An improved calculation of the mass for the resonant spring pendulum

Joseph Christensena)
Department of Physics, McMurry University, Abilene, Texas 79697
Received 28 January 2002; accepted 23 January 2004
When the appropriate mass is used to oscillate a spring, the vertical oscillations couple to a pendular
swing. Previous calculations of various aspects of this resonance assumed a massless spring as a
simple pendulum. This paper improves the estimate of the mass necessary to induce this resonance
by describing a massive spring as a physical pendulum and obtains an expression for the mass in
terms of the spring constant and various lengths associated with the spring. Several approximations
will be considered to simplify the complete expression. Comparisons of the predictive power of
these expressions are made for various values of the spring constants. An Appendix discusses the
assumption of uniform coil density of a hanging massive spring. 2004 American Association of Physics
Teachers.
DOI: 10.1119/1.1677269

I. RESONANT SPRING PENDULUM For simple pendula, Im 2 , and for small angles, sin .
Equation 2 also is solved by a trigonometric function
Several authors1 4 have discussed the resonance in which (t) 0 cos(pt), where p mg/I g/.
a vertically oscillating spring spontaneously oscillates be- When the autoparametric resonance occurs, it is observed
tween spring-bouncing and pendular-swinging. These papers that the period of the pendular motion, T p , is equal to twice
assume massless springs as simple pendula to solve the equa-
the spring-like period, T s :
tions of motion. The goal of this paper is to introduce mas-
sive springs as physical pendula in order to more accurately 2T s T p , 3
predict the mass that leads to this resonance. For details
about the coupling of the modes, the resonance, the paramet- where, for a massless spring and a simple pendulum,
ric instability, and the period of oscillation between these
modes, the reader should refer to Refs. 1 4 and references
therein. In brief, Olsson1 cautions that because this system is
T s 2 m
k
, T p 2
g
. 4

not linear, superposition is not applicable and, therefore, the As mentioned in Ref. 1, Eq. 3 requires that a spring with
general motion cannot be expressed as a combination of nor- the unstretched length s must be stretched to a length
mal modes. He also notes that the ... resonance effect is 4mg/k for the resonance to occur. Because a mass m will
more correctly known as an autoparametric resonance... stretch a spring to the length s mg/k, this resonance
due to the lack of an explicit time dependence in the differ- occurs when
ential equation. More importantly, he gives a more detailed
description of the equations of motion that describe this reso- k s
nance. Lai2 solves Olssons equations and discusses why the m . 5
3g
conversion between the oscillation modes does not merely
occur, but recurs.5 Equation 5 assumes a massless spring and gives only
A spring with length will oscillate vertically according to roughly approximate values for the various springs that we
the equation of motion: will consider. The detailed changes due to considering a mas-
sive spring as a physical pendulum should allow us to more
mz k z mg, 1 closely predict the experimental value of the required mass.
The main goal of this paper is to give a more accurate
where z is the vertical position with z0 at the top of the calculation of the required mass, while minimizing the com-
unstretched spring and positive downward, z is the second plexity of the final equation. A secondary goal is to help
derivative in time, m is the mass attached to the spring, k is students learn about approximations as well as provide an
the spring constant, and g is the magnitude of the local gravi- opportunity to numerically solve a cubic equation that is ap-
tational field. Equation 1 is solved by a trigonometric func- plicable to an observable phenomenon. This resonance phe-
nomenon makes a good problem for undergraduates for a
tion plus a constant stretch z(t)A cos(st)(mg/k). For
variety of reasons. The resonance is not necessarily easy to
the initial position z 0 , the spring oscillates with amplitude produce in an unfamiliar spring by trial and error because it
Az 0 mg/k about the equilibrium point mg/k. For is difficult to see how close the system is to resonance. Pre-
a massless spring, the angular frequency is s k/m. dicting the resonance mass forces the students to reconsider
Whereas Eq. 1 describes the forces acting on a spring, the assumptions massless springs and simple pendula in the
the pendular motion is due to the torques. A pendulum with equations that have been derived in class and probably used
length oscillates according to the equation of motion: in the laboratory.
In Sec. II, I will review how previous authors have intro-
I mg sin 0. 2 duced the spring mass into the vertical oscillations and com-

818 Am. J. Phys. 72 6, June 2004 http://aapt.org/ajp 2004 American Association of Physics Teachers 818
bine that with the physical pendulum to derive a cubic equa-
tion for the mass that induces the resonance. In Sec. III, I
consider three approximations to this cubic equation in order
0

d

0
1

0
mg z m s g
k

k
dz 7a

to find an expression that is more convenient. In Sec. IV, I

mg m s g
will discuss the experimental apparatus and the theoretical 0 , 7b
predictions, and compare the predictions to the masses that k 2k
experimentally induce a resonance in the springs. The con-
clusion is in Sec. V. An appendix gives more details about where z is measured from the top of the spring and (
the major underlying assumption. z)m s / is the fraction of the spring below the point z.
Equation 6 or 7 shows a correction to the static length in
which mmm s /2.
Note first that the term (Ni)m i g/k i in Eq. 6a, or
II. INTRODUCING THE SPRING MASS equivalently (z)m s g/k in Eq. 7a, implies that the coils
stretch more at the top, where the spring supports more
In this section, I will introduce the mass of the spring into weight. This difference means that the center of mass of the
the vertical spring oscillations and then into the pendular spring will be lower than half-way down the spring, which
swing. When we combine these expressions to predict the will be of interest when we discuss the pendular motion.
resonance, we will find that the distribution of mass is rela- Second, Mak7 started from Ref. 6 and derived a piecewise
tively important. function that generalizes the static correction to include the
effect of a constant force necessary to initially extend a
spring, thereby providing a discrete version of Eq. 20 in
A. Massive springs Ref. 8 and footnote 16 in Ref. 9. Finally, for convenience, I
The consideration of the correction to the spring oscilla- will consider the freely hanging length,
tions due to including the mass of a spring has led to many m sg
papers.6 14 There is a commonly used correction that em- s 0 , 8
ploys an approximation that should be reasonably valid for 2k
all but the softest springs. For the convenience of the reader,
to be a measured quantity rather than measuring 0 and add-
a brief summary will be repeated here with references to the
literature for details. ing m s g/2k.
There are two situations in which the mass of the spring The dynamic correction due to a bouncing, finite-mass
should be included. First, the spring mass shifts the equilib- spring affects the period of oscillation. This correction can be
rium position of a vertically hanging spring because more derived in a deceptively easy manner by using the kinetic11
mass must be supported by the upper portions of the spring. and potential energy in Lagrangian mechanics if one assumes
uniform stretch, or equivalently, an uniform mass density or
This static correction will modify T p , the pendular period,
uniform coil density. This derivation is deceptively easy be-
but will not affect T s , the period of the spring-like bouncing cause a more careful and complicated treatment gives a tran-
motion. Second, the inertia of a massive spring produces a scendental equation discussed in the following that reduces
dynamic correction to T s . The corrections due to these ef- to the Lagrangian result in the appropriate limit. Given a
fects are not the same and are more prominent in a softer spring of length 0 and mass m s and an uniform linear mass
spring.
density m s /, we can find the velocity v (z) of a differ-
To see the static effect, consider a spring with mass m s ,
ential portion of the spring at a distance z measured from the
spring constant k, and unstretched length 0 . Imagine it as a
top of the coils: v (z) v m z/, where v m is the velocity of
series of N springs, labeled from top to bottom as 1 through the added mass and is given by Eq. 6. The kinetic energy
N, each with mass m i m s /N, spring constant k i Nk, and of the spring between z and zdz is
unstretched length i0 0 /N. The ith spring supports the
(Ni) springs below it as well as the mass at the end and 1 1 vm
2 2
z
therefore each is stretched to length i i0 mg/k i (N dK v z dz
2
2 dz. 9
2 2
i)m i g/k i . The total length of the spring series is

N1 N1 The total kinetic energy of the spring is the integral of Eq.
mg m ig 9 plus that of the mass hanging from the spring:
i i0 Ni 6a

ki ki

i0 i0
1 1 vm
2 2
z 1 ms 2

N1 2
K m v m dz m vm . 10
1 mg m sg 2 0 2 2 2 3
0 Ni
i0 N k Nk
Equation 10 shows a correction to the dynamic length in
0
mg
k
N 2
N N1 m s g
2 N 2k which mmm s /3. When hung, the bottom of the un-
loaded spring sits at z 0 m s g/2k. The spring potential
energy plus the gravitational potential energy is
mg m s g
0 . 6b 1 z
k 2k U z k 0 z 2 mgzm s g . 11
2 2
I have used the fact that N1
iN(N1)/2 and taken the
large N limit so that 1/N0. In the continuum limit, Eq. 6 We have again assumed uniform mass density to write the m s
can be written as6 term.

819 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 819
 The kinetic and potential energies, Eqs. 10 and 11, can where D is given by the solution to their version of Eq. 15
be used in the Lagrangian formulation to derive the new with the limiting behavior
equation of motion. The generalization of Eq. 1 is
2

m
ms
3
z kzk 0 m
ms g
2 k
, 12
lim D3,
m s /m0
lim
m s /m
D
4
.

The dynamical correction is discussed further in Refs. 12,

17

with the solution 13, 14, and 9. Three methods for solving Eq. 15 are given
z t z 0 cos s t , 13 in Weinstocks later paper,12 in which it was shown that the
lowest frequency mode is sufficient to describe the small
where s (mm s /3)/k and 0 (mm s /2)g/k as in m s /m limit. The uniform density approximation was used to
Eqs. 6 and 7. A shift of the coordinate system by show that the m s /3 dynamic correction is reasonable up to
0 (mm s /2)g/k makes Eq. 13 look simpler, but hides m s m. McDonald13 showed that a closed form for D exists
the shift in the equilibrium position. It follows that the pe- for the conical spring, reproduced the result for D in Ref. 6,
riod of oscillation becomes and compared m s /D to the equivalent correction term for

conical springs. Bowen14 considered an unloaded slinky with
ms general initial conditions for the nonfixed end, emphasizing
m
3 the 2 /4 result. Cushing9 showed explicitly that the small
T s 2 . 14
k m s /m case necessarily transitions to the large m s /m case so
that the lowest normal mode is always the dominant one.
The m s /3 in Eq. 14 is the dynamic correction and may be It should be emphasized that adding a term m s /3 as in Eqs.
contrasted with the m s /2 equilibrium shift of Eq. 7, the 10 and 14 is based on the assumption of uniform stretch
static correction. Students who have not seen Lagrangians (m s /l is a uniform coil density. This assumption be-
can infer the dynamic correction to the period from Eq. 10, comes less applicable for softer springs and depends on the
because it is reasonable to expect the period to see the same
ratio m s /m, as seen in Eq. 17. In most cases all but the
mass as the kinetic energy, (mm s /3), rather than as the
smaller k springs cited in Refs. 7 and 9, it seems sufficient
equilibrium shift, (mm s /2).
to use m s /3 as was done here. The relevant results are dis-
Several papers6 10,1214 have been published on these cor-
cussed in Appendix B.
rections due to the effect of the spring mass. Those
authors6 10 that consider both corrections make a point of
distinguishing the static correction from the dynamic correc- B. Summary of the physical pendulum
tion. As alluded to earlier, a careful treatment of the dynamic
spring in terms of the position of the spring shows that in- As stated following Eq. 2, p mg/I. For a physical
cluding the mass of the spring leads to a transcendental equa- pendulum, we cannot use the simplification Im 2 , which
tion for the period. gives T p in Eq. 4. Rather, we must express the period more
In an early paper, Weinstock10 considered a mass in uni- generally as
form circular motion which stretches the spring the same
amount as a hanging mass and introduced an m s /3 correction
to the mass dependence of the frequency. He also considered
T p 2 I
mg
, 18
the mass oscillating about its equilibrium radius and found
that the angular frequency satisfies the transcendental equa- where the denominator is due to the torques that drive the
tion, which in my notation reads oscillation. To see how this general expression changes with

different assumptions, consider three cases. Case 1: A simple
2 m s /k 2 m s /k ms pendulum with a massless string has Im bob 2 . If we use
tan . 15 this value of I and that the torque on the bob is m bobg, we
Ts Ts m
have T2 /g. Case 2: A solid rod pendulum has I
Equation 15 reduces to Eq. 14, with dynamic correction 13 m rod 2 and, because the torque acts at the center of mass
mm s /3, in the small m s /m limit. of the rod, the denominator of Eq. 18 becomes m rodg(/2).
Heard and Newby8 and Cushing9 also consider both the The period of the rod is then
static and dynamic corrections. For the dynamic spring, they
solved the appropriate differential equations. Reference 8
went a step further by considering a vertical soft spring to
find that in the m s /m limit, T s 4/(2n1) m s /k,
T2 1
3 m 2
mg /2
2 2
3g
. 19

where n is any positive integer corresponding to the peri- Case 3: A massive support with a massive bob, like the case
odic nature of trigonometric functions. These periods corre- of interest, combines these two cases by separately adding
spond to the eigenvalues that solve the appropriate differen- the moments of inertia in the numerator, Im bob 2
tial equation given in their paper. These periods also solve 13 m rod 2 , and the torques in the denominator, m bobg
Eq. 15 in the large m s /m limit. Consequently, as discussed m rodg(/2), giving a period of

by Galloni and Kohen,6 the dynamical correction in Eq. 14
is different in the large m s /m limit than in the small m s /m m bob 31 m rod
limit. Equation 14 is given in Ref. 6 as T2 . 20
m bob 21 m rod g

T2 1
k
m
ms
D
, 16 Note that Eq. 20 reduces to either the massless string or the
solid rod pendulum when the appropriate mass is set to zero.

820 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 820
It is an interesting coincidence that the moment of inertia can then be found and is due to both the spring and the added
sees the same fraction of m s as the kinetic energy discussed mass. It also is a combination of rotations about the center of
in Sec. II A and that the torque sees the same fraction of m s mass, I c.m. , and the effect of the parallel axis theorem I pat :
as the stretching of the spring. However, the mathematical
II s,c.m.I s,patI m,c.m.I m,pat . 21
expression for the moment of inertia of the spring pendulum
will be further complicated by the length dependence as dis- The first two terms can be written as

cussed in Sec. II C. A careful treatment of the length depen- 2
dence in Eq. 18, specifically in terms of the mass distribu- 1 mg
I s,c.m.I s,pat m s s
tion, shows that the length does not cancel as simply as it did 12 k

in Eq. 20. 2
mg
s
C. Massive spring as a physical pendulum k
m s 1 . 22
2
Now that we have introduced the mass into the bouncing
spring and swinging pendulum, we can use Eqs. 14 and The first term is the moment of inertia of a slender rod the
18 to derive an expression for the mass that produces the spring rotating about its center of mass. If 1 0, the two
desired resonance between bouncing and swinging. As dis- terms in Eq. 22 would combine to give 31 m s ( s mg/k) 2 ,
cussed in Sec. II B, the distribution of mass in the system the moment of inertia of a slender rod about one end. The
must be understood in order to express the period of the definition,
pendular motion. To account for the geometry of the swing-
ing spring system, Fig. 1a shows the various relevant 1
x 1 , 23
lengths: the length s of the spring with no added mass, the s mg/k
amount of stretch mg/k due to the added mass, the distance
leads to the expression
1 from the top of the coils of the spring to the pivot point,
the distance 2 from the bottom of the coils of the spring to
the top of the added mass, and the distance m from the top I s,c.m.I s,patm s s mg
k
2
1
3
x 1 x 21 . 24
of the added mass to the center of mass. The initial length of
the hanging spring, s , is stretched only due to its own The third term in Eq. 21 is
weight as given by Eq. 8, and 2 and m depend on the I m,c.m. 121 m h m 2 121 m 2 m 2 31 m m
2
. 25
amount of mass added. Figure 1b distinguishes the theoreti-
cally convenient lengths, 2 and m , from the experimen- The length 2 m is used because m is half the height of the
tally convenient heights: the distance h s from the coils to the added mass, h m . The fourth term in Eq. 21 describes the
hook, the height h h of the hanger, and the height h m of the mass at the end of all of the lengths:
masses. These lengths are related according to 2 h s h h
h m and m h m /2. For simplicity, I will assume that 2
and m are independent of the amount of added mass. In-

I m,patm 1 s
mg
k
2 m 2
. 26

deed, one might expect that m is significantly less than s so Finally, the denominator of Eq. 18 can be expressed as
that the estimated value of m is irrelevant as long as it is
close enough to its true value. As a rough approximation, the
M gmg 1 s
mg
k
2 m

appropriate 2 and m can be estimated by measuring h m for
the mass given by Eq. 5. I will discuss this issue in Sec. IV.
The expressions for the period of a swinging coil can be
simplified by assuming that, when swinging, the spring is a m s g 1
s
mg
k
. 27
2
rigid rod with mass m s , length ( s mg/k), and center of
mass 1 ( s mg/k)/2 . A more accurate, but much less If we substitute Eqs. 2427 into Eq. 18 and combine
convenient expression for the center of mass is given in Ap- this result with T s from Eq. 14 as expressed by the square
pendix B. The moment of inertia in the numerator of Eq. 18 of Eq. 3, we find

m
ms
3
m s s
mg
k
2
1
3 1
x 1 x 21 m m
3
2

m s
mg
k
1 2 m 2

4 . 28
k mg 1 mg
mg s 1 2 m m s g s 2 1
k 2 k

The complicated expression for m in Eq. 28 explains why in Sec. III D. For compactness, we have written only one of
it is not usually considered in this form. Note that if 1 the contributions in Eq. 28 in terms of x 1 . If we gather
2 m 0, then the substitution of Eqs. 2427 into terms in powers of m without using the x 1 notation, we ob-
Eq. 18 for T p reproduces Eq. 20. This case will be useful tain

821 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 821
 the various lengths appropriate to the spring, the coefficients
can be included in a root-finding program to solve for what
should be the best prediction of the mass m that will cause
the spring to excite the autoparametric resonance.
We have made three assumptions to obtain Eq. 29. We
assumed that the spring is a rigid rod for calculating the
moment of inertia I s , that 2 and m are independent of the
mass, and that the spring has uniform density giving both
(mm s /3) and the expression for the center of mass in Eqs.
22 and 27. The first assumption is reasonable because
when the appropriate mass produces the autoparametric reso-
nance between swinging and bouncing, the few oscillations
in the purely swing-mode do not have any visible bounce,
that is, the swinging spring behaves as a rigid rod. The sec-
ond assumption will be reasonable if and only if h m can be
estimated to sufficient precision. The details of generalizing
the third assumption are relegated to Appendix B.

III. APPROXIMATIONS TO THE CUBIC

Solving the cubic polynomial in Eq. 29 for the added

mass m is straightforward using standard numerical root-
finding techniques. However, reasonable approximations
should simplify Eq. 28 considerably. In this section, I will
introduce three seemingly reasonable approximations to find
a more useful approximation to Eq. 28.
We will consider the case 1 2 m 0 in Sec. III A
and then m 0 in Sec. III B. We build on these approxima-
tions in Sec. III C where we keep all three lengths and Sec.
III D where we build a mathematically convenient approxi-
mation.

A. The long-spring approximation

When x 1 is set to zero, Eq. 28 remains cubic in m due to

its dependence on 2 and m . In the long-spring approxima-
tion, we set 1 and x 1 ) 2 m 0, and Eq. 28 becomes
Fig. 1. The hanging spring can be separated into a variety of distinct mea-
surements. a The theoretically useful quantities; s is the length of the ms

hanging spring with no mass attached but stretched due to its own mass; m m
is the distance to the center of mass for the mass added. The total length is 3 mg 1 mg
denoted as 1 s mg/k 2 m . b A close up of the added mass to 4 mg s m sg s
k k 2 k
relate the theoretically useful quantities to the experimentally easy-to-
measure quantities (h h ). 1
m s s
3
mg
k
2
m s
mg
k
2
. 30

We can cancel ( s mg/k) and (mm s /3), giving an ex-
k pression linear in m,
03m 3m s 2 s 1 2 m m 2
3
g
k s 2m s
2

m s2
3
k 8
g 3
13

s 1 2 m m s
3 m
3g

3
,

which is a minor change from the massless spring simple

31

1 k m
3 g 2

k
g
2

s 1 2 m 2 m
pendulum of Eq. 5.

2 2k
m 2 m
3 sg s
B. Weakening the long-spring approximation

m s
k
g 2
1 2
s 1 21
3 s . 29
Let us take m 0 and 1 2 0, because 1 and 2 are
always added to s , whereas m is not added to a larger
2
quantity in the 31 m m term. With this choice, Eq. 28 be-
Given the mass m s of the spring, the spring constant k, and comes

822 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 822
 m
ms
3 4 2 2
32x m x m
3
m 4
3 x m m s
3

4
k 4 2 k s 2 k s m s

2 2 12x m x m m m s2 0. 34
1 mg 1 mg 3 g 3 3g
m s s m m
2
m s m
3 k 3 k Note that Eq. 34 reduces to Eq. 31 in the limit x m 0.

.
mg 1 mg
mg s m m s g s
k 2 k
32
Because Eq. 32 is still cubic in m, I will assume that m C. Collecting small terms
s and define x m as
To this point, the theoretical results either are poor predic-
m m 3 m tors of the desired mass m, such as Eqs. 5 and 31, or are
x m , 33 higher-order expressions, such as Eqs. 29 and 34. Equa-
mg k s g 4 s
s s tion 31 will give poor predictions for m because the pre-
k 3g k
dictions of Eq. 5 are, in most cases, smaller than the nec-
where m has been replaced by the massless-spring, simple- essary physical mass.
pendulum approximation of Eq. 5. After we divide the nu- The terms in Eq. 28 do not readily cancel because of the
merator and denominator of the right-hand side of Eq. 32 different combinations of 1 , 2 , and m . Let us first re-
by ( s mg/k) and collect terms in powers of m, we obtain place each of these by ave , the average of these three quan-
a quadratic equation for m that can be solved given m , m s , tities. This replacement reduces Eq. 28, without the x 1 no-
k, and s , tation, to

m
ms
3
1
m
3 s s
mg
k 2
mg
3 ave s
k
3 ave
2

1
m ave
3
2

m s
mg
k
3 ave
2


4 . 35
k mg 1 mg
mg s 3 ave m s g s 2 ave
k 2 k
If we rewrite Eq. 35 in terms of L( s mg/k 3 ave) and complete the square in the m s term, we obtain

m
ms
3 m
3
ms 2 ms
L
3
3 ave s
mg
k
6 ave
2
1
m ave
3
2


4 . 36
k ms 1
m gL m s g ave
2 2

We define the hopefully small quantity

ave
4
g
k
m
ms
3 m
ms
2
y
L
, 37
m ms
3
L
divide the numerator and denominator by L, and find

ms

y 2 m
3
m s m sg
k
m s L ave
m ave
3
. 39

m
3
4 If we divide through by (mm s /3), solve for m, and collect
k
the y terms to O(y), we find

m
ms
3 Lm s y mg
k
s
1
2 ave my ave
3 k 2

. m 3 ave m s O y . 40
ms 1 3g s 3
m g m s gy
2 2
38 If we ignore terms of order y, Eq. 40 gives a better esti-
mate of the mass. Alternatively, given the experimental value
We then multiply both sides by (m m s /2)g 21 m s gy, col- for the mass, we can estimate the magnitude of the O(y)
lect terms of order y on the right, and obtain terms and evaluate the approximation l aveL.

823 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 823
 Table I. The physical dimensions of the springs used. For all but the two stiffest springs, a mass-hanger with
height h h 8.000(4) cm was used. The second stiffest spring used a hanger with h h 14.806(4) cm. The
stiffest spring used one with h h 28.2(1) cm. The other lengths can be found from 2 h s h h h m and m
h m /2. The height of the added mass h m is given for the mass predicted by the equation indicated. All lengths
are in centimeters.

Eq. 5 Eq. 42 Experimental

k N/m m s kg s 1 hs hm hm hm

3.0591 0.67792 293.31 2.005 2.005 3.4194 N/A N/A

6.6963 0.33902 121.41 0.00 2.005 3.2724 1.3464 1.2764
25.164 0.08292 26.61 1.235 1.325 2.8464 3.3164 3.4564
28.635 0.08242 27.61 1.405 0.955 3.2924 3.7424 3.8094
36.268 0.08842 27.11 1.305 1.305 4.1094 4.5354 4.9124
40.91 0.08462 27.11 1.205 1.205 4.3154 4.9924 5.5184
51.42 0.07862 27.21 1.505 1.305 5.0124 7.3464 6.8494
10022 0.06292 6.71 1.11 5.005 6.355 20.455 23.845

D. A mathematically convenient approximation down, I was able to investigate the results for k
We have seen that the spring-like oscillation is affected by 3.059 N/m and k6.696 N/m. Finally, a spring-scale was
the spring mass by a factor of (mm s /3). We also saw that removed from its casing to test a high k spring.
a rod-mass pendulum has the moment of inertia I(m The values of k in Table I are not needed because k ap-
m s /3) 2 and torque (m m s /2)g as if each saw the pears in the ratio k/g, which can be measured directly. A
additional mass at the bottom of the spring. Although physi- mass hanging in equilibrium from a spring will satisfy kz
cally inappropriate, if we simply extend Eq. 20 by includ- mg; therefore, the slope of the displacement versus the
ing the lengths 1 , 2 , and m and substitute this result into added mass the independent variable will give g/k. To ob-
Eq. 3 as before, we find tain the values of the spring constant, we assumed g

m
ms
3 ms
3
m
1 s
mg
k
2 m
2
9.80(1) m/s2 . These values are listed as k in Table I. Ap-
pendix A discusses some of the relevant details of this mea-

4 . 41 surement.
k ms mg While addressing the uniform density issues raised at the
m g 1 s 2 m
2 k end of Sec. II A, Appendix B discusses an alternative, direct
Equation 41 should be compared to Eq. 28 to see how it measurement of k. Because k can be measured directly, it is
differs from a more physically accurate treatment. Equation possible15 to combine the measurements of g/k and k to
41 is, however, mathematically convenient and easily re- calculate the local gravitational field.
duces to Table I also lists the physical lengths of the equipment.
The length s was measured with a meter stick. The lengths
k 2
m 1 2 m m s , 42 1 and h s were measured with a vernier caliper, which has a
3g s 3 precision of 0.002 cm. To account for slight awkwardness in
which is Eq. 40 without the O(y) term. The physical inter- measuring as well as for the possibility that the support ring
pretation is that if 1 , 2 , and m are sufficiently less than might stretch when weight was added, this uncertainty was
s , they can be handled conveniently as in Eq. 40, but even increased to 0.05 cm. The value of h h is not listed because
small values are not so small that they can be set to zero as in the same mass hanger was used for five of the seven springs.
Eq. 31. This hanger had h h 8.000(4) cm. The second stiffest
spring used a mass hanger with h h 14.806(4) cm. These
IV. ANALYSIS OF THE APPROXIMATIONS two were measured with a caliper with the error doubled to
account for slight awkwardness. The stiffest-spring mass-
After describing the experimental apparatus and the uncer- hanger was measured with a meter stick and had h h
tainty, we will compare the predictions of the various ap- 28.2(1) cm.
proximations to the mass that actually induces the resonance. As mentioned in Sec. II B and shown in Fig. 1b, the
A. The experiment other needed lengths can be found from 2 h s h h h m and
m h m /2. To find an appropriate h m , we estimated a value
Our goal is to find a useful expression for the mass that
for the added mass from Eq. 5 and, using standard masses
induces the autoparametric resonance between the vertical
oscillations and the pendular swinging of a spring. To ana- and a vernier caliper, measured h m . If Eq. 42 with this
lyze the accuracy of the predictions, seven springs were value of h m predicts a mass with a height that is significantly
used. The spring constants, masses, and various lengths are different than this value of h m , we measured the h m for the
given in Table I. The five springs with spring constants rang- predicted mass and recalculated Eq. 42. One or two itera-
ing from 25 to 52 N/m were selected from a standard under- tions were sufficient for consistent results. The last three col-
graduate laboratory set. A wave-demonstration spring was umns of Table I indicate how much h m can vary between the
used to test the predictions on a low k spring. By first using predictions from these equations. So, even if we are careful
the full length of the spring and then clamping it half-way to measure the lengths very precisely, there is an inherent

824 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 824
Table II. Comparison of the predicted mass values to the experimental masses. Except for Eq. 5, the calculation for the predicted mass requires a value for
the height of the mass being calculated, h m . As a rough estimate, Eq. 5 is used to estimate h m in the other equations. To improve the prediction, the result
of Eq. 42 is then used to re-estimate h m . This estimation process was self-consistent within three iterations. The two rightmost columns compare the rms
average difference i1
N
(m i m expt)2/N between the predicted and experimental masses. The second-to-last column is averaged over the seven stiffest
springs. The last column averages over six springs by excluding the softest and stiffest springs. All units are kilograms. Equations 29 and 42 are
significantly better predictors of the mass that experimentally excites the spring-pendulum resonance.

Mass prediction for eight springs, sorted by spring constant, k N/m rms difference
Eq. used to Eq. used to
predict m estimate h m 3.0591 6.6963 25.164 28.635 36.268 40.91 51.42 10022 all k k100

5 0.30523 0.27654 0.2281 0.2691 0.3341 0.3772 0.4763 2.293 3.217 0.123
29 5 0.1124 0.0647 0.257 0.306 0.389 0.438 0.692 9.8 0.375 0.032
42 5 0.1361(6) 0.0701 0.2503 0.2994 0.3815 0.4305 0.6847 9.21 0.605 0.034
29 42 0.1144 0.066 0.255 0.303 0.386 0.434 0.673 10.8 0.027 0.029
42 42 0.1339(5) 0.0721 0.2483 0.2964 0.3785 0.4255 0.6637 10.51 0.118 0.032
34 42 0.1460 0.051 0.185 0.231 0.302 0.354 0.488 9.64 0.449 0.102
Experimental range N/A 0.0655 0.2455 0.3188 0.42010 0.48010 0.63515 10.82

systematic uncertainty in that we do not actually know the diction. This can be done by checking the self-consistency
mass for which we should be measuring the height, h m . mentioned at the end of Sec. IV A.
To be explicit, the 4 term appears in Eq. 44 as the
uncertainty in s 1 2 m ; however, these uncertain-
B. Uncertainty and precision
ties are not actually all the same. Some terms are measured
Table II gives the predictions from the various approxima- with a meter stick precision of 0.05 cm and some with a
tions as well as the experimental values that produce the caliper precision of 0.002 cm. For awkward measurements,
resonance. The experimental masses were found by trial and these precisions were increased. Further, 2 h s h h h m
error starting from the predicted values. The central value is and each of these measurements increases the uncertainty.
that which subjectively seemed to give the most pronounced With this in mind, 4 is actually 0.1 cm from s )
effect. The error bars on the experimental mass are the extent 0.05 cm from 1 ) 0.05 cm from h s in 2 ) 0.004 cm
to which I could adjust the mass and see a pendular swing
from h h in 2 ) 0.004 cm from h m in 2 ) 0.002 cm
that retained a small amount of bouncing. Including masses
that produced a moderate swing without significantly dimin- from m h m /2) 0.210 cm. If the added mass or the
ishing the vertical oscillation would typically double or triple hanger is measured with a ruler as for the stiffer springs that
the listed uncertainty. require more applied mass, 4 becomes 0.256 cm.
Table II also estimates the precision of some predicted Regarding the systematic uncertainty, we see from Table I
values. Because Eq. 5 is a product, the relative uncertain- that h m can be off by as much as 2 cm if the mass is not
ties can be added in quadrature: predicted accurately. If is taken to be 0.5 cm so that the

m,5
k 0
3g

k
k
2


0
2

g
g
2
. 43
4 l term matches this 2 cm uncertainty, then m /m5%.
For the 51 N/m spring which requires an added mass of
0.635 kg, this 5% relative uncertainty, m /m, implies an
The precision of Eqs. 29 and 34 could be found by a uncertainty of m 0.032 kg. Unfortunately, Table II shows
Monte Carlo analysis, but was not due to the consistency that I only obtained a good resonance within 0.015 kg of
between Eqs. 29 and 42 and the measurements. The de- 0.635 kg. This variation indicates that it is important to mini-
termination of the precision of Eq. 42 is only slightly more mize the systematic uncertainty by verifying the consistency
complicated than Eq. 43 due to the additional terms: between the mass used to estimate h m and the predicted
k 0 3 ave mass. However, this may not be of dire necessity when one
m,42 considers that with 0.5 cm, the prediction of Eq. 29 is
3g
still within 2 m of the experimental value.

k
k
2

4
0 3 ave
2


g
g
2
2
ms .
3
The uncertainties listed for the masses in Table II reflect
the roughly 1% precision of the length measurements. That
these uncertainties do not overlap with the experimental val-
44
ues, especially using Eq. 29, suggests that there is an addi-
The relative uncertainty in 4 /( 0 3 ave) dominates the tional source of uncertainty. This discrepancy will be consid-
right-hand side of Eq. 44. The relative uncertainty of k and ered in Sec. IV C where we will discuss the underlying
g are below 1% and usually much less. The relative uncer- assumptions.
tainties / 0 and 4 /( 0 3 ave) are such that for every The precision quoted for the predictions in Table II indi-
millimeter that is increased, m /m increases by about cates only how much the result will vary due to the measure-
1%. This important point cannot be emphasized enough to ment uncertainties. It does not indicate how close the predic-
those who wish to utilize these equations. If the lengths are tion is to the experimental value, because it does not include
not measured carefully, then the precision of the predictions the systematic uncertainties such as the 2 cm variation in the
is reduced. Further, we must control the systematic uncer- prediction of h m . In addition, no attempt was made to indi-
tainty of using a precise measurement of an inaccurate pre- cate the size of the terms dropped in the approximations

825 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 825
because doing so would only indicate how far the various tions that the spring is a rigid rod and has uniform density.
approximations are from Eq. 29, not how far the prediction Equation 42 uses only these assumptions and the inclusion
is from the experimental value. of 1 , 2 , and m and gives predictions that are comparable
To estimate the quality of the predictive value of each to Eq. 29 for the five central springs. Equation 42 is
expression, the rightmost columns of Table II show the root- within about 3% of the experimental value for the very stiff
mean-square difference between each predicted mass and the spring, where one might expect these assumptions to be
experimental mass averaged over the various springs: more reasonable, and is about 11% off for the softest spring,

N where these assumptions are more suspect. Equation 29 is

1

N eq1
m eqm expt 2 . 45 close in both cases. As mentioned, the rigid rod assumption
should be reasonable because the swing mode at the reso-
nance mass does not have a discernible bounce. The uniform
Two values are given for each equation. The second-to-last density assumption, on the other hand, might warrant further
column is the rms error including all springs. The last col- investigation. See Appendix B for further details.
umn is the rms error not including the stiffest spring. Be-
cause the stiffest spring involves masses that are over an
order of magnitude larger, these differences can significantly
overwhelm the others. On average, Eq. 29 predicts the ex- V. CONCLUSIONS
perimental value to within 0.027 kg, even when the stiffest
spring is included. The purpose of this paper is to improve the prediction for
the mass, not to solve the differential equations of motion
C. The results that would generalize the analysis of Refs. 1 and 2 describing
the motion of the oscillating mass. Although the expression
As expected, Table II shows that Eqs. 29, 34, and 42 that predicts the mass for a massless spring as a simple pen-
are significant improvements over the assumption of a mass- dulum, Eq. 5, is easy to derive, it is possible to improve the
less spring in Eq. 5. For the smallest k spring in Table II, prediction by extending the derivation to describe the mas-
the negative mass values are consistent with not finding an sive spring as a physical pendulum. This extension gives Eq.
experimental value. Although the prediction of Eq. 5 for 29, a cubic polynomial for the mass that will induce the
this spring is 0.305 kg, the measured period of swinging and spring-pendulum resonance. A rough approximation also was
of bouncing indicates that T p cannot equal 2T s as needed for developed that gives a much simpler expression, Eq. 42,
the resonance. In fact, the massless spring approximation, and comparable results. It is clear from Table II that both
Eq. 5, only has a chance of being coincidentally correct for Eqs. 42 and 29 are much better predictors of the mass
k20 N/m. For k above this value, Eq. 5 predicts a mass than Eq. 5. Equations 29 and 42 assume that the swing-
that is too small and becomes worse for larger values of k, in ing spring is a rigid rod with uniform density. Because in full
spite of the expectation that this approximation might im- swing, it does not bounce, the rigid rod assumption is prob-
ably reasonable, at least for springs that are not too long.
prove for stiff springs. In fact, as k increases, the need to
However, the uniform mass assumption is somewhat suspect
include 1 , 2 , and m becomes more relevant, presumably
see Appendix B. The lengths 2 and m shown in Fig. 1
because a stiffer spring requires more mass to oscillate mak-
also play a significant role in the accuracy of the predictions.
ing at least m harder to ignore and the inclusion of these The precision of the predictions in Table II are due primarily
terms is a better measure of the pendular length, which con- to the uncertainty in the lengths.
trols the pendular period. Because the long-spring approxi-
mation, Eq. 31 ( 1 2 m 0), is necessarily smaller
than the massless spring of Eq. 5, the predictions of Eq.
31 were not included in Table II. APPENDIX A: DETAILS OF MEASURING kg
As seen in the rms averages of Table II, the consistency
between Eqs. 42 and 29 is impressive. Although the rms Eighteen masses were gradually added to each spring and
difference over all springs is 0.605 kg for Eq. 42, it predicts the corresponding stretch was measured using a mirrored
the mass to within 3% for the stiffest spring and has an rms ruler to minimize parallax. The masses used to cause the
difference of only 0.032 kg for the other springs. This pre- stretch ranged from 0.150 to 1 kg for the springs with mid-
dictive capability is not too different from Eq. 29, which range stiffness, 0.010 to 0.500 kg for the soft spring, and 3 to
has an rms difference below 0.03 kg whether the stiff spring 15 kg for the stiff spring. The first two mass ranges were
is included or not. If Eq. 42 does not give accurate results measured to a precision of 0.1 g; the third was measured to
and one wishes to avoid solving a cubic polynomial, one within 2 g. The corresponding displacement was measured to
could use the calculated mass to estimate the O(y) terms in 0.1 cm for all but the two softest springs. Due to limitations
Eq. 40 to determine how far this prediction is from Eq. in construction, these were measured to a precision of 0.2
29. cm. The slope was found by a least-squares fit16 to account
It is interesting that Eq. 29 does not give exact results. It for the measurement uncertainty in both the added mass and
is possible that the problem is due to the systematic uncer- the stretch. To give an idea of the value of the spring con-
tainty discussed in Sec. IV B. As noted there, for half of the stants for the springs used, we assume g9.80(1) m/s2 and
springs, the predictions of Eq. 29 are within m of the divide by the slope g/k. This value is listed as k in Tables
experimental value and within 2 m for the remaining springs IIII. The reduced 2 of these fits, 2 2 / with equal to
if we assume that 0.5 cm. This is a sizable uncertainty the degrees of freedom, are listed in Table III. Each spring
for the measurement of lengths and so this estimate may only has eighteen data points except for the k25 N/m spring.
account for a portion of the difference. Another explanation The value of 2 0.3 indicates that the precision in the dis-
for the inaccuracy of Eq. 29 might come from the assump- placement may have been overestimated, so the k values

826 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 826
 Table III. Tests of the uniform stretch coil density assumption. The values of k are from the slope of a graph
of m vs x, assuming that g9.80(1) m/s2 . The reduced chi-square, 2 , values indicate that the precision of
the displacement may be overestimated. The ratio Z c.m. / compares the distance from the top of the spring to
the calculated center of mass, Z c.m. , to the length of the spring, . This ratio is evaluated at the experimental
mass value and deviations from 50% indicate the nonuniformity of the stretch. A graph of (T/2 ) 2 vs mass has
slope s and intercept b. If we assume uniform stretch as discussed in Appendix B, the reciprocal slope (1/s)
should give k without assuming a value of g), the ratio 3b/s should give the mass of the spring, and the ratio
of b/(sm s ) should be 1/3. Differences from these values should correlate to differences in Z c.m. / from 50%.

k N/m 2 Z c.m. / % m s kg 1/s 3b/s b/(sm s )

3.0591 1.196 16 55.08 0.67792 3.2608 0.8503 0.4182

6.6963 0.336 16 52.67 0.33902 6.8379 0.3591 0.3531
25.164 0.309 15 50.71 0.08292 25.446 0.0984 0.402
28.635 0.355 16 50.60 0.08242 28.815 0.0923 0.371
36.268 0.316 16 50.51 0.08842 36.566 0.0953 0.361
40.91 0.379 16 50.44 0.08462 41.145 0.0943 0.371
51.42 0.164 16 50.31 0.07862 51.91 0.0854 0.362
10022 0.494 16 50.03 0.06292 10153 0.079 0.45

assume the common uncertainty: 2c i,ave 2 , which gives

mm s g

c 0.03 0.06 cm for all but the two softest cases, which 0
k 1

have c 0.3 cm. Z c.m. ,
m sg 0 mm s g/k
ln
k 0 mg/k
B5
APPENDIX B: NONUNIFORM COIL DENSITY IN A

m s /k0
MASSIVE SPRING 1 m s g/k m sg 2
Z c.m. O . B6
In a massive spring, each coil stretches all of the coils 2 12 0 mg/k k
above it and none of the coils below it, which implies a
nonuniform distribution of the coils of the spring. When Equation B5 shows that Eqs. 22 and 27 should be modi-
hung under its own weight, the density must increase toward fied significantly to account for the shift in the center of
the bottom. The mass per coil is still considered uniform; it is mass. Equation B6 shows that for either a light spring or a
the nonuniform distribution of the coils that produces a non- stiff spring, the center of mass approaches the midway point,
uniform distribution of the spring mass. In fact, the mass /2.
density is As an initial test of the nonuniformity of the springs, Eq.
B6 was used to calculate the location of the center of mass
ms for the static hanging spring. If the spring is perfectly uni-
z 0 , B1 form in stretch, then Z c.m. should be 50% of 0 mg/k
mg z m s g
0 m s g/2k. Table III lists the ratio Z c.m. / for the springs
k k used in this experiment. The mass used was that predicted by
Eq. 29. The value for the ratio will, of course, change with
where 0 is a dimensionless normalization constant given by

different m. The assumption of uniform density might war-
1 rant further investigation for better accuracy in the softer
mg m s g springs, but from Z c.m. , the assumption seems reasonable for
0
m sg k k all but the softest springs.
0 ln B2 After relaxing the uniform density assumption, Eq. 10
k mg
0 can still be integrated, but is rather more involved. It is pos-
k
sible to derive a fairly complicated formula for the effect of
the spring mass in terms of 0 , k, m, and m s , which does
and 0 (g/k) (m m s /2). Because the argument of the
not seem more useful than previous approximations.
logarithm also can be expressed as ( m s g/2k)/( On the other hand, it is possible to account for variations
m s g/2k), it is possible to write 0 as in the m s coefficient without doing any more work than it

1 would take to measure the spring constant, which can be

m sg m sg done using Eq. 14 rather than via Hookes law. First attach
0 tanh1 . B3
k 2k the spring to the desired support, which ideally will not flex
with the bouncing of the spring. Then, for a collection of
We can verify that 0 (z)dzm s , that the average density masses, let the spring oscillate and measure each period. If
matches the uniform density used in Sec. II A, and that we rearrange Eq. 14 as

Z c.m.
1
ms

0

z z dz, B4 T
2
2
1
m
k
ms
3k
, B7

827 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 827
we see that a plot of (T/2 ) 2 vs m will give a straight line
5
Equation 25 in Ref. 2 for the period with which the energy oscillates
between these modes involves elliptical integrals of the first kind, even for
with slope 1/k and an intercept that, when divided by the
the case of a massless spring as a simple pendulum.
slope, should be close to m s /3. Any deviation from m s /3 6
E. E. Galloni and M. Kohen, Influence of the mass of the spring on its
would indicate the value that should be used in Eq. 14. We static and dynamic effects, Am. J. Phys. 47 12, 1076 1078 1979.
can use the ratio of intercept divided by the slope and the 7
S. Y. Mak, The static effectiveness mass of a slinky, Am. J. Phys. 55
actual mass of the spring in place of the 1/3 that currently 11, 994 997 1987. See also P. K. Glanz, Note on energy changes in
multiplies m s in Eq. 14. In other words, the ratio of inter- a spring, ibid. 47 12, 10911092 1979 for a discussion of the effect of
cept divided by the slope is a direct measurement of the D pre-loading a spring on the conservation of energy.
8
T. C. Heard and N. D. Newby, Jr., Behavior of a soft spring, Am. J.
that appears in Eq. 16.
Phys. 45 11, 11021106 1977. See especially Eq. 20.
As a second test of the nonuniform density, the spring 9
J. T. Cushing, The spring-mass system revisited, Am. J. Phys. 52 10,
constant for each spring was found by measuring the period 925937 1984. See footnote 16.
of 18 different masses over 100 oscillations. Fifty oscilla- 10
R. Weinstock, Spring-mass correction in uniform circular motion, Am.
tions were used for the soft springs. The graphs of (T/2 ) 2 J. Phys. 32 5, 370376 1964. See also M. Parkinson, Spring-mass
vs m are available from the author. The last three columns of correction, ibid. 33 4, 341342 1965 for the same result from iterative
Table III are from this graph. The reciprocal of the slope 1/s methods rather than differential equations.
11
R. J. Stephenson, Mechanics and Properties of Matter Wiley, New York,
should give the spring constant. These differ from those
1952, pp. 113114. See also L. Ruby, Equivalent mass of a coil spring,
listed in Table I, because Table I assumes g9.80(1) m/s2 . Phys. Teach. 38, 140141 2000 for the same result from iterative meth-
Three times the intercept divided by the slope (3b/s) should ods to avoid the calculus for an algebra-based course. Ruby also attempts
be close to m s . Equivalently, the quantity b/(sm s ) should be a simplified nonuniform approximation, but incorrectly relates Maks Ref.
close to 1/3, the coefficient used for m s . These last two 7 static correction to his dynamic correction.
12
comparisons are equivalent measures of the accuracy of the R. Weinstock, Oscillations of a particle attached to a heavy spring: An
application of the Stieltjes integral, Am. J. Phys. 47 6, 508 514 1979.
assumption of uniform coil density. The value of b/(sm s ) for 13
F. A. McDonald, Deceptively simple harmonic motion: A mass on a
the stiffest spring also is consistent with 1/3, having a some- spiral spring, Am. J. Phys. 48 3, 189192 1980.
what imprecise result of 0.45. Interestingly, the softest 14
J. M. Bowen, Slinky oscillations and the motion of effective mass, Am.
spring has a larger ratio, 0.4182, which is inconsistent with J. Phys. 50 12, 11451148 1982.
the 4/ 2 0.405 284 7 predicted by Refs. 6, 8, 9, and 14. 15
As mentioned in this article, a mass hanging in equilibrium from a spring
Except for the stiffest and softest springs, the weighted av- will satisfy kzmg. A graph of displacement versus added-mass then
erage of the results is 0.3541, which indicates a difference gives a slope of g/k. As described in Appendix B, a mass that is hung from
from 1/3 and might explain why the cubic error bars do not a spring and set to oscillating has a characteristic period. A plot of T/2 2
overlap the experiment in every case. vs m then gives a line with slope 1/k. In principle, g/k divided by 1/k will
give a calculation of the local gravitational field. In a paper in preparation,
a these values are compared to the acceleration due to gravity measured by
Electronic mail: [email protected]
1
M. G. Olsson, Why does a mass on a string sometimes misbehave?, a spark-machine free-fall experiment and the gravitational field measured
Am. J. Phys. 44 12, 12111212 1976. by a swinging pendulum. The free-fall experiment gave repeatable results
2
H. M. Lai, On the recurrence phenomenon of a resonant spring pendu- consistently lower than expected. The springs and pendula independently
lum, Am. J. Phys. 52 3, 219223 1984. gave repeatable results that were higher than expected and consistent with
3
W. R. Mellen, Spring string swing thing, Phys. Teach. 32, 122123 each other. The systematic uncertainties are still being investigated.
16
1994. Philip R. Bevington and D. Keith Robinson, Data Reduction and Error
4
D. E. Holzwarth and J. Malone, Pendulum period versus hanging-spring Analysis for the Physical Sciences McGrawHill, New York, 1992, 2nd
period, Phys. Teach. 38, 47 2000. ed., Chaps. 6, 11.

828 Am. J. Phys., Vol. 72, No. 6, June 2004 Joseph Christensen 828