Eigenvalues and Musical Instruments

V. E. Howle and Lloyd N. Trefethen

August 20, 1997
Abstract

Most musical instruments are built from physical systems that oscillate at certain natural frequencies. The frequencies are the imaginary
parts of the eigenvalues of a linear operator, and the decay rates are
the negatives of the real parts, so it ought to be possible to give an approximate idea of the sound of a musical instrument by a single plot of
points in the complex plane. Nevertheless, the authors are unaware of
any such picture that has ever appeared in print. This paper attempts
to ll that gap by plotting eigenvalues for a guitar string, a ute, a
clarinet, a kettledrum, and a musical bell. For the drum and the bell,
simple idealized models have eigenvalues that are irrationally related,
but as the actual instruments have evolved over the generations, the
leading ve or six eigenvalues have moved around the complex plane
so that their relative positions are musically pleasing.

1 Introduction
Linear systems like strings and bars and drums have eigenvalues, which are
numbers in the complex plane C . If the system is governed by an equation
du = 1 Au
dt 2
where A is a matrix or linear operator, then an eigenvalue  2 C of A
corresponds to a solution

u(t) = e

t=2 u(0)

where u(0) is a corresponding eigenvector. An eigenvalue on the imaginary

axis corresponds to oscillation without decay, and an eigenvalue to the left
of the imaginary axis corresponds to oscillation with decay. Thus Im is the
(real) frequency of the oscillation and ;Re is its decay rate.
Nonlinear oscillators have eigenvalues too, if one linearizes by considering innitesimal motions. This may give a good initial picture of the
1

behavior, which can then be improved by considering changes introduced

by the nonlinearity.
In this report our goal is to bring these ideas to life by showing a sequence of more than a dozen pictures of eigenvalues in the complex plane
of various oscillating systems. The systems we consider are musical instruments. Typically we ask, what do the eigenvalues look like for the simplest
idealization, such as a perfect string or tube or membrane? How do they
change when terms are added to the model corresponding to sti ness, or
friction, or sound radiation? And how does this all match frequencies and
attenuation rates that have been measured in the laboratory?
Eigenvalues do not tell us everything about the sound of a musical instrument. If we wanted to fully distinguish a guitar from a ute, for instance,
even without getting into nonlinear e ects, we would want to know the relative amplitudes of the excited modes. Still, the eigenvalues represent a good
starting point.
There is nothing new in the study of the physics of musical instruments,
which is set forth beautifully in the books by Benade 2, 3] and Fletcher
and Rossing 7], nor in the portrayal of eigenvalues in the complex plane.
Nevertheless, we have not found a book or paper that combines the two
as we do here. Thus this paper is solely expository|but the basis of the
exposition, we think, is new, and well suited to applied mathematicians.

2 Guitar String
Let us start with the eigenvalues of an ideal guitar string. By \ideal" we
mean that the string is perfectly exible, that there are no internal losses
or losses to the string supports, and that there is no damping due to the
surrounding air. To be denite, let us suppose the length of our string is
L = 65 cm, the mass density is  = 1:15 g=cm3, and the radius is r =
0:032 cm. These values correspond approximately to the high E string on a
nylon string guitar. Typical tension values are between 50 and 80 Newtons
7, p. 212]. Given the values for mass density, cross-sectional area, and string
length just chosen, we can \tune" our ideal guitar string to high E (329:63
Hz  330 Hz) by setting the tension to T = 67:09 Newtons.
The motion of an ideal string is governed by the second-order wave equation, utt = c2uxx , where c2 = T=S is the square of the wave speed. Figure 1
shows the eigenvalues for these choices or parameters.
In this gure, the fundamental frequency, the rst eigenvalue, is a good
approximation to the expected fundamental frequency, 2;5=12440 Hz  330
Hz, for a high E guitar string (ve half-steps below concert A, which has a
frequency of 440 Hz). The remaining eigenvalues, the higher harmonics, are
integer multiples of the fundamental frequency. The eigenvalues have zero
2

1980
1650
1320

frequency (cycles/sec.)

990
660

; fundamental frequency

330

329.6 Hz

0
330
660
990
1320
1650
1980
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

Figure 1: Eigenvalues for an ideal high E guitar string. There are no losses,
so the eigenvalues are pure imaginary.
real part, corresponding to our ideal string experiencing no losses.
For every positive frequency, we always have a corresponding negative
frequency of the same magnitude. From now on, we will only show the upper
half-plane.
If we now gently touch the guitar string in the middle, we damp out the
fundamental and all of the odd harmonics. The odd eigenvalues all shift
into the left half-plane. Ideally speaking, the even eigenvalues are not much
a ected, since their corresponding eigenmodes already have a node at the
point we are touching. However, in reality they are a ected somewhat since
a nger is not sharp and cannot damp the string exactly in the middle,
but damps a small region at the middle of the string. Figure 2 suggests
schematically what happens to the eigenvalues when we gently touch the
middle of a high E string with a very narrow nger.
We are still assuming in this picture that the string experiences no losses
other than the damping from touching it in the middle. A real guitar is not so
simple (if it were, we couldn't hear it!). Some of the physical phenomena that
make it deviate from the ideal are sti ness in the strings and damping. The
e ect of sti ness is to stretch the frequencies slightly, making them inexact
harmonics of one another. However, the sti ness of a string depends on its
thickness, and guitar strings are not generally thick enough for sti ness to
3

3960
3630
3300

frequency (cycles/sec.)

2970
2640
2310
1980
1650
1320
990
660
330
0
400

300

200

100

100

decay rate (efoldings/(2 )

200

300

400

sec.)

Figure 2: The eect of gently touching the guitar string in the middle|
schematic. Odd harmonics are greatly attenuated, while even harmonics are
little aected, so the pitch goes up an octave. This eect is analogous to the
eect of opening the register hole in a ute, recorder, or clarinet.
be an important factor. On the other hand, losses from damping are indeed
important. Their e ect is to shift the eigenvalues into the left half of the
complex plane, some of them further than others.
In a real guitar, the most important phenomena a ecting the eigenvalues
are damping due to air viscosity and internal losses. Which is more important depends on the materials used. In a nylon string guitar, the higher
modes decay mainly as a result of internal damping in the string. In a steel
string guitar, on the other hand, the main damping mechanism is air viscosity. Steel strings are actually damped less by air viscosity than nylon
strings, since they are thinner, but the e ect is relatively more important
because the internal losses are negligible.
The energy losses associated with non-rigid end supports are in one sense
the most important, for it is these losses that couple the string to the soundboard and thus provide most of the sound volume that we hear. However,
this type of loss is actually not very large in magnitude, and thus it is not
essential to our picture of the eigenvalues.
Consider rst damping from the viscosity of the air in which the string
is vibrating. Viscous damping depends on the velocity of the string, and
so a ects each frequency di erently. In particular, the decay rate due to
4

air viscosity is roughly proportional to the square root of the frequency.

Specically, the decay rate is
p

a = af (2M22M + 1) 
where  is the density of the string material, a is the air density,
p f is
the frequency, S is the cross-sectional area of the string, r = S= is the
radius of cross-section
of the string, a is the kinematic viscosity of air, and
p
M = (r=2) f= a 7, p. 50].

Consider a steel string guitar. Let us assume that the string is still 65
cm long, the radius is r = 0:017 cm, the mass density is  = 7:8 g=cm3 ,
and the tension is T = 1:228  107 dynes. (Typical tension values for steel
strings are between 107 and 1:8  107 dynes 7, p. 212] we have again
\tuned" the tension to get a fundamental frequency of approximately 330
Hz.) Assume the air density is a = 0:0012 g=cm3 and the kinematic density
is a = 0:15 cm2=s. In Figure 3, the eigenvalues have shifted into the left
half-plane, the magnitude of the shift being proportional to the square root
of the frequency.
3960
3630
3300

frequency (cycles/sec.)

2970
2640
2310
1980
1650
1320
990
660
330
0
2

1.8

1.6

1.4

1.2

0.8

0.6

decay rate (efoldings/(2 )

0.4

0.2

sec.)

Figure 3: Steel guitar string with air damping. Higher frequencies are attenuated more than lower ones.
Going back to the nylon string guitar parameters we considered above,
when we include air damping, the picture changes much more than it did
for the steel string (Figure 4). The eigenvalues shift farther into the left
5

half-plane, since the nylon string is thicker and less dense.

3960
3630
3300

frequency (cycles/sec.)

2970
2640
2310
1980
1650
1320
990
660
330
0
2

1.8

1.6

1.4

1.2

0.8

0.6

decay rate (efoldings/(2 )

0.4

0.2

sec.)

Figure 4: Nylon guitar string with air damping. Since the nylon string is
thicker and less dense than the steel one, the losses are greater.
Next, let us consider internal damping. This type of damping occurs
because of energy lost in bending the string. It is generally negligible for
metal strings, but it can be important for nylon strings. Internal losses
can be represented by assuming a complex Young's modulus for the string
material. The characteristic decay rate i from internal damping is roughly
proportional to the frequency and is given by

Q2 
i = f Q
1

where f is the frequency and Q1 + iQ2 is the complex Young's modulus

7, p. 51]. Q1 relates to the elastic bond distortions and Q2 relates to
relaxation processes such as dislocation motion or the movement of kinks in
the polymer chains 7, p. 51]. Figures 5 and 6 show the pictures we get for
our same representative nylon and steel high E guitar strings, if we assume
somewhat arbitrarily that Q2 =Q1  3:75  10;6 for steel and 4:78  10;4
for nylon. (These ratios are chosen roughly to match the data on internal
damping in Pickering 12, Table 1] and Hancock 8, Table 5B].) Notice that
the e ect is large for the nylon string, but very small for the steel string.
Putting together both types of damping, we get the pictures of the nylon
and steel strings summarized in Figures 7 and 8.
6

3960
3630
3300

frequency (cycles/sec.)

2970
2640
2310
1980
1650
1320
990
660
330
0
2

1.8

1.6

1.4

1.2

0.8

0.6

decay rate (efoldings/(2 )

0.4

0.2

sec.)

Figure 5: Nylon guitar string with internal losses. The attenuation rates are
great, representing the chief energy loss mechanism for nylon strings at high
frequency.
3960
3630
3300

frequency (cycles/sec.)

2970
2640
2310
1980
1650
1320
990
660
330
0
0.2

0.18

0.16

0.14

0.12

0.1

0.08

0.06

decay rate (efoldings/(2 )

0.04

0.02

sec.)

Figure 6: Steel guitar string with internal losses. Since a steel string is thin
and elastic, these attenuation rates are small, much less signicant than the
damping introduced by air viscosity.
7

3960
3630
3300

frequency (cycles/sec.)

2970
2640
2310
1980
1650
1320
990
660

both
air
internal

330
0
5

4.5

3.5

2.5

1.5

decay rate (efoldings/(2 )

0.5

sec.)

Figure 7: Nylon guitar string with both types of damping. Internal losses
dominate at higher frequency, making the overall decay rate approximately
proportional to the frequency.
3960
3630
3300

frequency (cycles/sec.)

2970
2640
2310
1980
1650
1320
990
660

both
air
internal

330
0
1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

decay rate (efoldings/(2 )

0.2

0.1

sec.)

Figure 8: Steel guitar string with both types of damping. Air damping dominates, making the overall decay rate approximately proportional to the square
root of the frequency.
8

Of course, there are other instruments based upon strings than guitars,
and in these, other phenomena may be important in determining the eigenvalues. In guitars, energy loss from from non-rigid end supports is only
important in the bridge end, since the neck end is stopped with a rigid end
support, the fret. However, losses to end supports can be very important,
for example, in violins, where the string is stopped at the neck end with a
nger instead of a fret (especially in a violin played pizzicato 5, p. 15]). In
pianos, the increased spacing of eigenvalues caused by sti ness of the strings
becomes important.

3 Flute
In this section, we start with the eigenvalues of an ideal ute and gradually
add in the e ects of some additional physical phenomena, as we did with
the guitar string.
Consider a ute of length L with no tone holes. The oscillator in a
ute is the column of air in the tube, which is kept oscillating by the air
blowing across a sharp ridge. In the simplest model, inside the tube are
plane pressure waves that obey the same equation as do transverse waves
on a string,
utt = c2uxx
where c is the speed of sound in air under normal conditions, u is the deviation from equilibrium pressure, and the x-axis is oriented along the center
of the ute bore.
Since the ute is open at both ends and the openings are narrow compared with the sound wavelength, there are pressure nodes at both the head
and foot of the ute (where the air column is in contact with the surrounding air) 17, p. 513]. It follows that the eigenvalues are exactly the same as
those for an ideal string xed at both ends. They are given (in Hertz) by
nc=(2L), where n is the mode number.
Under typical conditions, the speed of sound in air is 34400 cm=s 10,
p. 222]. If we want a simple ute whose lowest note has a fundamental
frequency approximately equal to the C below concert A (approximately
261:6 Hz, corresponding, for example, to a modern ute), its length should
be L = c=(2;9=12  440s;1  2)  65:7 cm. The eigenvalues are all the
integer multiples of this frequency, exactly as in the case of the ideal guitar
string (Figure 9). The only di erence in the picture is that the fundamental
frequency is 261:6 Hz.
Opening the register hole in a ute is analogous to lightly touching a
guitar string in the middle. An ideally located register hole in a ute is at
half the acoustic length of the tube. Opening the hole greatly attenuates
9

3930
3668
3406

frequency (cycles/sec.)

3144
2882
2620
2358
2096
1834
1572
1310
1048

fundamental frequency

786
524

; 261.6 Hz

262
0
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

Figure 9: Eigenvalues of an ideal C ute, of length 65:7 cm. There are no

losses, and the ute has no nger holes.
the fundamental and all of the odd harmonics, raising the pitch of the note
by an octave, as in Figure 2.
In this model, as in our ideal guitar, we are ignoring all losses (other than
those from the register hole). To begin to develop a more realistic picture of
the ute, we consider length corrections, wall losses, and losses from sound
radiation.
In a real ute (even one without tone holes), the air pressure does not
reach equilibrium precisely at the ends of the tube. The pressure waves
extend slightly past the ends before coming to equilibrium with the outside
air. How far they extend depends on the cross-sectional area of the tube and
the acoustic conductivity of the open ends. This open-end correction lowers
the frequencies of all the harmonics, making the \acoustic length" of the
tube somewhat longer than its physical length. A reasonable approximation
is that the acoustic length is L + S=ca, where L is the physical length of the
tube, S is its cross-sectional area, and ca is the acoustic conductivity of the
open end 11, p. 613]. Therefore the nth eigenvalue becomes nc=2(L + S=ca).
Given a ute with an internal radius of 0:95 cm and assuming that the
acoustic conductivity is 1:9 cm (the acoustic conductivity is approximately
equal in magnitude to the diameter of the ute 11, p. 613]), for the ute
to have a fundamental frequency of approximately 261:6 Hz, its (physical)
10

length would have to be L = c=(2;9=12  440  2) ; S=ca  64:3 cm instead of 65:7 cm. This length is a little closer to the length of a real ute
(approximately 62 cm).
The two main mechanisms for energy loss in a ute are frictional and
thermal energy transfer to the instrument walls and sound radiation. The
wall losses have the greatest e ect on the eigenvalues 1, p. 142].
The damping rate due to wall losses is approximately proportional to the
square root of the frequency. A reasonable approximation to the damping
rate  is
 = 2A!1=2c=r
where A is a constant depending on the wall material and surface condition
(A  2:5  10;5 s1=2 for woodwinds), ! is the angular frequency, and c is
the speed of sound in air 1, pp. 142-43]. Figure 10 shows the e ect of wall
losses in the ute.
3930
3668
3406

frequency (cycles/sec.)

3144
2882
2620
2358
2096
1834
1572
1310
1048
786
524
262
0
50

45

40

35

30

25

20

15

decay rate (efoldings/(2 )

10

sec.)

Figure 10: Flute eigenvalues with wall losses. Note that the decay rates
are large. Fortunately, the player keeps blowing, supplying the system with
energy.
Next consider losses from sound radiation. It is this energy loss that lets
us hear the ute. In our ideal ute, without any tone holes, all of the sound
radiation is from the end. Losses from sound radiation at the end of a tube
are approximately proportional to the square of the frequency 7, p. 183],

 = 4 ( rl )2(2n ; 1)!
11

where r is the bore radius, l is the length of the ute, n is the mode number,
and ! is the angular frequency 1, p. 143]. Figure 11 shows the e ects
of radiation losses on the eigenvalues of our C ute. Figure 12 shows the
combined e ect of both types of losses.
3930
3668
3406

frequency (cycles/sec.)

3144
2882
2620
2358
2096
1834
1572
1310
1048
786
524
262
0
50

45

40

35

30

25

20

15

decay rate (efoldings/(2 )1 sec.)

10

Figure 11: Flute eigenvalues with radiation losses from the end the tube.

4 Clarinet
The most important di erence between a clarinet and a ute is that the
oscillations in a clarinet are driven by a reed. The main e ect of the reed is
to essentially close the reed end of the instrument.
A simple model of a clarinet (without tone holes) is therefore a cylindrical
tube open at one end and closed at the other. The closed end lowers the
fundamental frequency by an octave compared with that of a ute of the
same length, since we now have a pressure antinode at the closed end instead
of a node. The wavelength of the fundamental is thus four times the length
of the tube instead of twice the length, and the fundamental frequency is
halved.
The closed end also means that the overtones of the clarinet are only
the odd multiples of the fundamental. The even harmonics would have a
pressure node at the reed end of the instrument, which is impossible since the
closed end forces an antinode at that point. Figure 13 shows the eigenvalues
for an ideal B  clarinet.
12

3930
3668
3406

frequency (cycles/sec.)

3144
2882
2620
2358
2096
1834
1572
1310
1048
786
both
wall losses
radiation losses

524
262
0
100

90

80

70

60

50

40

30

decay rate (efoldings/(2 )

20

10

sec.)

Figure 12: Flute eigenvalues with both wall losses and radiation losses.

3861
3627
3393
3159

frequency (cycles/sec.)

2925
2691
2457
2223
1989
1755
1521
1287
1053
819

fundamental frequency
; 116.5 Hz

585
351
117
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

Figure 13: Eigenvalues of an ideal B  clarinet. The fundamental is an octave

lower than that of an ideal ute of the same length, and the harmonics are
all odd multiples of the fundamental.
13

3861
3627
3393
3159

frequecy (cycles/sec.)

2925
2691
2457
2223
1989
1755
1521
1287
1053
819
585
351
117
400

300

200

100

100

200

decay rate (efoldings/(2 )1 sec.)

300

400

Figure 14: The eect of opening the register hole on a clarinet|schematic.

The two-thirds of the harmonics that did not already have a node at the
register hole are greatly attenuated, and the pitch goes up a twelfth. Compare
Figure 2.
Because we only have the odd harmonics, the register hole in a clarinet
is situated di erently than in a ute. In a ute, the register hole is ideally
located at half the acoustic length of the tube. Opening the hole damps the
fundamental and all of the odd harmonics, raising the pitch by an octave. In
a clarinet, the register hole is ideally located at a third of the acoustic length
of the tube. Opening the hole damps the fundamental and all but every third
(odd) upper harmonic, raising the pitch by a twelfth. See Figure 14.
Unlike in the ute, in a clarinet the tone holes are not large enough
relative to the bore size to cut o the tube at their location. The open holes
of a clarinet change the character of the sound in a more complicated way.
The tube e ectively ends at a point somewhat past the rst open tone hole,
attening the fundamental. The acoustic length of the tube due to open
tone holes increases with frequency, attening the upper harmonics relative
to the fundamental by an appreciable fraction of a semitone. However, such
e ects are generally countered by appropriate shaping and placement of
the holes in the construction of real clarinets, so the eigenvalue picture is
actually much the same as in the ideal case 2].
Like the ute, the main causes of energy loss in a clarinet are wall losses
and sound radiation. Other than having a lower fundamental and only
having the odd harmonics, the eigenvalues after incorporating losses look
14

similar to those for the ute.

5 Drums
Strings, utes, and clarinets are all essentially one-dimensional resonators,
and that is why their eigenvalues fall naturally into integer ratios, making
the sound strongly musical. We turn now to instruments that are not essentially one-dimensional, where more complicated physical e ects have been
exploited by designers over the years|largely by trial and error|to move
the eigenvalues to favorable locations so as to achieve a musical e ect.
We begin with a drum. The simplest mathematical description of a drum
is as an ideal circular membrane with clamped edges. By ideal, we mean
that the membrane is perfectly uniform, has no sti ness, and experiences
no losses. In particular, in this simplest description we are neglecting the
surrounding air and the body of the drum.
The equation of motion for a circular membrane is
1
r2 = 

c2

tt

where (r ) is the displacement of the membrane from its equilibrium at

the point (r ) on its surface, c2 is T= , T is the tension, and  is the
mass density per unit area. The eigenmodes are obtained by assuming a
solution of the form (r ) = R(r) ()e;i!t. This leads to the solution
(r ) = AJm (z) cos(m), where Jm (z) is the mth order Bessel function of
the rst kind and z = !r=c. For a membrane of radius a with xed edge,
the allowed frequencies are the values ! such that Jm (!a=c) = 0 9].
For example, given a membrane of radius 0:33 m, tension T = 4415
N/m, and mass density  = 0:26 kg=m2, which corresponds to a typical
mylar membrane kettledrum, the eigenvalues up to 600 Hz are shown in
Figure 15. The mode (mn) refers to the nth zero of Jm (!a=c) = 0. The
eigenfunction corresponding to mode (mn) has m nodal diameter lines and
n circular nodes.
Notice that the eigenvalues of the ideal membrane are not in the least
harmonically related. A real kettledrum, on the other hand, has a denite
pitch. To explain this e ect we need to add some more realistic physical
properties to the kettledrum. The two most important properties to consider
are the membrane moving through the air and the presence of the kettle. Our
discussion follows the laboratory experiments and highly readable articles
of T. D. Rossing 14, 15].
According to Rossing, the most important phenomenon in terms of making the kettledrum sound with a denite pitch is the interaction of the membrane with the air. This \air loading" has two important e ects. It tends to
15

600
(51)

(03)

(22)

500
(41)

frequency (cycles/sec.)

(12)
(31)

400
(02)
(21)
300

(11)
200
(01)
100

0
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

Figure 15: Eigenvalues for an ideal drum, with angular (m) and radial (n)
mode numbers indicated in parentheses (mn). These zeros of Bessel functions are not at all harmonically related
an ideal drum would not sound
with any denite pitch.
lower certain modes, making them more closely harmonic 14, p. 280], and
it tends to damp more quickly the modes that are not nearly harmonic. We
will not go into the physics, but will sketch some of the consequences.
The (11), (21), (31), (41), and (51) modes are shifted in frequency so that
they are more nearly harmonically related. The (01), (02), and (03) modes
are not harmonic, even with the air loading, but these modes are damped
out very quickly. See Figure 16, which shows frequencies and decay rates
for a real kettledrum head (without the kettle) as measured by Christian,
et al. 4, Table V]. The drum has the same physical properties and tension
as the ideal membrane described above.
Notice that the (0n) modes, which are not harmonically related, decay
very quickly. In addition, the (11), (21), (31), and (41) modes have shifted
in frequency so that they are more nearly harmonically related. Observations more than a century ago by Rayleigh were the rst to show that the
inharmonic but heavily damped (0n) modes have little e ect on the sound
of the drum. He observed that touching the drum in the center, further
damping these modes, has little e ect on the sound 13, p. 348].
The e ect of the kettle is to further tune the modes that are least
damped. The modes that were made more nearly harmonic by including
16

519
(51)

frequency (cycles/sec.)

(03)

(22)

432.5

(41)
(12)
346
(31)
(02)

259.5

(21)

(11)

173

(01)

86.5

20

18

16

14

12

10

decay rate (efoldings/(2 )1 sec.)

Figure 16: Eigenvalues for a kettledrum head (without the kettle), estimated
from experimental measurements 4, Table V]. Notice that the least damped
frequencies are now roughly harmonic. (There are presumably other modes
above 500 Hz that were not included in these measurements.)
air loading are shifted even closer to harmonic by the presence of the kettle.
This tuning is fairly small, but the kettle has a more important e ect on
the damping rates. It tends to make the harmonic modes, (11), (21), (31),
(41), and (51), radiate less e"ciently and therefore sustain longer, making
the drum sound more musical 15, p. 177]. (Although these modes are harmonically related by a \missing fundamental" at one-half the frequency of
the (11) mode, the pitch of the drum is usually heard as that of the (11)
mode itself 15, p. 174].)
Figure 17 shows modal frequencies and decay rates for a real kettledrum
(with the kettle) as measured by Christian, et al. 4, Table II]. Note that
the tension on this drum is not the same as in the previous measurements.
This drum has tension T = 3710 Newtons/m.
There are other properties of a real kettledrum that a ect its sound,
such as bending sti ness and sti ness to shear in the membrane, but these
e ects are not as large 4, p. 1336].
Given that the e ects of air loading and the presence of a kettle tend to
make the modes approximately harmonic, why don't all drums sound with a
denite pitch? Part of the answer is that years of design have led to timpani
whose parameters (radius, tension, kettle size, etc.) have been favorably
17

580

frequency (cycles/sec.)

507.5

435

(51)
(03)

(22)

362.5

(41)

290

(12)
(31)
(02)
(21)

217.5

145

(11)
(01)

72.5

20

18

16

14

12

10

decay rate (efoldings/(2 )1 sec.)

Figure 17: Eigenvalues for a kettledrum head (with kettle), estimated from
experimental measurements 4, Table II]. Modes (11), (21), and (31) are now
almost exactly harmonically related. The open circle represents where the
\missing fundamental" would be. (Again, some higher modes are omitted.)
adjusted to achieve a musical e ect. Drums constructed \at random" are
hardly likely to sound the same. Another part of the answer is that many
drums are intended to blend with any key and are therefore intentionally
designed not to have a denite pitch. One method is to use two drumheads,
tuned to two di erent frequencies. This is the situation, for example, in a
bass drum. The two membranes are typically tuned as much as a fourth
apart and the numerous partial frequencies in combination give little sense
of a pitch 7, p. 512].
Figure 18 shows some modal frequencies and decay rates for a real bass
drum as measured by Fletcher and Bassett 6, Table II].

6 Bells
Bells, like drums, are multi-dimensional objects, and a bell constructed \at
random" will certainly not sound musical. Centuries of evolution, however,
have done remarkable things to the eigenvalues of certain bells. We conclude
our tour with Figure 19, which shows the measured eigenvalues of an actual
A4] minor-third bell. The fundamental frequency is 456:8 Hz. In bells, there
are two modes that have three nodal diameters and one nodal line. In one,
18

600

frequency (cycles/sec.)

500

400

300

200

100

0
5

4.5

3.5

2.5

1.5

decay rate (efoldings/(2 )1 sec.)

0.5

Figure 18: Eigenvalues for a bass drum as measured by Fletcher and Bassett
6, Table II]. For a drum of this kind, no attempt has been made to achieve
harmonic relationships among the frequencies.
the nodal circle is at the waist of the bell, and at the other, it is nearer
the mouth of the bell. The rst, which corresponds to the minor third, is
referred to in the gure as the (31) mode. The second corresponds to the
perfect fth and is referred to as the (31#) mode. Similarly, the (21#) mode
has a nodal diameter near the mouth of the bell instead of around the waist
of the bell. The (21#) mode is the fundamental frequency, and the (20)
mode an octave lower is known as the hum.
Eigenvalues of bells such as these are discussed in the beautiful dissertation of Roozen{Kroon 16]. For our purposes, it is enough to note the
astonishingly satisfying imaginary parts of the rst six eigenvalues in Figure 19. These six notes line up like a chord played on a piano, and with
decay rates as low as about half an e-folding per second, you can almost
hear this clean bell ring.

7 Acknowledgments
It is a pleasure to thank Prof. Allard Jongman for providing the facilities of
the Cornell Phonetics Laboratory for digitizing some guitar sounds, which
enabled us to conrm some of the decay rates of Figures 1-8. (Of course, real
data are never quite as clean as in our gures.) The work of both authors
19

A#

(61)

frequency (cycles/sec.)

1828

1371

914

A#
F
C##
A
A#

(41)

(31#)

685.5
548.4

(31)
(21#)

457

(20)

228.5

A#

(51)

1.8

1.6

1.4

1.2

0.8

0.6

decay rate (efoldings/(2 )1 sec.)

0.4

0.2

Figure 19: Eigenvalues of a minor third A4 ] bell, measured by Slaymaker

and Meeker 18], as given in Roozen{Kroon 16, Table 5.3.1] The grid lines
show the positions of the frequencies corresponding to a minor third chord
at 456:8 Hz, together with two octaves above the fundamental and one below.
All six of these modes are closely matched by eigenvalues of the bell, a tribute
to how far bell design has evolved over the centuries to achieve a musical
eect. The eigenvalue picture for an unmusical bell, such as one worn by a
cow, would look utterly dierent.
was supported in part by NSF Grant DMS-9500975CS.

References
1] A. H. Benade, On Woodwind Instrument Bores, J. Acoust. Soc. Am.
31 (1959), 137{146.
2] A. H. Benade, Fundamentals of Musical Acoustics, (1977, 2d. Revised
Ed., Dover, New York, 1990).
3] A. H. Benade, Horns, Strings, and Harmony, Anchor Books Doubleday
& Co., New York, 1960.

20

4] R. S. Christian, R. E. Davis, A. Tubis, C. A. Anderson, R. I. Mills, and

T. D. Rossing, Eects of Air Loading on Timpani Membrane Vibrations,
J. Acoust. Soc. Am. 76 (1984), 1336{1345.
5] N. H. Fletcher, Plucked Strings|A Review, J. Catgut Acoust. Soc. 26
(1976), 13{17.
6] H. Fletcher and I. G. Bassett, Some Experiments with the Bass Drum,
J. Acoust. Soc. Am. 64 (1978), 1570{1576.
7] N. H. Fletcher and T. D. Rossing, The Physics of Musical Instruments,
Springer{Verlag, New York, 1991.
8] M. Hancock, The Dynamics of Musical Strings 1], J. Catgut Acoust.
Soc., 2d Ser., 1 (1989), 33{45.
9] J. R. Kuttler and V. G. Sigillita, Eigenvalues of the Laplacian in Two
Dimensions, SIAM Review 26 (1984), 163{193.
10] P. M. Morse, Vibration and Sound, McGraw{Hill, New York, 1948.
11] The New Encyclopaedia Britannica, Macropaedia V. 27 15th ed., Article on Sound, Helen Hemingway Benton, Chicago (1991), 604{627.
12] N. C. Pickering, Physical Properties of Violin Strings, J. Catgut Acoust.
Soc. 44 (1985), 6{8.
13] J. W. S. Rayleigh, The Theory of Sound, V. 1, Volume I, 2nd ed. (1894
reprinted ed. Dover, New York 1945).
14] T. D. Rossing, Acoustics of Percussion Instruments|Part II, The
Physics Teacher 15 (1977), 278{288.
15] T. D. Rossing, The Physics of Kettledrums, Sci. Am. 247 (1982), 172178.
16] P. J. M. Roozen{Kroon, Structural Optimization of Bells, Ph.D. Thesis,
Technische Universiteit Eindhoven, 1992.
17] F. W. Sears, Mechanics, Heat, and Sound, 2nd ed., Addison{Wesley,
Reading, MA, 1958.
18] F. H. Slaymaker and W. F. Meeker, Measurements of the Tonal Characteristics of Carillon Bells, J. Acoust. Soc. Am. 26 (1954), 515{522.

21