r
q
12k1  k2  12

k21 4k22

2:8

Example 2.1. Describe the motion for spring constants k1 6 and k2 4 with
initial conditions x1 0; x_ 1 0, x2 0, x_ 2 0 1; 0; 2; 0.
p
It is easy to see that the roots of the characteristic equation are  2i and
p
2 3i. Thus the general solution to equations (2.3) and (2.5) is
p
p
p
p
xt c1 cos 2t c2 sin 2t c3 cos 2 3t c4 sin 2 3t
2:9
x1 0 1; 0; 2; 0. This leads
It is also easy to compute that x1 0; x_ 1 0; x1 0; EEE
to the two pairs of simultaneous equations
p
p
c1 c3 1
2c2 2 3c4 0
2:10
p
p
 2 2c2  24 3c4 0
2c1  12c3 2
whose solutions are c1 1, c3 0, and c2 c4 0. Thus the unique solution for
x1 t is
p
2:11
x1 t cos 2t
x2 0 2; 0; 2; 0. This leads
It is also easy to compute that x2 0; x_ 2 0; x2 0; EEE
to the two pairs of simultaneous equations
p
p
c1 c3 2
2c2 2 3c4 0
2:12
p
p
2c1  12c3 4
 2 2c2  24 3c4 0
whose solutions are c1 2, c3 0, and c2 c4 0. The unique solution for x2 t is
p
2:13
x2 t 2 cos 2t

Coupled spring equations

69

The motion here is synchronized and thus the weights move in phase with each
other, having the same period of motion, merely having dierent amplitudes; this
is shown in gure 2.1. Since the motion is simple periodic motion, the phase
portraits for x1 and x2 are simple closed curves (ellipses) as shown in the left-hand
frame in gure 2.2. Shown in the right-hand frame of gure 2.2 is a plot of x1
against x2 ; observe this plot is a straight line of slope 2.
The next example illustrates 1808 out of phase motion.
Example 2.2. Describe the motion for spring constants k1 6 and k2 4 with
initial conditions x1 0; x_ 1 0; x2 0; x_ 2 0 2; 0; 1; 0.
In this example, for x1 , we have c1 0, c3 2, and c2 c4 0, and accordingly

Figure 2.1.

Plot of x1 and x2 showing synchronized motion.

Figure 2.2.

Plots for Example 2.1.

70

T. H. Fay and S. D. Graham

p
x1 t 2 cos 2 3t

For x2 , we have c1 0, c3 1, and c2 c4 0; thus

p
x2 t cos 2 3t

2:14

2:15

While the rst weight is moving downward, the second weight is moving upward
and when the rst is moving upward, the second is moving downward. Again both
motions have the same period, the motions are 1808 out of phase. This is shown in
the left-hand frame of gure 2.3. In the right-hand frame we plot x1 against x2 and
observe a straight line of slope 1.
Generally speaking, the motions of the two weights willpnot
since
be periodic,
p
p
the solutions
for
x
and
x
are
linear
combinations
of
sin
2
t,
cos
2
t,
sin
2
3t,
1
2
p
p
p
and cos 2 3t and the ratio of the frequencies 2 and 2 3 is not rational. For more
on the period of such combinations we refer the reader to [2].
By tinkering with the parameters in this model, more interesting motions than
those just described can be obtained.
Example 2.3. Describe the motion for spring constants k1 0:4 and
k2 1:808 with initial conditions x1 0; x_ 1 0; x2 0; x_ 2 0 1=2; 0; 1=2; 7=10.
From equation (2.8), we see that it is the k1 and k2 values that completely
determine the period and hence frequency of the response of this symmetric
weight problem. Essentially the initial conditions only aect the amplitude and
phase of the solutions. Thus, using a numerical solver and tweaking the k values
and experimenting with the initial conditions led to the choice of parameters in this
example. This type of numerical and graphical interactive investigation can be
carried out with a computer algebra system almost eortlessly. The phase portraits
for the two solutions are shown in the top row of gure 2.4. These portraits show
an interesting (almost?) period motion. Plots of the actual solutions are shown in
the middle row. Plotting both solutions together on the same coordinate system
also shows an interesting pattern. This is shown in the left-hand frame of the
bottom row of gure 2.4. Finally, the plot of x1 against x2 shown in the right-hand
frame of the bottom row of gure 2.4 appears to be a Lissajous type curve.
Student problem. Verify analytically the solutions to this initial value problem
are periodic. Find the period. Use the analytic solution to produce other periodic
motions.

Figure 2.3.

Plots for Example 2.2.

Coupled spring equations

Figure 2.4.

71

Plots for Example 2.3.

2.3. Damping
The most common type of damping encountered in beginning courses is that of
viscous damping; the damping force is proportional to the velocity. The damping
of the rst weight depends solely on its velocity and not the velocity of the second
weight, and vice versa. We add viscous damping to the model by adding the term

72

T. H. Fay and S. D. Graham

1 x_ 1 to the rst equation and 2 x_ 2 to the second equation (2.1). We assume that
the damping coecients 1 and 2 are small. The model becomes
m1 x1 1 x_ 1  k1 x1  k2 x1  x2
m2 x2 2 x_ 2  k2 x2  x1

2:16

To obtain an equation for the motion x1 which does not involve x2 , we solve the
rst equation (2.16) for x2 and substitute into the second equation (2.16) to obtain
4

EEE
m1 m2 x1 m1 1 m2 2 x
x1
1 m2 k1 k2 m1 m2 1 2

k1 2 k2 1 2 x_ 1 k1 k2 x1 0 2:17
In a similar manner, we may solve the second equation (2.16) for x1 and substitute
into the rst equation to obtain a fourth-order equation that involves only x2 . We
obtain
4

EEE
m1 m2 x2 m1 1 m2 2 x
x2
2 m2 k1 k2 m1 m2 1 2

k1 2 k2 1 2 x_ 2 k1 k2 x2 0

2:18

So once again we have the same linear dierential equation representing the
motion of both weights.
The roots of the characteristic equation are now more complicated to discuss in
general, but not surprisingly, we obtain damped oscillatory motion for both
weights under simple assumptions on the parameters.
Example 2.4. Assume m1 m2 1. Describe the motion for spring constants
k1 0:4 and k2 1:808, damping coecients 1 0:1 and 2 0:2, with initial
conditions x1 0; x_ 1 0; x2 0; x_ 2 0 1; 1=2; 2; 1=2.
The phase portraits for x1 and x2 are shown in the top row of gure 2.5; one
sees a regular pattern of motion with diminishing amplitude. Damped oscillatory
motion is evident in plots of the solutions shown in the middle row; and in the lefthand frame of the bottom row, we plot both x1 and x2 and observe nearly
synchronized motion. Finally, in the right-hand frame of the bottom row, we
plot x1 versus x2 which also shows damped oscillatory motion of both weights.
Student problem. Investigate the model when 1 0 and 2 > 0. What happens if 1 > 0 and 2 0?
Student problem. What are the conditions on the parameters of the model that
give rise to critically damped motion and over-critically damped motion, or do
these concepts not apply to this model?

3. Adding nonlinearity
If we assume that the restoring forces are nonlinear, which they most certainly
are for large vibrations, we can modify the model accordingly. Rather than
assuming that the restoring force is of the form kx (Hookes law), suppose we
assume the restoring force has the form kx x3 . Then our model becomes

Coupled spring equations

Figure 2.5.

73

Plots for Example 2.4.

m1 x1 1 x_ 1  k1 x1 1 x31  k2 x1  x2 2 x1  x2 3
m2 x2 2 x_ 2  k2 x2  x1 2 x2  x1 3

3:1

The range of motions for the nonlinear model is much more complicated than that
for the linear model. To get an idea of this range of motions for a single spring

74

T. H. Fay and S. D. Graham

model see [4]. Moreover, accuracy questions arise when solving these equations.
No numerical solver can be expected to remain accurate over long time intervals.
The accumulated local truncation error, algorithm error, roundo error, propagation error, etc., eventually force the numerical solution to be inaccurate. This is
discussed in some detail in the interesting paper by Knapp and Wagon [7], see also
[3] and [5].
Example 3.1. Assume m1 m2 1. Describe the motion for spring constants
k1 0:4 and k2 1:808, damping coecients 1 0 and 2 0, nonlinear
coecients
1 1=6
and
2 1=10,
with
initial
conditions
x1 0; x_ 1 0; x2 0; x_ 2 0 1; 0; 1=2; 0.
The motion in this example is quite nice. We have no damping, so the motions
are oscillatory and appear to be periodic (for more on detecting and describing
periodic solutions to nonlinear dierential equations see [6]). Phase plane trajectories are shown in the top row of gure 3.1 for 0 4 t 4 50; solutions are shown in
the middle row. A plot of x1 and x2 shows the motions that appear to be 1808 out of
phase, see the left-hand frame of the bottom row. A plot of x1 against x2 is given in
the right-hand frame.
Because of the nonlinearity, the model may exhibit sensitivity to initial
conditions.
Example 3.2. Assume m1 m2 1. Describe the motion for spring constants
k1 0:4 and k2 1:808, damping coecients 1 0 and 2 0, nonlinear coecients 1 1=6 and 2 1=10, with initial conditions x1 0; x_ 1 0; x2 0;
x_ 2 0 0:5; 1=2; 3:001; 5:9.
Phase plots and a plot of x1 versus x2 are shown in gures 3.2 and 3.3. Very
pleasing motions can be observed here. Next we change only the x1 0 value by 1/
10, and we obtain quite dierent motions.
Example 3.3. Assume m1 m2 1. Describe the motion for spring constants
k1 0:4 and k2 1:808, damping coecients 1 0 and 2 0, nonlinear coecients 1 1=6 and 2 1=10, with initial conditions x1 0; x_ 1 0; x2 0;
x_ 2 0 0:6; 1=2; 3:001; 5:9.
The phase plots for this example are shown in gure 3.4 for 0 4 t 4 200. A plot
of x1 versus x2 is shown in gure 3.5.
4. Adding forcing
It is a simple matter to add external forcing to the model. Indeed, we can drive
each weight dierently. Suppose we assume simple sinusoidal forcing of the form
F cos !t. Then the model becomes
m1 x1 1 x_ 1  k1 x1 1 x31  k2 x1  x2 2 x1  x2 3 F1 cos !1 t
m2 x2 2 x_ 2  k2 x2  x1 2 x2  x1 3 F2 cos !2 t

4:1

The range of motions for nonlinear forced models is quite vast. We can expect to
nd bounded and unbounded solutions (nonlinear resonance), periodic solutions
that share the period with the forcing (called harmonic solutions) and solutions
that are periodic of period a multiple of the driving period (called subharmonic

Coupled spring equations

Figure 3.1.

75

Plots for Example 3.1.

solutions), and steady state periodic solutions (limit cycles in the phase plane). The
conditions under which these motions occur are by no means easy to state. We
conclude with a simple forced example.
Example 4.1. Assume m1 m2 1. Describe the motion for spring constants
k1 2=5 and k2 1, damping coecients 1 1=10 and 2 1=5, nonlinear

76

T. H. Fay and S. D. Graham

Figure 3.2.

Figure 3.3.

Phase plots for Example 3.2.

x1 versus x2 , 0 4 t 4 200, for Example 3.2.

Coupled spring equations

Figure 3.4.

Figure 3.5.

Phase plots for Example 3.3.

x1 versus x2 , 0 4 t 4 200, for Example 3.3.

77

78

T. H. Fay and S. D. Graham

Figure 4.1.

Plots for Example 4.1

coecients 1 1=6 and 2 1=10, forcing amplitudes F1 1=3 and F2 1=5,

and forcing frequencies !1 1 and !2 3=5, with initial conditions
x1 0; x_ 1 0; x2 0; x_ 2 0 0:7; 0; 0:1; 0.
Because of the damping, we expect to see dierent behaviour for small values of
t and steady-state behaviour for large values of t. Thus we can expect to see a limit

Coupled spring equations

Figure 4.2.

79

Plot of x1 and x2 for Example 4.1.

cycle in the phase plane for both x1 and for x2 . The trajectories and limit cycles are
shown in the top and middle rows of gures 4.1. In the bottom row we show plots
of x1 and x2 for 0 4 t 4 150. In the left-hand frame of gure 4.2, we plot both x1
and x2 for 110 4 t 4 170 to show the steady-state solutions more clearly. We plot
x1 against x2 in the steady state in the right-hand frame.
5. Conclusions
We have developed a simple model for two coupled springs, examined both the
linear case and one possible form for the nonlinear case, and have included free
motion, damped motion, and forced motion examples. These examples produced
interesting solutions and are no more dicult for a student to produce than the
elementary examples commonly encountered in beginning courses. This model
reinforces much of the theory for linear equations and provides a nice elementary
modelling example that can be used in a computer laboratory component of a
beginning course or for an individual or a small-group undergraduate research
project.
The model has many features that permit the meaningful introduction of many
concepts including: accuracy of numerical algorithms, dependence on parameters
and initial conditions, phase and synchronization, periodicity, beats, limit cycles,
harmonic and subharmonic solutions. The use of a computer algebra system
permits almost eortless numerical explorations, graphical interpretations, and
motivation for analytical verications.
References
[1] Boyce, W. E., and DiPrima, R. C., 2001, Elementary Dierential Equations, 7th edition
(New York: John Wiley & Sons).
[2] Fay, T. H., 2000, Int. J. Math. Educ. Sci. Technol., 31, 733747.
[3] Fay, T. H., and Joubert, S. V., 1999, Math. Comput. Educ., 33, 6777.
[4] Fay, T. H., and Joubert, S. V., 1999, Int. J. Math. Educ. Sci. Technol., 30, 889902.
[5] Fay, T. H., and Joubert, S. V., 2000, Math. Mag., 73, 393396.
[6] Fay, T. H., and Joubert, S. V., 2000, Int. J. Math. Educ. Sci. Technol., 31, 825838.
[7] Knapp, R., and Wagon, S., CODEE (winter 1996), pp. 913.