University of Durham

Department of Mathematical Sciences

Modelling the Indian Drum

Supervisor:
Dr. Bernard Piette

Author:
James Hagues

April 30, 2010

Abstract
This project studies a number of different drum models and finds the frequencies of their vibrational modes. The standard drum is investigated, along
with the Indian drum called the tabla. A tabla consists of the right-handed,
concentrically loaded drum and the left-handed, eccentrically loaded drum.
Two and three-density models are used to approximate the right-handed
tabla and the three-density model is found to produce eigenfrequencies that
have very good agreement with experimental values. The effect of damping
on the drum is also calculated and a different method for solving the twodensity model is investigated, which can also be used to solve the left-handed
drum.

Acknowledgements
I would like to thank Dr. Bernard Piette for his help and direction during
the writing of this project.
Additionally, I would like to thank Johnny for computer-sitting lengthy
Maple runs and Marty and Hannah for proof-reading the project.

Contents
1 Introduction
1.1 The Tabla . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 The Harmonicity of the Tabla . . . . . . . . . . . . . . . . . . . . . . . . . .

4
4
6

2 The Model of an Ordinary Drum

8
2.1 Solving the wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Frequencies of Vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 The 2-Density Model
14
3.1 Solving the Wave Equation for the 2-Density Model . . . . . . . . . . . . . 14
3.2 Tuning the 2-Density Drum . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.1 Optimizing and k . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4 Modes of Vibration of the Drums and a Comparison of the Models
22
4.1 Graphs of the Vibrational Modes . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2 Comparison of Harmonicity . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5 The 3-Density Model
5.1 Solving the Wave Equation for the 3-Density Model . . . . . . . . .
5.2 Tuning the 3-Density Drum . . . . . . . . . . . . . . . . . . . . . . .
5.2.1 Bringing the Parameter Values Within their Physical Limits
5.2.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3 Graphs of the Vibrational Modes of the 3-Density Model . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

6 A Damping Term

27
27
34
37
38
40
42

7 The Variational Method for the 2-Density Model

44
7.1 The Variational Method for solving the Left-Handed Indian Drum . . . . . 45
7.2 The Eigenfrequencies for the Concentric 2-Density Drum . . . . . . . . . . . 53
8 Conclusion

58

Bibliography

59

A Bessel Functions
A.1 Properties of Bessel functions . . . . . . . . . . . . . .
A.1.1 Identities . . . . . . . . . . . . . . . . . . . . .
A.1.2 Bessel equation form with a solution of Jn (x)
A.2 Bessel functions using Fourier theory . . . . . . . . . .
A.2.1 Deriving the recurrance relations . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

61
62
62
62
63
65

B Bipolar Coordinates
B.1 Properties of Bipolar Coordinates
B.2 Finding the Circle Equations . .
B.3 Deriving the Bipolar Scale Factor
B.4 Modified Bipolar Coordinates . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

C Maple Programs
C.1 Solution and Graphs of the 2-Density Drum
C.2 Optimizing the 2-Density Parameters . . . .
C.3 Optimizing the 3-Density Parameters . . . .
C.4 Minization in the Variational Method . . .
D CD Contents

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

67
67
68
71
73

.
.
.
.

75
75
76
77
79
80

Chapter 1

Introduction
The Tabla is a precussion instrument that originates from northern India and consists of
two drums. The instrument has widespread use within both classical and popular music
and is comprised of a pair of small drums that are played by striking the membrane with
the hand. The tabla is shown in Figure 1.1 .

Figure 1.1: The tabla1 . The right-handed drum is on the left and the
left-handed drum is on the right.

1.1

The Tabla

The precise origins of the tabla are unclear but it is thought to have evolved in the
18th century. The older instruments of the pakhawaj and mridangam are believed to
have influenced its development. These drums have a membrane with properties that are
similar to the tabla.
The two drums that make up the tabla have many common characteristics, but also distinct
differences. The right-handed drum is called the tabla and the left-handed drum is called
the dayan
. Both drums have a membrane that is made from goatskin. Upon the edge of
this membrane an outer halo of skin is placed. This is made from a different skin, commonly
1

Image from http://artdrum.com/

a cows or a buffalos, and is known as the Chati. In addition to this, a loaded region
known as the Syahi, is added to the middle of the membrane [CT]. The special properties
of the tabla result mainly from the contribution of the syahi. It gives the central area of
the membrane a significantly greater radial density than the surrounding membrane. The
syahi consists of a combination of flour, water, and a mixture made from iron and other
substances. It is constructed by adding thin layers of this mixture to the drum membrane
and is crucial for creating the tone quality of the instrument. The successive layers are
initially applied with a constant radius, then the radius of application is gradually reduced
to the smallest radius of application of about a centimeter. This process is repeated a few
times creating a thickness gradient within the syahi [CK]. This can clearly be seen in
Figure 1.2 . The construction process means the difference in thickness between the six or
seven layers within the syahi are much smaller than the difference in thickness between
the edge of the syahi and the membrane. Ramakrishna and Sondhi [RS] give experimental
values (stated later in Section 3.2.1) for the density ratio and the radius ratio between the
two regions of the tabla in normal use.

Figure 1.2: The membrane of the tabla with the syahi and chati
labelled2 . The picture shows that the thickness of the syahi is not constant.

Despite these similarities, the two drums differ in construction. The circular loaded region of the right-handed drum is concentrically placed, whereas the loaded portion of
the left-handed drum has an off-centre position (as seen in Figure 1.1). The shell of the
right-handed drum is constructed from wood, normally rosewood, and is hollow for approximately half its depth. In contrast, the shell of the left-handed drum is commonly
made from a metal, preferably brass or copper, although it is sometimes shaped from wood
or clay. The left-handed drum is also significantly larger than the right-handed drum.
Ramakrishna and Sondhi [RS] list the fundamental frequency of the right-handed tabla as
between 110Hz and 200Hz. Experiments carried out by Kapur [K] find the fundamental
frequency of the tabla used in his experiment to be 172Hz, which corresponds to the F
below middle C. In this work the exact frequency of F3 to 4 significant figures, 174.6 Hz,
will be used. The left-handed tabla is normally tuned to the fourth or fifth below the
right-handed (frequencies 116.5 Hz and 130.8 Hz) [TT]. Tuning of the tabla is achieved by
altering the tension of the membrane. A wooden peg, located between the straps that hold
the membrane and the shell together, can be moved up and down to tighten or slacken
the membrane. Fine tuning is attained by striking the braided ring around the membrane
with a hammer.
2

Image from Wikimedia Commons and edited using Microsoft Powerpoint

The technique used in playing the tabla can greatly affect the sound that is produced.
There are twelve basic tabla strokes involving different methods for striking the membrane
and all have distinct sounds and tone qualities [CKT].
Much of the early scientific work on the Tabla was undertaken by the nobel laureate Sir
Chandrasekhara Venkata Raman. His investigations showed that the overtones of the
tabla were harmonic and degenerate.

1.2

The Harmonicity of the Tabla

If a note is played on an instrument then additional vibrations at higher frequencies are

produced alongside the fundamental frequency (the pitch of the note). The human ear
perceives these as one note instead of hearing the harmonics separately. This constructs the
tone quality of the instrument. If an instrument is perfectly harmonic, the harmonics are
in integer multiples of the fundamental frequency. This can be described in the following
equation where fh is the harmonic frequency and f0 is the fundamental frequency:

fh = nf0

with

n = {1, 2, ...} .

(1.1)

A perfectly harmonic instrument sounds pleasing to the human ear because it replicates
the harmonics that are naturally produced by the human voice. An example of a harmonic
instrument is a violin or a flute. If an instrument is inharmonic, like an ordinary drum,
then the harmonics bear no relation to the integer ratio.
Ramans experiments revealed the harmonic properties of the tabla. He found that a
number of the vibrational modes of the tabla share the same frequency, all of which are
close to perfectly harmonic frequencies. These degenerate modes do not appear in the
ordinary drum. Raman also found that the 6th harmonic and any higher harmonics are
sufficiently damped so that their contribution to the tone of the instrument is insignificant
[IMD]. If these higher harmonics did contribute to the tablas tone then their effect could
only be detrimental to the tuning of the tablas harmonics as this would mean that a
greater number would require tuning.
The damping of the higher harmonics is achieved by the outer halo of skin (Chati). The
ampliltudes of vibration of the higher frequency harmonics are greater towards the edge of
the drum than those of the lower harmonics, relative to the amplitudes of the respective
vibrations near the centre (see Section 4.1). The damping effect of the Chati is therefore
much greater at higher frequencies while lower frequencies are comparatively uninfluenced.
Experiments carried out by Ramakrishna and Sondhi give values to the relative frequencies
of the harmonics for the right-handed tabla (Table 1.1). They state that the drummer
tunes to the second harmonic so this is the frequency that is used to calculate the relative
frequencies. The different modes of vibration are explained in more detail later on (Section
2.2) but are denoted nj , where n is the order and j is the root. As n and j increase, the
frequency of the mode increases. The different modes are excited with varying amplitudes
of vibration depending on where the membrane is struck. A strike near the edge of
the drum will excite the higher modes of vibration more than a strike at the centre.
6

The different strikes that are used by tabla players are called notes. Different notes are
produced by strikes in different locations. Also, a number of strikes involve lightly touching
the membrane with a finger other than the striking finger to prevent some vibrational
modes.

Mode
01
11
02
21
12
31
41
22
03

Frequency ratio
Tabla in
Same Tabla with
normal use
opened shell
1.10
1.03
2.00
2.00
3.00
3.00
3.00
3.00
4.01
4.00
4.03
4.00
5.07
5.03
5.07
5.08
5.11
5.04

Table 1.1: Table showing the relative frequency of the modes of vibration of the right-handed tabla, taken from Rama. and Sondhi [RS]. The
2nd harmonic is defined as 2.00.

The models that will be produced in this work will not take into account the effect of the
air pressure within the shell, so the frequency values from the open shell are more relevant
to the subsequent models. These are also notably closer to the integer ratios.

Chapter 2

The Model of an Ordinary Drum

To mathematically find the allowed frequencies of vibration for a drum, the wave equation
must be solved. The wave equation describes the propagation of a wave over time and
takes the form of a second order partial differential equation.
Initially the wave equation will be solved for an ordinary drum, modelled as a one-density
membrane. The method is described in Music: A Mathematical Offering, written by D.
Benson [B].

2.1

Solving the wave equation

The wave equation is given by

2z
= c2 2 z ,
t2

s
with

c=

T
.

(2.1)

z is the displacement of the medium from its equilibrium position, c is the propagation
speed of the wave and T and are the tension and density of the medium. The membrane
of the drum is taken to be within the region of radius 0 r a and a number of
boundary conditions for the drum are imposed on the equation. The displacement must
be finite within the drum membrane and at the edge of the drum there is no displacement.
Additionally, the displacement is 2 periodic with respect to . These boundary conditions
are stated mathematically as
z(r) R

when

0 r a,

(2.2)

z(r) = 0

when

r = a,

(2.3)

z() = z( + 2n) .

(2.4)

As the boundary conditions in the case of the drum depend upon the radius of the drum
and the angular coordinate polar coordinates are used as the coordinate system (shown

in Figure 2.1). The Laplacian must therefore be of polar coordinate form, resulting in the
wave equation being expressed as
2z
= c2
t2

2 z 1 z
1 2z
+
+
r2
r r r2 2


.

(2.5)

Figure 2.1: The axes of the coordinate system of the drum with radius a

The solution to the wave equation will be in the form of a separable solution z(r, , t) =
f (r) g() h(t). If this is substituted into equation (2.5) then it becomes


1
1
2
f (r) g() htt (t) = c frr (r) g() h(t) + fr (r) g() h(t) + 2 f (r) g () h(t) .
(2.6)
r
r
To separate the variables both sides are now divided by f (r) g() h(t) :
htt (t)
= c2
h(t)

frr (r) 1 fr (r)

1 g ()
+
+ 2
f (r)
r f (r)
r g()


.

(2.7)

The left-hand side of this equation depends only on the variable t and the right-hand side
depends only on the variables r and . As each side of the equation depends on different
independent variables, the value of each side must be independent of all the variables, and
therefore a constant. The choice shall be made to call this constant k1 , giving

htt (t)
= c2
h(t)

frr (r) 1 fr (r)

1 g ()
+
+ 2
f (r)
r f (r)
r g()


= k1 .

(2.8)

The equation that must be solved to find h(t) is therefore

htt (t) = k1 h(t) .

(2.9)

The general solution to this is

h(t) = B1 e
3

k1 t+1

Image created using Microsoft Powerpoint

+ B2 e

k1 t+2

(2.10)

A drum produces a note and therefore oscillates. To create oscillations the system needs
to be underdamped, requiring the condition k1 < 0 to hold, giving k1 = 2 and
h(t) = B1 ei(t+1 ) + B2 ei(t+2 ) .

(2.11)

The result needs to be real, so using the identity ei = cos + i sin and then taking the
real part produces
h(t) = B1 cos(t + 1 ) + B2 cos(t + 2 ) .
This can be rewritten as
h(t) = B1 cos(t + 1 ) + B2 cos(t + 1 3 ) .
Using the trigonometric identity cos(u v) = cos u cos v + sin u sin v allows this to be
written in the form


h(t) = B1 cos(t + 1 ) + B2 cos(t + 1 ) cos(3 ) + sin(t + 1 ) sin(3 ) .
Grouping the terms together and rewriting the constants gives
h(t) = B3 cos(t + 1 ) + B4 sin(t + 1 ) .
The constants can again be restated as


h(t) = B5 b1 cos(t + 1 ) + b2 sin(t + 1 ) ,

where

b21 + b22 = 1 .

Knowing that cos2 + sin2 = 1, this allows h(t) to be given in the form


h(t) = B5 sin(4 ) cos(t + 1 ) + cos(4 ) sin(t + 1 ) .
Using the trigonometric identity sin(u + v) = sin u cos v + cos u sin v, this can then be
rewritten
h(t) = B5 sin(t + 1 + 4 ) ,
which then can finally be written as
h(t) = B sin(t + ) .

(2.12)

This now gives the equation for z as z(r, , t) = f (r) g(r) B sin(t + ). The next variable
to solve is .
tt (t)
The hh(t)
equality is discarded and the last two equalities from equation (2.8) are rearranged to give


1 fr (r)
c2 g ()
2 frr (r)
c
+
+ 2 = 2
.
f (r)
r f (r)
r g()

If both sides are multiplied by

r2

r2
c2

then the resulting equation is

frr (r)
fr (r) 2 r2
g ()
+r
+ 2 =
.
f (r)
f (r)
c
g()

10

(2.13)

Now with independent variables on separate sides of the equation, both sides can again
be set equal to a constant, this time called k2 :
r2

fr (r) 2 r2
g ()
frr (r)
+r
+ 2 =
= k2 .
f (r)
f (r)
c
g()

(2.14)

This gives an equation for g() that is identical to the equation (2.9) for h(t) :
g () = k2 g() .
This can be be solved in the same manner as equation (2.9) using the 2 periodicity of
instead of the requirement of oscillations. This method produces the equation for g() ,
giving
g() = C sin(n + ) ,

(2.15)

where n = k2 Z . To satisfy the 2 periodicity (the 3rd boundary condition, equation

(2.4)) n must be an integer.
The equation from (2.14) that is now left to solve is
r2

frr (r)
fr (r) 2 r2
+r
+ 2 = n2 .
f (r)
f (r)
c

This is multiplied by f (r) and rearranged as

 2

2
2
2
r frr (r) + rfr (r) +
r n f (r) = 0 .
c2

(2.16)

The solution
is
given in Appendix 1
as equation (A.9), and is a linear combination

to this r
r
and
Y
will tend to infinity as r tends to zero. This
of Jn r
n
c
c . However, Yn c
produces a singularity at r = 0 , which contradicts the boundary condition given by
equation (2.2). Therefore this part of the general solution is discarded. The z dependence
on r is now written

f (r) = Dn Jn

 r 
c

(2.17)

It is noted that the amplitude Dn depends on the order n of the Bessel function. Combining
all the components (2.12), (2.15) and (2.17) gives the solution to the original wave equation
(2.1) :

z = An Jn

 r 
c

sin(t + ) sin(n + ) ,

(2.18)

where An = B C Dn and is the amplitude of the vibration for the particular mode of
vibration.
11

Equation (2.18) describes the displacement for the mode of vibration at each position on
the drum membrane for the allowed frequencies. To find the permitted frequencies the
final boundary condition, equation (2.3) , is used. If this condition is used in equation
(2.18) , the following equation is obtained:
 a 
Jn
= 0.
(2.19)
c
Taking a and c as constants, the substitution can be made that x = a = c a. As a and c
are constant the ratios of the different values of x are equal to the ratios of the different
values. The equation then simply becomes the Bessel function set equal to zero:
Jn (xnj ) = 0 .

(2.20)

When the equation is solved it will give the allowed ratios of the different frequencies of
the drum, ie. the eigenfrequencies. If equation (2.18) is written without time-dependence
then it becomes

nj = Anj Jn (nj r) sin(n + ) ,

(2.21)

where nj is restricted by the solutions of equation (2.20).

2.2

Frequencies of Vibration

Equation (2.20) is solved using Maple 12 software and the different solutions are labelled
in the following way. The root is denoted xnj , where the indices refer to the particular
solution. n is identical to the n from the above equations and indicates the order of the
Bessel function. j refers to the root number of this solution subset. The first root greater
than 0 for the given n is labelled j = 1 , the second root is labelled j = 2 , and so on.
To show more clearly the frequency ratios of the harmonics, every root is divided by the
x01 root (the solution relating to the fundamental frequency), hence defining the x01 root
as equal to 1. To comply with later notation and the notation of existing literature, the
relative frequency of the xnj root will be listed as the relative frequency of the nj mode
of vibration.
The vibrational modes with a frequency up to 5.5 times the fundamental frequency are
listed in the following table (Table 2.1). The work by Raman, described in the introduction,
found the harmonics of the tabla above the fifth harmonic of insignificant amplitude due
to damping effects. Although this damping is not present on the ordinary drum, this is
the harmonic chosen as the last entry of this table. This is, in part, due to the purpose of
this table being a comparison with the equivalent tables for the tabla models but also, as
there are infinitely many solutions, a place to stop needs to be chosen at some point.
Note that the 02 root is incorrrectly listed as having a frequency ratio of 2.40 in both
the Ramakrishna and Sondhi paper [RS] and the recent Gaudet et al. paper [GGL]. This
result has been double checked.
The graphs of these modes of vibration are shown in Section 4.1 .
12

Mode Frequency ratio Mode Frequency ratio

01
1.000
51
3.647
11
1.593
32
4.056
21
2.136
61
4.132
02
2.295
13
4.230
31
2.653
42
4.601
12
2.917
23
4.832
41
3.155
04
4.903
22
3.500
52
5.131
03
3.598
33
5.412
Table 2.1: Relative frequencies of the vibrational modes of the 1density drum membrane

13

Chapter 3

The 2-Density Model

The model used here was first layed out in Ramakrishna and Sondhis 1954 paper, Vibrations of Indian Musical Drums Regarded as Composite Membranes [RS]. It models the
right-handed tabla which is approximated to a drum membrane that has two regions of
different density; the density of each region is constant. The syahi of the tabla is not of
constant density so this model will not fully reproduce the vibrations of the tabla. This
model also does not account for the effect of air pressure within the shell of the drum.

3.1

Solving the Wave Equation for the 2-Density Model

The radius of the inner region of the drum is a and the radius of the outer region is b (see
Figure 3.1). The right-handed drum will be solved here and the inner section (0 r a)
shall have a density = 1 , and the outer section (a r b) a density = 2 .
As in the 1-density case, the solution of the wave equation, extended to two dimensions,
is required for a model of the drum:
2z
= c2 2 z .
t2

(3.1)

The method for solving the wave equation is again by separation of variables, and to
separate the variable t, z can be written in the form z(r, , t) = (r, )h(t), allowing the
wave equation to be written as
(r, )h(t)tt = c2 h(t)2 (r, ) .

(3.2)

The equation is divided by z(r, , t). This produces independent variables on each side
and allows both sides of the equation to be set equal to the constant 2 . The equation
then becomes

14

htt
2
= c2
= 2 .
h

(3.3)

The same method used in the 1-density case solves the function h(t) and again gives
h(t) = B sin(t + ). The wave equation is now time-independent and written using the
function of (r, ). It is rearranged to give

2 +

2
= 0,
c2

(3.4)

where, as before, c2 = T . The value of c is different for the different regions of the drum
as the density differs. The tension is uniform across both regions because the loading is
applied to the surface of the membrane. These different values define i as
r
1

1 =
=
when
0 r a,
c1
T
r
2

2 =
when
a r b.
=
c2
T
Two different equations are now written for the two differents regions of the drum:
2 1 + 21 1 = 0

0 r a,

(3.5)

2 2 + 22 2 = 0

a r b.

(3.6)

The regions and their densities are shown in Figure 3.1 .

Figure 3.1: The regions of the 2 density model4

Next boundary conditions are imposed on the equations. The edge of the membrane is
fixed, so the displacement must equal zero at the edge of the drum. Also, where the
inner and outer sections of the membrane meet the function and its gradient must be
equal for both regions. As the tension is constant for both sections the gradients are not
4

Image from http://bobbysingh.com.au and edited using Microsoft Powerpoint

15

stress-weighted. Additionally the function is 2 periodic and must be constant within the
region of the drum, giving the boundary conditions:

(r, ) R

when

0 r b,

2 (b, ) = 0 ,
1 (a, ) = 2 (a, ) ,

1 (a, ) =
2 (a, ) ,
r
r
i (r, ) = i (r, + 2n) .

(3.7)
(3.8)
(3.9)
(3.10)
(3.11)

The equations (3.5) and (3.6) can then be solved in the same manner as the ordinary drum
but, unlike the normal drum, the solution to 2 must include the Bessel function of the
second kind because 2 does not exist at r = 0, so will not produce a singularity there.
This gives the solutions
1 (r, ) = An Jn (1 r) sin(n + n ) ,

(3.12)

2 (r, ) = [Bn Jn (2 r) + Cn Yn (2 r)] sin(n + n ) .

(3.13)

These equations must be altered so that they have equal displacement and gradient where
they meet. The frequencies of the vibrational modes can then be determined by solving
the equation at the edge of the drum.
If the solutions to the wave equation for the different regions (3.12 and 3.13) are entered
into the boundary conditions (3.8) and (3.9) they become:

Bn Jn (2 b) + Cn Yn (2 b) = 0 ,

(3.14)

An Jn (1 a) = Bn Jn (2 a) + Cn Yn (2 a) .

(3.15)

Entered into equation (3.10) the two halves of the equation for the gradient are:




n
1 (a, ) = 1
An Jn (1 a) sin(n + n ) + An Jn1 (1 a) sin(n + n ) , (3.16)
r
1 a


n
2 (a, ) = 2
Bn Jn (2 a) sin(n + n ) + Bn Jn1 (2 a) sin(n + n )
r
2 a

n

Cn Yn (2 a) sin(n + n ) + Cn Yn1 (2 a) sin(n + n ) .

2 a

(3.17)

These two are set equal to one another to satisfy (3.10). The sin(n + n ) term cancels

16

out to give



n
n
1
An Jn (1 a) + An Jn1 (1 a) = 2
Bn Jn (2 a) + Bn Jn1 (2 a)
1 a
2 a

n

Cn Yn (2 a) + Cn Yn1 (2 a) .
2 a
(3.18)
The next stage is to find an equation which is independent of all the coefficients so that it
can be solved to find the allowed frequency ratios. First (3.18) is divided by (3.15), which
cancels An out and now leaves the equation independent of one of the coefficients. This
equation is
1 Jn1 (1 a)
Bn Jn1 (2 a) + Cn Yn1 (2 a)
=
.
2 Jn (1 a)
Bn Jn (2 a) + Cn Yn (2 a)

(3.19)

Dividing both the numerator and denominator of the right hand side by Cn produces
1 Jn1 (1 a)
=
2 Jn (1 a)

Bn
Cn Jn1 (2 a)
Bn
Cn Jn (2 a)

+ Yn1 (2 a)
+ Yn (2 a)

(3.20)

To eliminate Bn and Cn from this equation, equation (3.14) is rearranged to the form
Yn (2 b)
Bn
Bn
Cn = Jn (2 b) , which is the substituted in for Cn . The equation becomes
(2 b)
YJnn (
Jn1 (2 a) + Yn1 (2 a)
1 Jn1 (1 a)
2 b)
=
.
Y
(
b)
n
2 Jn (1 a)
Jn (22 b)
Jn (2 a) + Yn (2 a)

Rearranging this by multiplying the right by

Jn (2 b)
Jn (2 b

(3.21)

produces

Yn (2 b)Jn1 (2 a) Jn (2 b)Yn1 (2 a)
1 Jn1 (1 a)
=
.
2 Jn (1 a)
Yn (2 b)Jn (2 a) Jn (2 b)Yn (2 a)

(3.22)

There is now a change to one variable that depends on the frequency, for which the
equation can be solved. The quantities are defined as
b
,
c2
1
2
2 =
= 12 ,
2
2
a
k= .
b
x = 2 b =

It is noted that 2 = 21 is a fixed ratio equal to the ratio of densities and k is a fixed ratio
2
equal to the ratio of radii.
17

The new quantities are now subsituted into (3.22) and this gives an equation that can be
numerically solved for x with a given 2 and k. This is the equation whose solutions are
the frequency eigenvalues:

Jn1 (kx)
Yn (x)Jn1 (kx) Jn (x)Yn1 (kx)
=
.
Jn (kx)
Yn (x)Jn (kx) Jn (x)Yn (kx)

(3.23)

It is noted that this equation is undefined when k = 1 but the 1-density case can still be
solved by using = 1. When equation (3.23) is solved for x it will provide the permitted
relative frequencies of the modes of vibration. Provided the values for b (the
q radius of the

whole membrane) and c2 , which can be calculated from the equation c = T2 , are known
the frequency of the vibration can be found. Alternatively, knowing the values of b and
2 and the desired frequency, the solution can tell the tabla player the tension to tighten
the drum to to achieve that frequency.
There will be more than one solution to this equation for a given n, and k, so the jth
root of equation (3.23) will be denoted xnj .
To get the equations of displacement of the two regions for each distinct xnj there will
be different ratios of Anj , Bnj and Cnj . Just one coefficient is wanted, Anj , which will
be the amplitude of the system. To obtain Bnj in terms of Anj the equation (3.14),
B Jn (xnj )
Cnj = nj
Yn (xnj ) , is substituted into equation (3.15), getting

Anj Jn (kxnj ) = Bnj Jn (kxnj )

Bnj Jn (xnj )
Yn (kxnj ) .
Yn (xnj )

(3.24)

Rearranging to make Bnj the subject, this becomes

Bnj = Anj

Jn (kxnj )Yn (xnj )

.
Jn (kxnj )Yn (xnj ) Jn (xnj )Yn (kxnj )
C

Y (x

(3.25)

n nj
Similarly substituting from equation (3.14) Bnj = nj
into equation (3.15), gives
Jn (xnj )
Cnj as
Jn (kxnj )Jn (xnj )
.
(3.26)
Cnj = Anj
Jn (kxnj )Yn (xnj ) Jn (xnj )Yn (kxnj )

The solutions of for the two different regions are now


1nj (r, ) = Anj Jn

2nj (r, ) = Anj


kxnj
r sin(n + n ) ,
a

Jn (kxnj )
Jn (kxnj )Yn (xnj ) Jn (xnj )Yn (kxnj )
h
x

x
i
nj
nj
Yn (xnj )Jn
r Jn (xnj )Yn
r sin(n + n ) .
b
b
18

(3.27)

(3.28)

3.2

Tuning the 2-Density Drum

The harmonics frequencies of the tabla that are relevant to the 2-density model are the
relative frequencies of the open wooden shell from Table 1.1. The model doesnt take into
account the air pressure, so the values for the tabla in normal use are not as applicable to
this model. The frequencies for the open shell tabla are very close to perfect harmonicity
so Ramakrishna and Sondhi, whose values depart much further from perfect harmonicity
than the real world values, chose to tune their membrane towards perfect harmonicity
rather than the observed values of tabla. This is the goal towards which tabla-makers
have evolved the tabla and it is assumed that this would be the tabla-makers ideal for
their constructed tabla. The 2-density model also has errors greater than the error between
perfect harmonicity and the experimental values. The approach of [RS] is therefore the
approach taken here, with the error calculated as departing from the perfect harmonics
rather than the observed frequencies.

3.2.1

Optimizing and k

There now exists a set of solutions to the eigenvalue equation that are denoted xnj . Each
solution is related to the frequency by the equation
r
2
xnj = 2f b
.
(3.29)
T
b, 2 and T are constants so there is a linear relationship between xnj and the frequency.
For the drum to be perfectly in tune all the harmonics have to be integer ratios of the
x
fundamental frequency ie. xnj
= 1, 2, 3... .
01
To get different solutions of xnj , and k can be varied and as this is done the proximity
x
of the xnj
values to the integers changes. The harmonic error is a calculation of how far
01
the harmonics are from the integer multiples. The error shall be calculated from the error
defined by Gaudet et al. [GGL]. This weights the lower harmonics to mirror the increased
amplitude of these harmonics in the tabla and is given by

Error =

5 X
D
X

xnj
x01

h
h

h=2 d=1

!2
,

(3.30)

where h is the integer value of the harmonic that xnj

is closest to and D is the number of
01
degenerate modes. Harmonics h 6 are ignored due to the damping effect of the outer
halo. This error equation was not used by Ramakrishna and Sondhi who set k = 0.4 and
arrived at their value for by graphical methods.
Starting with the values given by Ramakrishna and Sondhi [RS], Maple 12 is used to vary
and k to 3 significant figures until the minimum error was reached (program C.2). This
optimizes and k to produce the lowest harmonic error within the model, assuming there
isnt a local minimum between the [RS] values and the global minimum. It is recognised
that in practice it would be very difficult for tabla makers to achieve the three significant
figure level of precision using their construction techniques for the tabla (changing k by
19

the third significant figure is a change of approximately 0.1mm). This level of precision is
included as it can still affect the harmonic ratios so that the frequency changes are within
the limit of the human ear ( 4Hz).
x

nj represents the modes of vibration relating to xnj

. The optimal values in this case were
01
found to be = 3.15 and k = 0.401, with an error of 0.0044226 produced. The values of
and k found by [RS] for real-world tablas are
3 < < 4 ,

(3.31)

0.45 <k < 0.55 ,

(3.32)

meaning this value of k lies well outside the range. For a completely accurate model of
the right-handed tabla, the value of k must fall within this range. A comparison of the
frequencies of the different vibrational modes for the above values of k and are shown
in Table 3.1 .

Mode

Frequency ratio
Rama. and Sondhi Maple 12 optimization
= 3.125 k = 0.4
= 3.15 k = 0.401
1.00
1.00
1.94
1.94
2.95
2.94
3.05
3.05
3.97
3.96
4.10
4.10
4.82
4.84
4.97
4.96
5.14
5.16
0.0044500
0.0044226

01
11
21
02
31
12
03
41
22
Error
Table 3.1: Comparison of frequency ratios of harmonics by varying
and k, with 03 as the 5th harmonic

From Table 3.1 it can be seen that the 2-density model has good agreement with perfect
harmonicity. As the tabla is approximately harmonic the model also has good agreement
with the tabla.
Gaudet, Gauthier and Leger [GGL] found that if 03 was set to be the 6th rather than 5th
harmonic then the error was significantly reduced. They also include the 13 harmonic as
part of their analysis as it then falls close to the 5th harmonic.
This departs even further from the real-world than the idealized membrane used by Ramakrishna and Sondhi. If the model is trying only to emulate the real-world tabla then
this could not be used as a technique for lowering the error as the 03 harmonic has been
experimentally found to fall close to the 5th harmonic. The significant lowering of the
error by this change raises interesting questions about the evolution of the drum. The
drum must have evolved through a stage similar to the two-density model and better
harmonicity at this stage would be achieved by a 6th harmonic for the 03 model.
Again and k have been varied to 3 significant figures. The optimal values in this case
20

were found to be = 2.84 and k = 0.385, with an error of 0.0018872 produced. The
frequency eigenvalues for this result are shown in Table 3.2. With the placing of 03
at the 6th harmonic the values for both and k depart further from their experimental
values.

Mode

Frequency ratio
Gaud., Gauth. and Leger Maple 12 Optimization
= 2.9 k = 0.38
= 2.84 k = 0.385
1.00
1.00
1.96
1.95
3.00
2.98
3.05
3.04
4.02
3.99
4.05
4.03
4.95
4.91
5.09
5.06
5.14
5.12
5.99
5.96
0.0021392
0.0018872

01
11
21
02
12
31
22
41
13
03 *
Error
Table 3.2: Comparison of frequency ratios of harmonics by varying and k,
with 03 as the 6th harmonic, therefore not included in error calculation

The difference between the errors of placing 03 at the two different harmonics shows
the greatly superior harmonicity of the 03 frequency when near the 6th harmonic. This
shows that for the idealized 2-density membrane greater harmonicity can be achieved with
a model that departs from the tabla.
Using the values of = 3.15 and k = 0.401 the tension in the membrane required to
produce a certain frequency can be found. The fundamental frequency x01 = 0.998 is used
in the equation. Rearranging equation (3.29) gives
T = 4 2 f 2

b2 2
x201

(3.33)

A typical diameter and pitch of the right-handed tabla are 5.5" (radius of b=0.06985 m)1
and F3 (f = 174.6 Hz) is used. Using equation (3.33) and the density (0.24 kg m2 ),
which was given for a synthetic drum membrane by Howle and Trefethen [HT], equates
to a required tension of 1415 N m1 . As the density of the tabla membrane is likely
to be greater than a synthetic drum membrane and the tightening mechanisms arent as
powerful this appears reasonable when compared to their value of tension for a kettledrum:
T=4415 N m1 .

kksongs.org/tabla/chapter34.html

21

Chapter 4

Modes of Vibration of the Drums

and a Comparison of the Models
The solutions to two different drums have now been calculated and, from these solutions,
graphs showing their vibrational characteristics can be created. A comparison between
the eigenfrequency ratios and the harmonicity of the drums can also be made.

4.1

Graphs of the Vibrational Modes

Using the displacement equation (2.21) for the 1-density drum and equations (3.27) and
(3.28) for the 2-density model with = 3.14 and k = 0.401, a time-independent graph for
each mode of vibration can be drawn. These graphs show the displacement at maximum
amplitude and are rotated to provide the best view rather than having a particular orientation for . The graphs can be thought of as the actual membrane of the drum with
an exaggerated displacement. The edge of the graph represents the edge of the drum and
the surface of the graph represents the membrane of the drum at a moment in time.
The values of n and j are physically evident as n is equal to the number of diameter lines
that have a time-independent displacement of zero and j is equal to the number of circles
of constant r, including the edge of the drum, that have a time-independent displacement
of zero. The central high density region causes the 2-density model to have a different
profile of vibration than the 1-density drum. The graphs are notably different even though
the same restrictions due to n and j apply. The graphs are plotted by Maple 12 (using
program C.1).

22

(a) 1-density

(b) 2-density

Figure 4.1: 01 (r, ) for one-density and two-density concentric drum

(a) 1-density

(b) 2-density

Figure 4.2: 02 (r, ) for one-density and two-density concentric drum

(a) 1-density

(b) 2-density

Figure 4.3: 03 (r, ) for one-density and two-density concentric drum

23

(a) 1-density

(b) 2-density

Figure 4.4: 11 (r, ) for one-density and two-density concentric drum

(a) 1-density

(b) 2-density

Figure 4.5: 12 (r, ) for one-density and two-density concentric drum

(a) 1-density

(b) 2-density

Figure 4.6: 21 (r, ) for one-density and two-density concentric drum

24

(a) 1-density

(b) 2-density

Figure 4.7: 22 (r, ) for one-density and two-density concentric drum

(a) 1-density

(b) 2-density

Figure 4.8: 31 (r, ) for one-density and two-density concentric drum

(a) 1-density

(b) 2-density

Figure 4.9: 41 (r, ) for one-density and two-density concentric drum

25

4.2

Comparison of Harmonicity

This section compares the eigenfrequencies from Tables 2.1 and 3.2 . The allowed frequencies of the ordinary drum bear no relation to the frequencies of the harmonics creating
a completely inharmonic instrument. In contrast to this, if a constant region of higher
density is added to the centre of the drum then a model with good levels of harmonicity
can be created. In addition to the absense of any relation between the harmonics and
the eigenfrequency ratios, the 1-density drum also has a much greater number of allowed
frequencies between the fundamental frequency and the 5th harmonic. These differences
can clearly be seen in Figure 4.10 .

Figure 4.10: The relative frequencies of the harmonics belonging to

the 1-density and 2-density models, compared with ideal harmonicity

From the enclosed CD it can be heard that these sets of harmonics create effects that
are audibly very different from each other. The harmonics of the ordinary drum have no
discernable tuning and as expected produce a sound that bears relation to a normal drum.
The harmonics of the 2-density model produce a note that is audibly close to a harmonic
note, although there is clearly a level of tuning error.

26

Chapter 5

The 3-Density Model

A 3-density model is a logical step to improve upon the harmonicity of the 2-density model
. This also mirrors the construction of the tabla whose loaded portion has an increase in
thickness towards its centre. After an abrupt change in radial density between the edge
of the syahi and the outer membrane the differences are then less pronounced within the
syahi. The 3-density model should better replicate the real-world tabla and reduce the
harmonic error. The relative densities of the inner two regions are expected to be closer
to equality than the relative densities of the outer two regions.

5.1

Solving the Wave Equation for the 3-Density Model

Using the same method as was used to solve the right-handed tabla, the wave equation of
a drum with three different densities is now solved. The inner-most section ( 0 r a )
shall have a density = 1 , the middle section ( a r b ) shall have = 2 and the
outer section ( b r l ) shall have a density = 3 , l being the edge of the drum.
Again the two-dimensional wave equation is started with:
2z
= c2 2 z .
t2

(5.1)

As before, to use a separation of variables method z(r, , t) is written in the form z(r, , t) =
(r, )h(t), allowing the wave equation to be written as
(r, )h(t)tt = c2 h(t)2 (r, )) .

(5.2)

The equation is divided by z(r, , t) = (r, )h(t). This produces independent variables
on each side and allows both sides of the equation to be set equal to the constant 2 .
The equation then becomes

27

htt
2
= c2
= 2 .
h

(5.3)

The solution to h(t) is the same as the previous cases: h(t) = sin(t + n ).
Rearranging the right two equivalences of equation (5.3) produces the time-independent
equation

2 +
where the value of c2 =

T
.

2
= 0,
c2

(5.4)

The next step is to define for each region of the drum.

r
1

=
1 =
c1
T
r
2

=
2 =
c2
T
r

3
3 =
=
c3
T

when

0 r a,

when

a r b,

when

b r l.

Now the equations for each region of the drum can be written in the form
2 1 + 21 1 = 0

0 r a,

(5.5)

2 2 + 22 2 = 0

a r b,

(5.6)

3 3 + 23 3 = 0

b r l.

(5.7)

These regions and their densities are shown in Figure 5.1 .

28

Figure 5.1: The regions of the 3 density model5

As before, the next step is to impose boundary conditions on the equations. The same
boundary conditions as the 2-density model are used but with the addition of two more.
There are now two different radii, a and b, where different sections of the membrane meet
and at both of these the function and its gradient must be equal in the adjacent regions.

(r, ) R

when

0 r l,

3 (l, ) = 0 ,
1 (a, ) = 2 (a, ) ,
2 (b, ) = 3 (b, ) ,

1 (a, ) =
2 (a, ) ,
r
r

2 (b, ) =
3 (b, ) ,
r
r
(r, ) = (r, + 2n) .

(5.8)
(5.9)
(5.10)
(5.11)
(5.12)
(5.13)
(5.14)

Equations (5.5), (5.6) and (5.7) can then be solved in the same manner as the normal
drum but, unlike the normal drum, the solutions to 2 and now 3 must include the
Bessel function of the second kind because the domains of both 2 and 2 do not include
r = 0, so will not produce a singularity. This property and the solution of the wave
equation for each region gives the general solutions

1 (r, ) = An Jn (1 r) sin(n + n ) ,

(5.15)

2 (r, ) = [Bn Jn (2 r) + Cn Yn (2 r)] sin(n + n ) ,

(5.16)

Image from http://bobbysingh.com.au and edited using Microsoft Powerpoint

29

3 (r, ) = [Dn Jn (3 r) + En Yn (3 r)] sin(n + n ) .

(5.17)

If these results are entered into the boundary conditions (5.9), (5.10), (5.11), (5.12) and
(5.13), then (5.9), (5.10) and (5.11) become

Dn Jn (3 l) + En Yn (3 l) = 0 ,

(5.18)

An Jn (1 a) = Bn Jn (2 a) + Cn Yn (2 a) ,

(5.19)

Bn Jn (2 b) + Cn Yn (2 b) = Dn Jn (3 b) + En Yn (3 b)

(5.20)

and the two sides of equation (5.12) become




n
1 (a, ) = 1
An Jn (1 a) sin(n + n ) + An Jn1 (1 a) sin(n + n ) , (5.21)
r
1 a

n

2 (a, ) = 2
Bn Jn (2 a) sin(n + n ) + Bn Jn1 (2 a) sin(n + n )
r
2 a

n
Cn Yn (2 a) sin(n + n ) + Cn Yn1 (2 a) sin(n + n ) . (5.22)

2 a
These two are put equal to one another to satisfy (5.12), with sin(n + n ) cancelling:



n
n
1
An Jn (1 a) + An Jn1 (1 a) = 2
Bn Jn (2 a) + Bn Jn1 (2 a)
1 a
2 a

n

Cn Yn (2 a) + Cn Yn1 (2 a) .
2 a
(5.23)
Similarly, differentiating 2 and 3 and setting them equal to each other at b the boundary
condition (5.13) is observed producing


n
n
2
Bn Jn (2 b) + Bn Jn1 (2 b)
Cn Yn (2 b) + Cn Yn1 (2 b) =
2 b
2 b


n
n
3
Dn Jn (3 b) + Dn Jn1 (3 b)
En Yn (3 b) + En Yn1 (3 b) .
(5.24)
3 b
3 b
The next step is find an equation which is independent of all the coefficients which will
Dn
n
provide the frequency eigenvalues when solved. First B
Cn and En are needed in a form
independent of the coefficients. Rearranging (5.18) gives
Dn
Yn (3 l)
=
.
En
Jn (3 l)
To get

Bn
Cn

(5.25)

equation (5.23) is divided by (5.19)

1 Jn1 (1 a)
Bn Jn1 (2 a) + Cn Yn1 (2 a)
=
.
2 Jn (1 a)
Bn Jn (2 a) + Cn Yn (2 a)
30

(5.26)

Dividing both the numerator and denominator of the right-hand side by Cn produces
1 Jn1 (1 a)
=
2 Jn (1 a)

Bn
Cn Jn1 (2 a)
Bn
Cn Jn (2 a)

+ Yn1 (2 a)
+ Yn (2 a)

(5.27)

Rearranging this produces the equation

Yn1 (2 a)Jn (1 a) 21 Jn1 (1 a)Yn (2 a)
Bn
= 1
.
Cn
2 Jn1 (1 a)Jn (2 a) Jn1 (2 a)Jn (1 a)

(5.28)

This equation shall be defined as (n, 1 , 2 , a)

(n, 1 , 2 , a)

Yn1 (2 a)Jn (1 a)

1
2 Jn1 (1 a)Yn (2 a)

1
2 Jn1 (1 a)Jn (2 a)

Jn1 (2 a)Jn (1 a)

(5.29)

Now the two unused boundary conditions are used. Equation (5.24) is divided by (5.20)
to give
2 Bn Jn1 (2 b) + Cn Yn1 (2 b)
Dn Jn1 (3 b) + En Yn1 (3 b)
=
.
3 Bn Jn (2 b) + Cn Yn (2 b)
Dn Jn (3 b) + En Yn (3 b)

(5.30)

The numerator and denominator of the left-hand side of the equation are divided by Cn
and the numerator and denominator of the right-hand side of the equation are divided by
En to produce
2
3

Bn
Cn Jn1 (2 b)
Bn
Cn Jn (2 b)

+ Yn1 (2 b)
+ Yn (2 b)

Dn
En Jn1 (3 b)
Dn
En Jn (3 b)

+ Yn1 (3 b)
+ Yn (3 b)

(5.31)

Substituting in equations (5.25) and (5.28) to this equation and multiplying the right by
Jn (3 c)
Jn (3 c) then gives

2 (n, 1 , 2 , a)Jn1 (2 b) + Yn1 (2 b)

Yn (3 l)Jn1 (3 b) Jn (3 l)Yn1 (3 b)
=
. (5.32)
3 (n, 1 , 2 , a)Jn (2 b) + Yn (2 b)
Yn (3 l)Jn (3 b) Jn (3 l)Yn (3 b)
There is now a change to one variable so that the equation can be solved. The new

31

quantities are defined as

y
2
2
q
k

r
l
3
= 3 l =
= l
,
c3
T
c2
2
1
= 21 = 22 ,
=
2
2
c1
2
2

c2
=
= 22 = 23 ,
3
3
c2
a
= ,
b
b
= .
l

It is noted that 2 and 2 are a fixed ratio equal to the ratio of densities and k and q are
equal to the ratio of the radii.
The new quantities are now subsituted into equation (5.32) and this gives an equation that
can numerically solve for a given 2 , 2 , k and q. This is the equation whose solutions are
the frequency eigenvalues:

(n, y)Jn1 (ky) + Yn1 (ky)

Yn (y)Jn1 (ky) Jn (y)Yn1 (ky)
=
,
(n, y)Jn (ky) + Yn (ky)
Yn (y)Jn (ky) Jn (y)Yn (ky)

(5.33)

where
(n, y)

Yn1 (kqy)Jn ( kqy) Jn1 ( kqy)Yn (kqy)

.
Jn1 ( kqy)Jn (kqy) Jn1 (kqy)Jn ( kqy)

(5.34)

Note that the eigenvalue equation needs both k 6= 1 and, from the term, 6= 1 to be
defined.
The 2-density case can be solved in two ways from this 3-density model. If
k q = k2-density
=1
= 2-density
and equation (5.33) is solved then the eigenvalues will be the eigenvalues of the
2-density case. Alternatively, if
= 2-density
k = k2-density
q=1
the 3-density model can still solve the 2-density case. If = 1 then the 1-density
case can also be solved with this method.
The results from the 3-density model for both the 2-density and 1-density cases are
identical to their respective models.
32

Equation (5.33) will be solved for y, which can in turn be used to find either w, l, 3 or T if
the others are known. If this equation is defined there will be an infinite number solutions
for a given , , k, q and each n. As in the previous cases the jth root of equation (5.33)
will be ynj where j = 1 is the first root greater than 0.
For each distinct ynj there will be different ratios of Anj , Bnj , Cnj , Dnj and Enj . Just
one coefficient, Anj , is wanted which will be the amplitude of the system. Bnj , Cnj , Dnj
and Enj are now found in terms of Anj .
To start with Bnj and Cnj will be obtained in terms of Anj . The equation (5.28) will
be rearranged to get Bnj = (ynj )Cn and substituted into equation (5.19) getting the
equation

Anj Jn ( kqynj ) = (ynj )Cn Jn (kqynj ) + Cn Yn (kqynj ) .

(5.35)

If this is rearranged then the equation provides Cnj in terms of Anj :

Cnj = Anj

Jn ( kqynj )
.
(ynj )Jn (kqynj ) + Yn (kqynj )

(5.36)

Again a substitution is made from equation (5.28) Bnj = (n, y)Cn this time into equation
(5.35). Bnj is solved in terms of Anj to become
Bnj = Anj

(ynj )Jn ( kqynj )

.
(ynj )Jn (kqynj ) + Yn (kqynj )

(5.37)

The next stage is to find Dnj in terms of Anj . The above values of Bnj and Cnj are
J (y )
substituted, along with the rearranged equation (5.18) Enj = Dnj Ynn (ynj
, into equation
nj )
(5.20), getting

Dnj Jn (kynj ) Dnj

Jn (ynj )
(ynj )Jn ( kqynj )
Yn (kynj ) =Anj
Jn (kynj )
Yn (ynj )
(ynj )Jn (kqynj ) + Yn (kqynj )
Jn ( kqynj )
+ Anj
Ynj (kynj ) .
(ynj )Jn (kqynj ) + Yn (kqynj )
(5.38)

This is rearranged to produce the result of Dnj in terms of Anj :

Dnj = Anj

(ynj )Jn (kynj ) + Yn (kynj )

Jn ( kqynj )Yn (ynj )
. (5.39)
(ynj )Jn (kqynj ) + Yn (kqynj ) Jn (kynj )Yn (ynj ) Jn (ynj )Yn (kynj )

Using the same method to find Enj gives

Enj = Anj

(ynj )Jn (kynj ) + Yn (kynj )

Jn ( kqynj )Jn (ynj )
.
(ynj )Jn (kqynj ) + Yn (kqynj ) Yn (kynj )Jn (ynj ) Yn (ynj )Jn (kynj )

The solutions of now only depend on one amplitude, Anj :

33

(5.40)


1 (r, ) = Anj Jn

2 = Anj

3 = Anj

5.2


kqynj
r sin(n + n ) ,
a

(5.41)






Jn ( kqynj )
kynj
kynj
(ynj )Jn
r + Yn
r sin(n + n ) ,
(ynj )Jn (kqynj ) + Yn (kqynj )
b
b
(5.42)
Jn ( kqynj ) [(ynj )Jn (kynj ) + Yn (kynj )]
(ynj )Jn (kqynj ) + Yn (kqynj )

y

Yn (ynj )
nj

Jn
r
Jn (kynj )Yn (ynj ) Jn (ynj )Yn (kynj )
l
y

Jn (ynj )
nj
+
Yn
r sin(n + n ) .
Yn (kynj )Jn (ynj ) Yn (ynj )Jn (kynj )
l

(5.43)

Tuning the 3-Density Drum

The set of solutions to the eigenvalue equation (5.33) are denoted ynj . Each solution is
related to the frequency by the equation
r
3
.
(5.44)
ynj = 2f l
T
l, 3 and T are constants so there is a linear relationship between ynj and the frequency.
As in the 2-density case, for the drum to be perfectly in tune all the harmonics have to
y
be integer multiples of the fundamental frequency, ie. ynj
= 1, 2, 3....
01
Varying , , k and q will produce different solutions of ynj and as these are varied the
y
proximity of the values ynj
to the integers also changes. The objective is to get the
01
harmonics as close to the integers as possible and, as before, this is achieved with the
error equation, which is defined as

Harmonic Error =


5 X
D 
X
ynj /y01 h 2
h=2 d=1

(5.45)

where h is the integer value of the harmonic that nj is closest to and D is the number of
degenerate modes. As in the two-density model harmonics h 6 are ignored.
Minimizing the Harmonic Error produces a significantly greater problem when four, as
opposed to two, constants can be varied. A method of minimizing the error by varying
one constant at a time is impossible due to the erratic nature of the function, caused by
the large number of Bessel function terms on both the numerator and denominator. A
brute force method is used here for minimizing the error. A set of values for each of the
constants , , k and q are input into a Maple 12 program, written to trial every possible
permutation of these sets and return an error value for each set (C.3). The program then
selects the value with the lowest error and returns the error, along with the constants that
produced it. After the program is run with the initial sets of parameters another set of
parameters are input. These are chosen so the parameter that produced the minimum
34

error in the previous program run is the median value of the new run and so the intervals
between the values in each set are reduced. Fine tuning is then done after this to get the
minimum error when the parameters are adjusted to three significant figures.
It is recognised that this method is not guaranteed to locate the global minimum. To do
this, a calculation would be needed with a number of permutations that would be orders of
magnitude greater, requiring a considerably more powerful computer and a large amount
of time available to access it.
After some sampling runs, using the two-density model and the physical tabla as guidelines,
it was decided to do two initial runs. This decision was made based on the limited available
time and computing power, along with the experimentally found values for the tabla. The
first set of parameters were chosen to be

[2.6, 2.7 ... 4.1] [1.01, 1.06 ... 1.36] [0.3, 0.35 ... 0.7] [0.3, 0.35 ... 0.7] ,

(5.46)

and the second set

[1.01, 1.06 ... 1.36] [2.5, 2.6 ... 4.0] [0.3, 0.35 ... 0.7] [0.3, 0.35 ... 0.7] .

(5.47)

After this, runs with smaller intervals, centered around the lowest value of each set, were
made to locate a minimum.
When the 03 mode was placed at the 6th harmonic the optimal parameters were found
to be = 2.81, = 1.06, k = 0.390 and q = 0.339. These parameters produce an error
of 0.0015487 . In this case the minimal error is not greatly reduced from the two-density
model (error = 0.001886) and despite the extra calculations needed the more complex
model produces only a marginally superior result.
When the 03 mode was placed at the 5th harmonic the optimal parameters were found
to be = 4.62, = 1.21, k = 0.456 and q = 0.6132; these parameters produce an error of
0.00070534 . With this configuration the minimal error is significantly reduced from the
two-density model and the 6th harmonic 3-density model, producing an error of 0.0044226.
Despite the simplest model of the tabla (the 2-density model) achieving considerably better
harmonicity with the 03 mode placed at the 6th harmonic, when the more realistic 3density model is used the 03 mode placed at the 5th harmonic produces a significantly
lower error. This explains why the tabla possesses its particular harmonics.
As this method isnt guaranteed to find the global minimum, if different sets of parameters
were chosen then the method may produce a lower error. The relative frequencies for the
different modes of vibration are given in Table 5.1 .
The real tabla is not perfectly harmonic as the harmonics depart slightly from the integer
ratios. To model the tabla the harmonic values of the actual tabla must be reproduced.
As the effect of the air pressure inside the tabla is not modelled the experimental values
with the shell opened out is relevant to this model (Table 1.1). If the error is calculated
by using the distance of the harmonics from the experimentally determined values then
35

2-density drum
Mode
01
11
21
02
12
31
22
41
13
03
Error

= 2.84

k = 0.385

1.00
1.95
2.98
3.04
3.99
4.03
4.91
5.06
5.12
0.0018906

Frequency ratio
3-density drum 03 at 6th
= 2.81 = 1.06
k = 0.390 q = 0.339
1.00
1.96
2.99
2.97
3.99
4.04
4.94
5.07
5.12
0.0015487

3-density drum 03 at 5th

= 4.62 = 1.21
k = 0.456 q = 0.613
1.00
1.96
2.99
2.99
3.99
4.02
5.01
5.05
5.04
0.00070534

Table 5.1: Comparison of error with respect to perfect harmonicity from a 2-density and a 3-density
model

different values for the frequency ratios are produced when the error is minimized. The
03 5th harmonic is used to replicate the properties of the real-world tabla. As in the
experimental results the 2nd harmonic (y11 ) is defined with a frequency ratio of 2.00 .
To give the same significance to the overall error, and acknowledging the fact that the
fundamental frequency has a lower amplitude in practice than the 2nd harmonic, the error
for the 01 root is weighted

Error 01 =

2y01 /y11 1.03

2

2
.

(5.48)

The other roots have the same weighting as before.

The result of the error minimization when the error is taken with respect to the experimental values produces an error of 0.00024740 when the parameters are = 2.74, = 1.14,
k = 0.476 and q = 0.513. The frequency ratios for the modes of vibration are shown below
in Table 5.2 .
Table 5.2 shows that there is excellent agreement between the optimized 3-density model
and the experimental values of the frequency ratios for the tabla. The 3-density model for
the tabla produces results that model the right-handed tabla to a high degree of accuracy.
The results are considerably better, with a significantly lower error, than the 2-density
model and the 3-density model also better replicates the physical properties of the tabla.
If the resulting frequencies for each model are listened to on the enclosed CD then it
can be heard that, with the correct parameters, the 5th harmonic 3-density model creates
overtones that are more harmonic than a 2-density model.

36

Mode

Frequency ratio
3-density drum
Experimental values
= 2.74 = 1.14
k = 0.476 q = 0.513
1.03
1.06
2.00
2.00
3.00
2.99
3.00
3.00
4.00
4.00
4.00
4.03
5.03
5.04
5.04
5.05
5.08
5.08
0
0.00024740

01
11
21
02
12
31
41
03
22
Error
Table 5.2: Comparison of harmonic frequencies with the error calculated with
respect to the harmonics from experimental results

5.2.1

Bringing the Parameter Values Within their Physical Limits

To get a wholly accurate model of the tabla, not only do the harmonics have to match the
harmonics of the tabla but also the experimentally determined values for the radius and
density ratios. The values of these are given by Ramakrishna and Sondhi as 0.45 < k <
0.55 and 3 < s < 4 . k has the same representation in both the 2 and 3-density models
but due to the extra higher density region in the 3-density model the value of s must be
calculated from , and q. s is assumed to be calculated from the average density of the
syahi and is equal to for the 2-density model, but for the 3-density model it is defined
as
s
s2 =
,
(5.49)
3
where s is the average density of the combined inner two regions relative to the outer
membrane. The 2-dimensional density for the syahi s is given by
Pn
i Areai
Mass
s =
= i=1
.
(5.50)
Area
Area
For the inner two regions of the 3-density model this becomes
s =
Using q =

a
b

1 a2 + 2 (b2 a2 )
.
b2

(5.51)

this simplifies to
s = 1 q 2 + 2 2 q 2 .

Factorizing out 2 and with the identity 2 = 12 this gives



1 2
s =
q + 1 q 2 2 ,
2

(5.52)

(5.53)

and then


s = q 2 ( 2 1) + 1 2 .
37

(5.54)

If this is substituted back into the equation (5.49) the result is

Using 2 =

2
3


 2
s2 = q 2 ( 2 1) + 1
.
3

(5.55)



s2 = q 2 ( 2 1) + 1 2 .

(5.56)

this becomes

This equation allows the calculation of s from the parameters of the 3-density model. For
the parameters that produce the minimal error with respect to the experimental values,
s evaluates to 2.85 : outside the real-world values. The optimal parameters within the
allowed values of s are therefore calculated (giving s = 3.01 with = 2.88, = 1.15,
k = 0.480 and q = 0.526). Both values of k fall comfortably within the real world values.

Mode

Experimental values

Frequency ratio
3-density drum
minimal error
= 2.74 = 1.14
k = 0.476 q = 0.513
1.06
2.00
2.99
3.00
4.00
4.03
5.04
5.05
5.08
0.00024740

3-density drum
within s values
= 2.88 = 1.15
k = 0.480 q = 0.526
1.06
2.00
2.99
3.01
4.00
4.03
5.04
5.08
5.08
0.00029252

01
1.03
11
2.00
21
3.00
02
3.00
12
4.00
31
4.00
41
5.03
03
5.04
22
5.08
Error
0
Table 5.3: Comparison of harmonic frequencies with the error calculated with respect to the
harmonics from experimental results, including the values when the parameters are within their
physical limits

Table 5.3 shows that if the parameters are within the real world limits then the agreement
with the experimental values is only marginally decreased. If this restriction is placed
on the parameters then the model still provides good agreement with the experimentally
determined frequency ratios.

5.2.2

Summary

The 3-density model provides a very good method for replicating the eigenfrequency ratios
of the right-handed tabla and can greatly reduce the harmonic error from the 2-density
model. The model can almost exactly reproduce the experimental harmonic frequencies
but the density ratio of the syahi and membrane falls outside the experimental values. It is
noted, however, that the precision of the parameters that produced these results would be
difficult to achieve using the construction techniques of tabla-makers (see Section 3.2.1).
If the density ratio is limited to the experimental values the harmonic error is slightly, but
38

not greatly, increased. It is audibly difficult to discern a difference between these models
and the experimental values (hear enclosed CD).

39

5.3

Graphs of the Vibrational Modes of the 3-Density Model

The vibrational modes for the 3-density model, optimized to produce minimal error with
respect to the experimental values for the tabla ( = 2.74, = 1.14, k = 0.476, q = 0.513)
are shown below.

(a) 01

(b) 02

Figure 5.2: 01 (r, ) and 02 (r, ) for 3-density concentric drum

(a) 03

(b) 11

Figure 5.3: 03 (r, ) and 11 (r, ) for 3-density concentric drum

40

(a) 12

(b) 21

Figure 5.4: 12 (r, ) and 21 (r, ) for 3-density concentric drum

(a) 22

(b) 31

Figure 5.5: 22 (r, ) and 31 (r, ) for 3-density concentric drum

(a) 41

Figure 5.6: 41 (r, ) for 3-density concentric drum

41

Chapter 6

A Damping Term
In a real world system damping affects the movement of the simple harmonic oscillator.
This effect can be included in the model for a drum.
When a force opposes the movement of the oscillating sytem, damped harmonic motion
occurs. In a real world system there can be many contributing factors resisting the motion.
The frictional force produced is linearly proportional to the velocity [FR]:
F = z

(6.1)

where is the damping coefficient and z is the velocity. The minus sign is present as
the force is resisting the motion. If this equation is made equal to Newtons second law

and the acceleration is made the subject then the equation becomes z = m
z.
If this is
included into the 2D wave equation, as used for modelling the tabla, the wave equation
with the damping term is

z = c2 2 z z ,
(6.2)
m
If separation of variables is used as before (setting each side equal to the constant 2 ),
and the damping term is moved to the left, the result is
htt (t)
ht (t)
2 (r, )
+
= c2
= 2 .
h(t)
m h(t)
(r, )

(6.3)

This equation is rearranged to find the solution of h:

htt +

ht + 2 h = 0 .
m

(6.4)

This can be written

+ 2h + 2 h = 0 ,
h

(6.5)

where = 2m
. To solve this equation a solution in the form h = Aet+1 can be taken.
This gives, when differentiated and substituted into equation (6.5),

42

Aet+1 ( 2 + 2 + 2 ) = 0 .
This is solved as a quadratic equation and gives the solutions
p
= 2 2 ,

(6.6)

(6.7)

which can be rewritten as

p
= i 2 2 .

(6.8)

The two roots being referred to as 1 and 2 , this is then put back into h = Aet+1 to
give the time dependent displacement as the general solution
h = Be1 t+1 + Ce2 t+1 .

(6.9)

Substituting in the two roots and factorizing produces




2
2
2
2
h = et Bei( t+2 ) + Cei( t2 ) .

(6.10)

As shown previously in Chapter 2 (2.9) this equation can be written in the form
p
h = Aet cos( 2 2 t + ) .

(6.11)

The angular frequency of this system is therefore

p
d = 2 2 ,

(6.12)

with being the angular frequency of an undamped system. From equation (6.11) it can
be seen that the amplitude of the system decays in the form Aet .
It is noted that for harmonic motion to occur the system must be underdamped. An
overdamped or critically damped system will not oscillate. Therefore 2 2 > 0. If
equation (6.12) is written instead in terms of frequency using = 2f then the result is
r
2
(6.13)
fd = f 2 2 .
4
If this is written as the natural frequency multiplied by a damping term it becomes
s
2
fd = f 1 2 2 .
(6.14)
4 f
An approximate value of was found by varying the damping term in a synthesizing
program and aurally determining when the decay time seemed closest to the sound of a
tabla. The result found as that 6.5. This value is used in equation (6.14), along with
the natural frequency of 174.6 Hz . The damped frequency is then calculated to be 174.597
Hz , which is indistinguishable to the human ear from the natural frequency. It can be
seen from equation (6.14) that the effect on the higher harmonics will be even smaller.
The damping of the membrane will therefore have a negligible effect on the frequency and
no correction needs to be made to the models.
43

Chapter 7

The Variational Method for the

2-Density Model
Here a different method is used to solve the 2-density model of the tabla. This method
also has the potential to find the frequency eigenvalues of the asymmetric left-handed
tabla. The method used for the two and three density models cannot be used for the
asymmetric case as there is no solution to the partial differential equation that results
from the boundary conditions.
The left-handed tabla is of a similar composition to the right-handed tabla, but is loaded in
an eccentric manner. Bipolar coordinates can be used to describe the eccentric loading of
the membrane. The method used here includes the bipolar coordinate system which allows
the asymmetric case to also be solved using the same method. This method of finding the
vibrational modes of the Indian drum was first set out by Sarojini and Rahman [SR].
Using the modified bipolar coordinates and the scale factor which are derived in Appendix
B, along with the results of bipolar coordinates, the following operators result:

r2
4

+ r cos

2

2
1
1 2
=
+
+
2
r2 r r r2 2


2
r

1
4 + r cos
r +
=
,

r
r
2
dS = 
2 rdrd .
r2
+

r
cos

4
2


,

(7.1)
(7.2)
(7.3)

By employing these modified bipolar coordinates, the same wave equation and boundary
conditions can describe the asymmetric tabla.
These are
2 1 + 21 1 = 0

for

0 r a,

(7.4)

2 2 + 22 2 = 0

for

a r b,

(7.5)

44

where
r

1
=
1 =
c1
T
r

2
=
2 =
c2
T

when

0 r a,

when

a r b,

and the boundary conditions are

(r, ) R

0 r b,

when

(7.6)

2 (b, ) = 0 ,

(7.7)

1 (a, ) = 2 (a, ) ,

(7.8)

1 (a, ) =
2 (a, ) ,
r
r
(r, ) = (r, + 2n) .

(7.9)
(7.10)

However, with the modified bipolar scale factor present in the Laplacian, the wave equation
is not separable unless the scale factor is 1, which only happens in the concentric case
when . This means a different method must be used to solve the equation if the
possibility of solving the asymmetric case is to remain open. The method used is the
Variational Method.

7.1

The Variational Method for solving the Left-Handed

Indian Drum

The wave equation for the drum is written in the form of a single equation
2 +
where

(
1 ,
=
2 ,

2
= 0,
T
if
if

0 r < a,
a r b.

(7.11)

(7.12)

If equation (7.11) is rearranged then it produces

2
2
=
.
T

(7.13)

The right-hand side can then be multiplied by

, then both the numerator and denominator integrated with respect to the surface, taken over the whole drum membrane:
R
2 dS
2

= R
.
(7.14)
T
2 dS

It can be shown, using the Divergence Theorem, that the numerator is equal to
Z
Z
2
dS = ()2 dS .
45

(7.15)

This is demonstrated below.

The Divergence Theorem states that
Z
Z
( F) dV = F n dS .

(7.16)

If F = then the divergence theorem gives

Z
Z
( ()) dV = n dS .

(7.17)

Using the identity (A) = A + A the obtained result is

Z
Z
Z
2
dV + dV = n dS .

(7.18)

As
R boundary condition (7.6) states that = 0 on the boundary, this means that
n dS = 0 , leaving

2 dV = 0 .

dV +

(7.19)

This can be rewritten as

Z
Z
2
dV = ()2 dV .

(7.20)

and the scale factor are functions of only two variables (r and ) and dV = dS dz.
This means the additional integration element dz will evaluate to the same constant
on both sides of the equation. The constant then cancels out, leaving
Z
Z
2 dS = ()2 dS .
(7.21)
This shows that equation (7.15) holds. The right-hand side of this can, using equations
(7.2) and (7.3), be written as

!2
Z
Z
r2
+

r
cos

2
4
r +

()2 dS =

2 rdrd .

r
r
r2
+

r
cos

4
(7.22)
The scale factors and orthogonal vectors cancel out when the bracket is squared and leaves
  #
Z
Z " 2

1 2
2
() dS =
+ 2
rdrd .
(7.23)
r
r

The scale factor has been removed and the equation is therefore identical to the orthodox
polar coordinate equation. Equation (7.14) now becomes

2
=
T


R  2
r

1
r2

+
R
2 dS
46

2 

rdrd
.

(7.24)

As in the first method for solving the 2-density drum, dimensionless quantities are now
introduced:
r = bs ,

(7.25)

b
,
c2
1
2
2 =
= 12 ,
2
2
a
k= ,
b
b
=
.
2
x = 2 b =

(7.26)
(7.27)
(7.28)
(7.29)

The scale factor is divided through on both the numerator and denominator by to
produce
1
1
=
.
2r cos
r2
1 2s cos + 2 s2
1
+ 2
2

and the dimensionless quantities are subsituted into equation (7.24) . Together, this gives

R2 Rk h 1
x2
=
b2 2

1 2
s

b2

0 0

R2 Rk
0 0

1
b2 s2

1 2

b2 s dsd +

12
b2 s dsd
(12s cos +2 s2 )2

R2 R1 h 1

2 2
s

b2

0 k
R2 R1

+ 2

0 k

1
b2 s2

2 2

b2 s dsd
.

22
b2 s dsd
(12s cos +2 s2 )2

(7.30)
The b s on the numerator cancel and the left-side b2 cancels with the denominator b s.
The 2 is brought over the right-hand side, producing the equation

R2 Rk h
2

x =

0 0

1 2
s
R2 Rk
0 0

1
s2

1 2

s dsd +

12
s dsd
(12s cos +2 s2 )2

R2 R1 h
0 k
R2 R1

0 k

2 2
s

1
s2

2 2

s dsd
.

(7.31)

22
s dsd
(12s cos +2 s2 )2

The scale factor disappears as tends to zero and the equation becomes the variational
integral for the concentric drum. A binomial expansion is now made on the square of the
scale factor and terms of order 3 or higher are discarded:
1

= 1 + 4s cos 22 s2 + 3(42 s2 cos2 + O(3 )),

(1 2s cos + 2 s2 )2
Using the identity cos 2 = 2 cos2 1 gives
1
= 1 + 4s cos 22 s2 + 3(22 s2 (cos 2 + 1) + O(3 )) .
(1 2s cos + 2 s2 )2
This means the scale factor squared is now
1
(1 2s cos +

2 s2 )2

1 + 4s cos + 42 s2 + 62 s2 cos 2 .

47

(7.32)

As is 2 periodic, it can be written in the form of a Fourier series:

1 (s, ) =
2 (s, ) =

X
n=0

n (s) cos(n) +
n (s) cos(n) +

n=0

X
n=1

n (s) sin(n),

0 s k,

(7.33)

n (s) sin(n),

k s 1.

(7.34)

n=1

The functions n and n can have a singularity at s = 0 , as is the case with Bessel
functions of the second kind.
The next stage is to substitute the Fourier series identities of 1 and 2 into equation (7.31)
R2 Rk  1 2
s dsd
and evaluate the integrals. To start with, the integral from the term
s
0 0

2
1
is evaluated.
is equal to
s

1
s

2
=

!
0
n0 (s)m
(s) cos(n) cos(m)

n=0 m=0

X
X

+2

!
0
n0 (s)m
(s) sin(n) sin(m)

n=1 m=1

!
0
n0 (s)m
(s) cos(n) sin(m)

+ 00 (s)

n=1 m=1

n0 (s) sin(n)

(7.35)

n=1

This function is integrated with respect to P

in equation
(7.31). One term of the above
P
0 (s) 0 (s) cos(n) cos(m) term

function shall be integrated at a time. The

m
n=0
m=0 n
will be the first integrated and, to help notation, the definition
f (s) =

0
n0 (s)m
(s)

n=0 m=0

is made. If the integral from equation (7.31) is evaluated for this term then, as n, m 0 ,
the result is

Z2
0

0 ,

f (s) cos(n) cos(m) d = 2f (s) ,

f (s) ,

if
if
if

n 6= m ,
n = m = 0,
n = m 1.

(7.36)

This allows the evaluated term to be written as

Z2 X
X

0
n0 (s)m
(s) cos(n) cos(m) d

0 n=0 m=0

2(00 )2

(n0 )2

(7.37)

n=1

P
P
0
0
The term
n=1
m=1 n (s)m (s) sin(n) sin(m), from equation (7.35), gives a slightly
different result when integrated with respect to between 0 and 2 as n = m = 0

48

is not P
within the summation limits for this term. This time the definition f (s) =
P

0
0
m=1 n (s)m (s) is made and the results are:
n=1
Z2

(
0,
f (s) sin(n) sin(m) d =
f (s) ,

n 6= m ,
n = m 1.

if
if

(7.38)

This allows the term to be written as

Z2 X
X

0
n0 (s)m
(s) sin(n) sin(m) d =

0 n=1 m=1

(n0 )2

(7.39)

n=1

For the remaining two terms the integral evaluates to zero:

Z2

!
0
n0 (s)m
(s) cos(n) sin(m)

d = 0

(7.40)

n0 (s) sin(n) d = 0

(7.41)

n=1 m=1

Z2

00 (s)

X
n=1

The evaluated integral for the function found in equation (7.35) is therefore just equations
(7.37) and (7.39) added together, giving
Z2 

1
s

2

d = 2(00 )2 +

(n0 )2 +

n=1

The same method is used for the

1
s2

2

(n0 )2

(7.42)

n=1

term from equation (7.31) and a similar result

is found. The differentiation with respect to causes the function to be multiplied by n2 ,

so there is no term for n = 0. The term evaluates to
Z2

1
s2

2
d =

X
X
1 2 2
1 2 2

n
+

n
n
2
s
s2 n

n=1

(7.43)

n=1

Adding (7.42) and (7.43) together gives the total 1 part of the numerator, which is
Zk
2
0

s ds(00 )2

Zk
+

s ds
0


X

(n0 )2

n=1

49

(n0 )2




1 2 2
2 2
+ 2 n n + n n
.
s

(7.44)

Exactly the same result comes from 2 and its respective functions, but with different
integration limits. The resulting total numerator is

Zk
2

s ds(00 )2

Zk
+

s ds

+ 2

(n0 )2

(n0 )2

n=1

Z1


X

s ds(00 )2 +

Z1
s ds
k

1
+ 2 n2 n2 + n2 n2
s



X


1
(0n )2 + (0n )2 + 2 2n n2 + 2n n2 .
s

(7.45)

n=1

The denominator of (7.31) produces a more complex result. Taking only 1 to begin with,
then 12 in the denominator becomes

12 =

n (s)m (s) cos(n) cos(m) +

n=0 m=0

X
X

n (s)m (s) sin(n) sin(m)

n=1 m=1

n (s)m (s) cos(n) sin(m) + 0 (s)

+2

n=1 m=1

n (s) sin(n)

(7.46)

n=1

The 1 term of the denominator (equation 7.31) with the scale factor included then becomes
Z2

12 (1

2 2

2 2

+ 4 s + 4s cos + 6 s cos 2) d =

Z2h X
X

X
X

n (s)m (s) sin(n) sin(m) + 2

n=1 m=1

+ 0 (s)

n (s)m (s) cos(n) cos(m)

n=0 m=0

X
X

n (s)m (s) cos(n) sin(m)

n=1 m=1

i
n (s) sin(n) (1 + 42 s2 + 4s cos + 62 s2 cos 2) d .

(7.47)

n=1

The scale factor gives three different terms to multiply with 12 and integrate: a constant,
a cos term and a cos 2 term. Different integrals are needed for when the scale factor
has cos and cos 2. For the (1 + 42 s2 ) term of the scale factor the same integrals as
the numerator apply, giving the result

Zk

2 2

(1 + 4 s

)[202

(n2 + n2 )]s ds .

(7.48)

n=1

Finding the results of only the functions of , the 4s cos from the scale factor multiplies

50

with the 12 to give integrals of

Z2

1
cos n cos m cos d =
4

sin((m n 1)x) sin((m n + 1)x)

+
mn1
mn+1

sin((m + n 1)x)
sin((m + n + 1)x)
+
+
m+n1
m+n+1
2

Z
1 sin((m n 1)x) sin((m n + 1)x)
sin n sin m cos d =
+
4
mn1
mn+1

 2

 ,


(7.49)


sin((m + n 1)x)
sin((m + n + 1)x) 2
+
+
 ,
m+n1
m+n+1
0

Z2
1 cos((m n 1)x)
cos((m n + 1)x)
cos n sin m cos d =
+
4
mn1
mn+1
0

cos((m + n 1)x)
cos((m + n + 1)x) 2
+
+
 .
m+n1
m+n+1
0

(7.50)

(7.51)

For the special case of n = 0 in the first term of equation (7.49), the integral is
Z2
0


sin sin m + m cos cos m 2
cos m cos d =
 .
1 m2
0

(7.52)

Looking at the denominator in (7.52), the integral evaluates to zero unless m = 1. If the
integral is calculated with m = 1 then a normal cos2 integral results, which evaluates to
:
Z2
Z2
cos cos d = cos2 d = .
(7.53)
0

The same value also exists for m = 0 , so multiplying this by 2 for the n, m = 0 case of
the 4s cos scale factor on the denominator produces
4s[20 1 ] .
(7.54)
P
Now the sums of (7.49) to (7.51) have the limits
m,n=1 as the m, n = 0 case has been
solved. Equations (7.49) to (7.51) will all evaluate to 0 unless one of the denominators is
zero. The conditions that must be satisfied are
mn1 = 0,

mn+1 = 0,

m+n1 = 0,

m + n + 1 = 0 . (7.55)

The third condition and fourth conditions cant be satisfied if m, n 1 , so the only
conditions needed to find the integral are m = n + 1 or m = n 1. Using this relation in
equation (7.49) gives
Z2
0

1
cos n cos((n + 1)) cos d =
8

 2

sin 2nx sin(2(n + 1)x)

+
+ 2x + sin 2x  = .
n
n+1
2
0
(7.56)
51

This is multiplied by 2 as the other condition will have the same result, which gives
4s(

n n+1 ) .

(7.57)

n=1

As the integral for sin n sin m cos gives the same result, the function that result from
the sin n sin m cos is

X
4s(
n n+1 ) .
(7.58)
n=1

The equation
(7.51) evaluates to 0 when n = m + 1, n = m 1 for n, m 1 . Similarly
P
0 (s) n=1 n (s) sin(n) cos evaluates to zero when integrated between 0 and 2. These
integrals will not contribute to the denominator.
The total denominator section for the 4s cos part of the scale factor is therefore
"
#

X
4s 20 1 +
(n n+1 + n n+1 ) .
(7.59)
n=1

The term for the 62 s2 cos 2 part of the scale factor is derived in the same way and
produces the term
#
"

(7.60)
( 2 12 ) + 20 2 +
(n n+2 + n n+2 ) .
62 s2
2 1
n=1

The total denominator part from 1 is now

Zk

2 2

s ds(1 + 4 s

)[202

"

(n2
n=1

n2 )]

+ 4s 20 1 +

#
(n n+1 + n n+1 )

n=1

X
2
2
+6 s
( 1 ) + 20 2 +
(n n+2 + n n+2 ) .
2 1
n=1
(7.61)
Added to this is a similar expression for 2 in terms of i and i , but with the change of
integral limits to between k and 1 and without the 2 weighting. The total denominator,
leaving out 3 and higher terms, can be written
"

2 2

Zk

(
s ds (1 + 42 s2 )[202 +

"
(n2 + n2 )] + 4s 20 1 +

n=1

"
2 2

+6 s

#
(n n+1 + n n+1 )

n=1

#)

X
2
2
( 1 ) + 20 2 +
(n n+2 + n n+2 )
2 1
n=1
Z1
s ds {a similar expression in i and i } .

+
k

The numerator was given back at equation (7.45).

All the cross product terms are of the form i j , i j , i j and i j . When = 0 these
all disappear. There are no cross product terms of the form i j , i j or i j , i j . This
52

demonstrates that there is separation between the even and odd modes of vibration and
they can be stated separately. For the even modes of vibration all the terms with and
are utilized and and are set = = 0. For the odd modes of vibration all the terms
with and are utilized and and are set = = 0.
The even modes of vibration are therefore given by the following expression, if the sums
of the cross-terms are written out explicitly:
Zk

(
s ds 2002 +

(n02 + n2 n2 /s2 )

n=1

Z1

)
+

(
s ds 202
0 +

)
2 2
2
(02
n + n n /s )

(7.62)

n=1

is the numerator of the expression and

Zk


s ds 2(1 + 42 s2 )02 + 2(4s)0 1 + 2(62 s2 )0 2 + (1 + 72 s2 )12

+ (4s)1 2 + (62 s2 )1 3 + (1 + 42 s2 )22 + (4s)2 3 + (62 s2 )2 4 + ...
Z1
s ds {a similar expression in i }

(7.63)

is the denominator.
The odd modes of vibration replace i and i with i and i in equations (7.62) and (7.63),
although there are no 0 and 0 terms and the (1 + 72 s2 ) coefficient that multiplys the
12 and 21 terms in the denominator becomes (1 + 2 s2 ) when multiplying 12 and 21 .
With 1 and 2 written as Fourier series the boundary conditions (7.7), (7.8) and (7.9)
become
n (1) = 0 ,

n (1) = 0 ,

(7.64)

n (k) = n (k) ,

n (k) = n (k) ,

(7.65)

n0 (k)

n0 (k)

(7.66)

0n (k) ,

0n (k)

with n=0,1,2... .

7.2

The Eigenfrequencies for the Concentric 2-Density Drum

The concentric case will now be investigated. This is the case where = 0, which relates
to the 2-density case investigated by Ramakrishna and Sondhi [RS]. For this case there is
no distinction between the even and odd modes of vibration so either n and n or n and
n may be used. The cases n = 0, 1, ... can be treated separately as s and are separable
when = 0 .
After using different trial functions for n (s) and n (s), Sarojini and Rahman found that
the best results were related to the Taylor series of the Bessel function of the first kind
(A.5). n (s), which represents the Bessel fucntion of the second kind in the exact case,
also has a logarithmic infinity. The functions used are:
53

for n = 0
0 (s) = a00 + a02 s2 + a04 s4 + a00 s6 ,

(7.67)

0 (s) = b00 + b02 s2 + c00 log s ;

(7.68)

for n = 1
1 (s) = a11 s + a13 s3 + a15 s5 ,
1 (s) =

b11

b13 s3

c11 s log s

(7.69)
+

c13 s3 log s ;

(7.70)

for n = 2
2 (s) = a22 s2 + a24 s4 + a26 s6 + a28 s8 ,
2 (s) =

b22 s2

b24 s4

c22 s2 log s

(7.71)

c24 s4 log s ;

(7.72)

Looking at the n = 0 case, the 3 boundary conditions (7.7), (7.8) and (7.9) can be used
to express 3 parameters from (7.67) and (7.68) in terms of the other parameters and k.
Using the boundary condition 0 (1) = 0, equation (7.68) becomes
0 (1) = b00 + b02 12 + c00 log 1 ,

(7.73)

0 = b00 + b02 .

(7.74)

b00 = b02 .

(7.75)

which then simplifies to

This gives

Using equations (7.67) and (7.68) the boundary condition 00 (k) = 00 (k) becomes

2a02 k + 4a04 k 3 + 6a06 k 5 = 2b02 k +

c00
.
k

(7.76)

This gives c00 as

c00 = 2a02 k + 4a04 k 3 + 6a06 k 5 2b02 k k .

(7.77)

Substituting (7.75) and (7.77) into equation (7.68), the boundary condition 0 (k) = 0 (k)
is written

a0 + a02 k 2 + a04 k 4 + a06 k 6 = b02 + b02 k 2 + 2a02 k + 4a04 k 3 + 6a06 k 5 2b02 k k log k . (7.78)
This gives a00 as


a00 = b02 + b02 k 2 + 2a02 k + 4a04 k 3 + 6a06 k 5 2b02 k k log k a02 k 2 a04 k 4 a06 k 6 .
(7.79)
54

Equations (7.75), (7.77) and (7.79) are then substituted into (7.67) and (7.68), and then
in turn into (7.33) and (7.34). As these are even modes of vibration the n and n terms
are discarded and, for the case where n = 0, 1 and 2 are given by

1 (s, ) = (b02 + b02 k 2 + 2a02 k + 4a04 k 3 + 6a06 k 5 2b02 k k logk a02 k 2
a04 k 4 a06 k 6 + a02 s2 + a04 s4 + a00 s6 ) cos n ,
2 (s, ) =

b02

b02 s2

(2a02 k

4a04 k 3

6a06 k 5

2b02 k)k log s .

(7.80)
(7.81)

These expressions for 1 and 2 are then substituted into equation (7.31).
The system will exist in the lowest possible energy state allowed by the inital consitions
and the functions used for n (s) and n (s). For an oscillator the lowest-energy system
will be the system with the lowest frequency. The minimization of x will minimize the
frequency of the vibration as x = 2 b = b
c2 .
The right-hand side is minimized by using Maple 12 to vary the parameters a02 , a04 , a06 and
b02 . This gives the value of the lowest root for the concentric case where n = 0, k = 0.4,
= 3.125 and = 0. This evaluates to give x01 = 1.0068.
The different solutions for xnj must be orthogonal to one another. The equation that
must hold for orthogonal solutions is given in the appendix of Ramakrishna and Sondhis
paper [RS] and, when a change of variables from r to s is made, is of the form
Zk Z2

Z1 Z2

1nj 1n0 j 0 s ds d +
0

2nj 2n0 j 0 s ds d = 0 .
k

(7.82)

This equation guarantees orthogonality when n 6= n0 . This is because the integral

evaluates to 0 for these cases as shown previously in equation (7.36). The equation is valid
for the concentric case because the scale factor of the Laplacian is equal to 1.
To find x02 , equation (7.82) is used to restrict the values of the parameters. The values of
the parameters that produced the minimum value of x are used to give 0 and 0 for x01 :
01 2
01 4
01 6
01 (s) = a01
0 + a2 s + a4 s + a0 s ,

(7.83)

01 2
01
01 (s) = b01
0 + b2 s + c0 log s .

(7.84)

01
All the parameters are known from the minimization with a01
2 = 3.3610, a4 = 4.3383,
01
01
01
01
a01
6 = 10.091 and b2 = 0.088800. The other parameters a0 , b0 and c0 are calculated
from equations (7.75), (7.77) and (7.79).

0 and 0 for x02 are denoted

02 2
02 4
02 6
02 (s) = a02
0 + a2 s + a4 s + a0 s ,

(7.85)

02 2
02
02 (s) = b02
0 + b2 s + c0 log s ,

(7.86)

and these equations, along with equations (7.33) and (7.34), and the found numerical
parameters of 1 , are substituted into equation (7.82) and the equation is evaluated. b02
2
55

is then given in terms of the 02 parameters as

02
02
02
b02
2 = 0.57748a2 0.17125a4 0.039508a6 .

(7.87)

The right-hand side of equation (7.31) is again minimized with this new restriction on b02
2
and x02 is found to be equal to 3.0799 .
The x03 root has to satisfy orthogonality with both the x02 and x01 roots. This reduces
the free parameters for minimization to two, as the second orthogonality relation gives
03
03
a03
2 = 0.12266a4 0.011471a6 .

(7.88)

When the minimization is carried out x03 is found to be equal to 4.897 .

The same method is used to find the values of x11 , x12 , x21 and x22 . Due to the guaranteed
orthogonality when n 6= n0 , the minimization of the j = 1 functions can be implemented
with only the boundary conditions satisfied, just like the x01 case.
For x31 the following functions were used:
3
31 5
31 7
31 9
3 (s) = a31
3 s + a5 s + a7 s + a9 s ,

3 (s) =

3
b31
3 s

5
b31
5 s

3
c31
3 s log s

5
c31
5 s log s .

(7.89)
(7.90)

After minimization this produces the result x31 = 3.9675 . An attempt was made to find a
solution for x41 but Maple 12 could provide no evaluation with any accuracy or consistency
for the trial functions that were used.
The resulted for the roots are divided by x01 to give the frequency ratios. The following
table then compares them with the results from the previous method in Section 3 and the
results found by Sarojini and Rahman [SR].
Frequency ratio
Saro. and Rahman Maple 12 minimization
1.00
1.00
1.94
1.94
2.97
2.99
3.06
3.06
3.94
4.15
4.15
4.89
4.86
5.15

Section 3
01
1.00
11
1.94
21
2.95
02
3.05
31
3.97
12
4.10
03
4.82
22
5.14
Table 7.1: Comparison of frequency ratios of harmonics with = 3.125 and
k = 0.4, using the variational method

These results show that the variational method produces eigenfrequencies that have a high
degree of accuracy in replicating the previously solved analytical model. The boundary
conditions that are solved exactly by the 2-density model are solved numerically by this
method, hence the methods effectiveness is determined by its agreement with the analytical solution. Unlike the method used in Section 3, this method can be extended to the
asymmetric case to model the left-handed drum. Some alterations are needed before this
56

case can be solved. As 6= 0 for the asymmetric case, the cross product terms i j , where
i 6= j, now remain in the denominator given by equation (7.63) . Because of the presence
of these terms, slight changes to the minimization method are required.

57

Chapter 8

Conclusion
This project has found the frequencies of the different modes of vibration belonging to
three different drum models. The vibrational modes of the right-handed tabla have been
found experimentally to have frequencies that are very close to perfect harmonicity (integer multiples of the fundamental frequency). The physical properties and construction
techniques for the tabla were initially discussed, along with the work on vibrational modes
by Sir C. V. Raman [IMD] and Ramakrishna and Sondhi [RS]. From this information,
models for the tabla were created to replicate these vibrational modes.
The initial problem solved was that of the standard, one-density, drum. The frequency
ratios of the vibrational modes of this model were found to have no apparent pattern.
The project then focused on modelling the right-handed tabla. The simplest model for
the right-handed tabla consists of two different concentric regions, each of constant density.
The model was solved and the parameters were optimized, which then enabled the model
to produce eigenfrequencies close to perfect harmonicity and in good agreement with the
experimental values of the right-handed tabla (Table 3.1) .
The report then constructed a three-density model for the right-handed tabla, which
produced eigenfrequency ratios in excellent agreement with the experimental values found
by Ramakrishna and Sondhi (Table 5.3). The results produced by the 3-density model
replicate the eigenfrequency ratios of the right-handed tabla more accurately than other
existing models, including the results of recent numerical work [SA] [MS]. It is noted,
however, that the precision of the parameters used to produce these results would be
difficult to achieve in practice (Section 3.2.1) . The loaded region of the tabla is constructed
with approximately seven layers and a model with even more density regions could better
replicate this. This would provide better agreement with the experimental values and
more stability in the parameters.
The next step of the project was to show that any damping within the membrane of the
tabla would have a negligible affect on the frequency. This allowed the damping term to
be excluded from the model.
A different method for solving the model was investigated that can also be used to solve
the left-handed tabla. This method produced results that have a high degree of accuracy
in replicating the previously solved analytical model.
58

Bibliography
[ARF] G. B. Arfken
Mathematical Methods for Physicists, 4th edition, Academic Press Inc., pg 628,
ISBN 0-12-059815-9 (1995)
[B]

D. J. Benson
Music: A Mathematical Offering, Cambridge University Press, Cambridge, IBSN
978-0-521-85387-3, (2007)

[BP]

Wolfram Mathworld
http://mathworld.wolfram.com/BipolarCoordinates.html

[CT]

Crafting the Tabla

http://www.sadarang.com/Dhonkal.htm

[CK]

The Syahi
http://chandrakantha.com/tablasite/articles/pudi4.htm

[CKT] Techniques
http://chandrakantha.com/tablasite/bsicbols.htm
[FR]

N. Fletcher, T. Rossing
The Physics of Musical Instruments, Springer-Verlag, New York, (1991)

[GGL] S. Gaudet, C. Gauthier and S. Leger

Vibration of Indian Musical Drums, Journal of Sound and Vibration, 291, 1-2,
388-394, (2006)
[HT]

V. E. Howle, L. N. Trefethen
Eigenvalues of musical instruments, Journal of Computational and Applied
Mathematics, 135, 23-40, (2001)

[K]

A. Kapur
The Electronic Tabla, Computer Science Thesis, Princeton University, (2002)
http://www.cs.princeton.edu/sound/research/controllers/etabla/
Petthesis.pdf

[IMD] Sir C V Raman,

The Indian Musical Drums, Procedures of the Indian Academy of Sciences,
1A, 179-188 (1934)
[MS]

S. S. Malu and A. Siddharthan

Acoustics of the Indian Drum, arXiv:math-ph/0001030v1, February 6, (2008)
59

[RS]

B. S. Ramakrishna and M. M. Sondhi

Vibrations of Indian Musical Drums Regarded as Composite Membranes,
Journal of the Acoustical Society of America, 26, 4, 523-529, (1954)

[SA]

G. Sathej and R. Adhikari

The eigenspectra of Indian musical drums, Journal of the Acoustical Society of
America, 2, 831-838, (2009)

[SR]

T. Sarojin and A. Rahman

Variational Method for the Vibrations of the Indian Drums, Journal of the
Acoustical Society of America, 30, 3, 191-196, (1958)

[TT]

Tabla Tuning,
http://home.comcast.net/ tabla/tuningdetails.html

60

Appendix A

Bessel Functions
Bessel functions are solutions of Bessels Differential Equation, which is defined as:
x2

dy
d2 y
+x
+ (x2 n2 )y = 0 .
dx2
dx

(A.1)

Bessel functions discussed here have the common case where n is an integer. The general
solution to this equation is
y(x) = a1 Jn (x) + a2 Yn (x) ,

(A.2)

where A and B are constants and Jn (x) and Yn (x) are Bessel functions of the first and
second kind respectively.
Bessel functions of the second kind have a singularity at x = 0. Therefore if the function
has a finite value at x = 0 then a2 = 0 and the Bessel function of the second kind can be
ignored.

(a) Jn (x)

(b) Yn (x)

Figure A.1: Bessel functions of the first and second kind for values of n = 0 (red), n = 1
(blue) and n = 2 (green)

61

A.1
A.1.1

Properties of Bessel functions

Identities

Bessel functions have the following properties:

Jn+1 (x) =

2n
x Jn (x)

Jn1 (x) ,

Jn0 (x) = 21 [Jn1 (x) Jn+1 (x)] ,

Yn+1 (x) =

2n
x Yn (x)

Yn1 (x) ,

Yn0 (x) = 21 [Yn1 (x) Yn+1 (x)] .

(A.3)

If the two equations of each kind are substituted into one another then the following two
equations for each kind are produced:
Jn0 (x) = Jn1 (x) nx Jn (x) ,

Yn0 (x) = Yn1 (x) nx Yn (x) ,

Jn0 (x) = nx Jn (x) Jn+1 (x) ,

Yn0 (x) = nx Yn (x) Yn+1 (x) .

(A.4)

The Bessel function of the first kind can be defined by its Taylor Series expansion around
x = 0, [ARF]:

X
(1)s  x n+2s
.
(A.5)
Jn (x) =
s! (n + s)! 2
s=0

If n replaces n in this equation, and noting the fact that (s n)! for s =
0, ..., (n 1) , the series can be started from s = n. It can then be seen that if n is negative
then the following identity holds:
Jn (x) = (1)n Jn (x) .

(A.6)

The Bessel function of the second kind can be defined in the following way:
Y (x) =

J (x) cos() J (x)

,
sin()

Yn (x) = lim Y (x) ,

n

(A.7)
(A.8)

where is a non-integer and n an integer. The non-integer Bessel function of the first
kind subsititues for n and replaces the factorial term (n + s)! with the gamma function
( + s + 1) .

A.1.2

Bessel equation form with a solution of Jn (x)

If the equation to be solved is of the form

x21

d2 y
dy
+ x1
+ (2 x21 n2 )y = 0 ,
2
dx1
dx1

62

(A.9)

then it can be solved by the substitution of x1 = x2 . The chain rule

 
d2 y
dy
d dx2
d
d
d
is used, along with dx
=
2
dx1 dx1 and dx1 = dx2 dx1 = dx2 .

dy
dx1

dy dx2
dx2 dx1

dy
= dx
2

This gives

d2 y
dx21

d y
= 2 dx
2 , and the resulting equation is
2

x22
Giving the general solution

d2 y
dy
+ x2
+ (x22 n2 )y = 0 .
2
dx2
dx2
y(x2 ) = a1 Jn (x2 ) + a2 Yn (x2 ) ,

or

y(x1 ) = a1 Jn (x1 ) + a2 Yn (x1 ) .

A.2

(A.10)

Bessel functions using Fourier theory

This section is outlined in Bensons Music: A Mathematical Offering [B].

Fourier theory states that any periodic function can be written in the form

X
1
f () = a0 +
(an cos(n) + bn sin(n)) ,
2

(A.11)

n=1

where the coefficients are given by

1
an =

1
bn =

cos(n)f ()d ,

(A.12)

sin(n)f ()d .

(A.13)

Bessel functions can be thought of as the fourier series applied to sin(x sin ) and cos(x sin ).
The function sin(x sin ) is odd: sin(x sin ) = sin(x sin()). This means its fourier
coefficient an equals 0. It is also a half-period antisymmetric function: sin(x sin ) =
sin(x sin( )). This gives the coefficients b2n equal to zero, meaning only b2n+1
coefficients exist.
Now writing the general Fourier equation for sin(x sin ) and using the Bessel function as
the coefficient along with a factor of 2 to tidy the equation up later, the equation obtained
is

X
sin(x sin ) = 2
J2n+1 (x) sin((2n + 1)) .
(A.14)
n=0

The function cos(x sin ) is even: cos(x sin ) = cos(x sin()). This means its fourier coefficient bn equals 0. It is also a half-period symmetric function: cos(x sin ) = cos(x sin(
)). This gives the coefficients a2n+1 equal to zero, meaning only a2n coefficients exist.

63

Now writing the general Fourier equation for cos(x sin ) , as was done for the sin(x sin )
case, gives:

X
cos(x sin ) = J0 (x) + 2
J2n (x) cos(2n) .
(A.15)
n=1

Now the Bessel funciton is made the subject of the equation. Using the coefficient equation
(A.13) and setting bn = 2J2n+1 (x) produces
Z
1
sin((2n + 1)) sin(x sin )d .
(A.16)
2J2n+1 (x) =

Using the properties of an even function results in the equation
Z
1
J2n+1 (x) =
sin((2n + 1)) sin(x sin )d .
0

(A.17)

As seen while deriving equation (A.14) the function cos((2n+1)) cos(x sin ) is half-period
antisymmetric and has the value 0 at /2 , so
Z
1
cos((2n + 1)) cos(x sin )d = 0 .
(A.18)
0
Adding this to equation (A.16) gives
Z
1
[sin((2n + 1)) sin(x sin ) + cos((2n + 1)) cos(x sin )]d .
J2n+1 =
0

(A.19)

Using the trigonometric identity cos(a b) = sin a sin b + cos a cos b , this can be simplified
to
Z
1
J2n+1 =
cos ((2n + 1) z sin ) d .
(A.20)
0
Using a similar method for the an coefficient produces
Z
1
J2n (x) =
cos(2n x sin )d .
0

(A.21)

This gives use a common equation for both the odd and even integers and another definition
for the Bessel function:
Jn (x) =

cos(n x sin )d .

(A.22)

Using the relation Jn (x) = (1)n Jn (x) , in addition to the even and odd properties of
sin(2n + 1) and cos(2n) , allows equations (A.12) and (A.13) to be written as sums from
minus infinity to infinity as follows:
sin(x sin ) =

J2n+1 (x) sin((2n + 1)) ,

n=

64

(A.23)

cos(x sin ) =

J2n (x) cos(2n) .

(A.24)

n=

Using
P the same properties and equation
P (A.21) again it can also be seen that
J
(x)
sin(2n)
=
0
and
2n
n=
n= J2n+1 (x) cos((2n + 1)) = 0 , leaving us able
to simplify the equations as

sin(x sin ) =

Jn (x) sin(n) ,

(A.25)

Jn (x) cos(n) .

(A.26)

n=

cos(x sin ) =

n=

A.2.1

Deriving the recurrance relations

Now the equations found in (A.3) for Bessel functions of the first kind can be derived.
Starting with (A.25) it is differentiated with respect to giving
x cos cos(x sin ) =

Jn (x)n cos(n) .

(A.27)

Jn (x)x cos cos(n) .

(A.28)

n=

Now (A.26) is multiplied by x cos , producing

x cos cos(x sin ) =

n=

Then the trigonometric identity

right hand side to produce
x cos cos(x sin ) =

cos a cos b = 12 (cos(a + b) + cos(a b))

x
Jn (x) (cos((n + 1)) + cos((n 1))) .
2
n=

is used on the

(A.29)

The equation can then be separated into two parts and reindexed, as the limits of the sum
are and the cos(n) s are independent:
x cos cos(x sin ) =

x
(Jn1 (x) + Jn+1 (x)) cos n .
2
n=

If equation (A.27) is now used to get the left hand side as

after rearranging, the produced relation is
Jn+1 (x) =

(A.30)

n= Jn (x)n cos(n),

2n
Jn (x) Jn1 (x) .
x

then,

(A.31)

To derive the other equation a similar method is used. Starting with (A.25), it is differentiated with respect to x:
sin cos(x sin ) =

X
n=

65

Jn0 (x) sin(n) .

(A.32)

Now (A.25) is multiplied by sin getting

sin cos(x sin ) =

Jn (x) sin cos(n) .

(A.33)

n=

Then the trigonometric identity sin a cos b = 12 (sin(a + b) + sin(a b)) is used on the
right hand side to produce, along with the odd property of sine, producing
sin cos(x sin ) =

1
Jn (x) (sin((n + 1)) sin((n 1))) .
2
n=

(A.34)

The equation can then be separated into two parts and reindexed, as the limits of the sum
are and the sin(n) s are independent, giving
sin cos(x sin ) =

X
1
(Jn1 (x) Jn+1 (x)) sin n .
2
n=

If equation (A.32) is now used to get the left hand side as

then rearranged to get the relation

1
Jn0 (x) = (Jn1 (x) Jn+1 (x)) .
2

66

0
n= Jn (x) sin(n).

(A.35)
This is

(A.36)

Appendix B

Bipolar Coordinates
Bipolar coordinates are a two-dimensional, orthogonal system of coordinates that creates
two sets of circles on the two-dimensional plane.

B.1

Properties of Bipolar Coordinates

The bipolar coordinate system is defined by the transformation equations [BP]. The foci
are located at (see Figure B.1).
The transformation equations are
x=

sinh
,
cosh cos

y=

sin
.
cosh cos

(B.1)

Circles are produced when is constant of the form

x2 + (y cot )2 =

2
,
sin2

(B.2)

and when is constant circles are produced of the form

y 2 + (x coth )2 =

2
.
sinh2

(B.3)

The scale factor is given by the equation

h = h =

.
cosh cos

(B.4)

The blue circles in Figure B.1 are the circles of constant and the red circles are the circles
of constant . The asymmetric drum in the bipolar coordinate system is represented by
67

Figure B.1: Bipolar coordinates with the foci located at (-1,0) and
(1,0)

two of the circles of constant , one within the other. The outer circle represents the edge
of the drum membrane and the inner circle the loaded region. The circles of constant
are then the equivalent of radial lines in normal polar coordinates.

B.2

Finding the Circle Equations

To have Bipolar coordinates two different sets of circles must exist. One set when is
constant and one set when is constant.
First the circle of constant will be derived. If the x equation from (B.1) is divided by
the y equation, it gives

x
sinh
=
.
y
sin
p
Using sinh = cosh2 1 this can also be written
p
x
cosh2 1
=
.
y
sin

(B.5)

(B.6)

Multiplying both sides by sin produces

q
x
cosh2 1 = sin .
y
This is rearranged to make cosh the subject, which gives
s
x2
cosh =
sin2 + 1 .
y2
68

(B.7)

This is then substituted into the y equation (B.1) to give y as a function of one variable:

y=

sin
q
.
2
xy2 sin2 + 1 cos

(B.8)

To show this can satisfy a circle equation of the form

(x a)2 + (y b)2 = r2 it must be rearranged in the following manner:
s

!
x2
2
sin + 1 cos y = sin .
y2

(B.9)

Both sides are then divided by y and the cos is moved across the equality giving
s
x2

sin2 + 1 = sin + cos .

(B.10)
2
y
y
Squaring both sides then produces
2
2
x2
2
sin

+
1
=
sin2 +
sin cos + cos2 .
2
2
y
y
y
Both sides are next multiplyied by
x2 +

y2
sin2

to give

2
cos
y2
2
2 cos
=

+
2y
,
+
y
sin
sin2
sin2

and are rearranged, resulting in



2
cos
2
2 1 cos
2y
x +y
2 = 0 .
2
sin
sin
Using 1 cos2 = sin2 and

1cos2
sin2

(B.11)

(B.12)

(B.13)

= 1 now gives

x2 + y 2 2y

cos
1 cos2
2
= 0.
sin
sin2

(B.14)

Factorizing this gives the circle equation when is constant, which has the form

x2 + y

2
2
=
.
tan
sin2

(B.15)

If is constant then circles are produced as shown in Figure B.2 .

The same methodis used to find the circles of constant . Starting with equation (B.5)
and using sin = 1 cos2 gives
x2
sinh2
=
.
y2
1 cos2
69

(B.16)

Figure B.2: Circles of constant with the foci located at (-1,0) and
(1,0) for values of = 0.5 (blue), = 1 (red) and = 1.5 (green)

This is rearranged to get

r

y2
sinh2 ,
x2
and substituted into the x equation from (B.1) to give x as a function of :
cos =

x=

sinh
q
.
y2
2
cosh 1 x2 sinh

(B.17)

(B.18)

Through rearranging this can become

1


2
y2

2
sinh

=
cosh

sinh

.
x2
x

(B.19)

Multiplying out the bracket produces

1

2
2
y2
2
2
sinh

=
cosh

cosh

sinh

+
sinh2 .
x2
x
x2

Multiplying both sides by

x2
sinh2

(B.20)

results in the equation

2
x2
cosh
2
2 cosh
2
=
x

y
2
2 2x sinh + .
sinh
sinh

(B.21)

After grouping the terms together the equation becomes

x2

cosh2 1
cosh
+ 2 + y 2 = 0 .
2x
2
sinh
sinh

(B.22)

Factorizing gives the circle equation for constant :



cosh 2
2
x
+ y2 =
.
sinh
sinh2

When is constant the circles produced will be located as shown in Figure B.3 .
70

(B.23)

Figure B.3: Circles of constant with the focus located at (1,0) for
values of = 0.5 (blue), = 1 (red)and = 2 (green)

B.3

Deriving the Bipolar Scale Factor

The simplist way to obtain the scale factor is through the line element, using the equation
(ds)2 = (dx)2 + (dy)2 = h (d)2 + h (d)2 .

(B.24)

Differentiating x from equation (B.1) produces


dx =

cosh
sinh2
+
cosh cos (cosh cos )2


d

sinh sin
d .
(cosh cos )2

(B.25)

The d terms are grouped together as one fraction giving

dx =

cosh2 cos cosh sinh2

sinh sin
d
d .
2
(cosh cos )
(cosh cos )2

(B.26)

Using the identity cosh2 x sinh2 x = 1 this becomes

dx =

1 cos cosh
a sinh sin
d
d .
(cosh cos )2
(cosh cos )2

(B.27)

Similarly, differentiating y from equation (B.1) gives

dy =

sin sinh
d +
(cosh cos )2

cos
sin2
+
cosh cos (cosh cos )2


d ,

(B.28)

and when the d terms are written in the same fraction gives
dy =

cosh cos cos2 sin2

sin sinh
d
+

d .
(cosh cos )2
(cosh cos )2
71

(B.29)

Using the identity cos2 x + sin2 x = 1, the next step is

dy =

sin sinh
cosh cos 1
d +
d .
2
(cosh cos )
(cosh cos )2

(B.30)

Now, if (B.27) and (B.30) are squared and added together then they produce
2 
2 !
1 cos cosh
sin sinh
+
(d)2
(cosh cos )2
(cosh cos )2
2 
2 !

cosh cos 1
sinh sin
+
(d)2 + f (, )dd ,
(cosh cos )2
(cosh cos )2

(dx)2 + (dy)2 = 2
+2

(B.31)

where f (, )dd is the crossterm from squaring dx and squaring dy. If a common denominator is found for the terms of each integral then the equation formed is
(dx)2 + (dy)2 = 2

1 2 cos cosh + cos2 cosh2 + sin2 sinh2

(d)2
(cosh cos )4
sinh2 sin2 + cosh2 cos2 2 cosh cos + 1
(d)2 + f (, )dd . (B.32)
(cosh cos )4

+2

Using the substitution sin2 = 1 cos2 , gives

(dx)2 + (dy)2 = 2

1 2 cos cosh + cos2 cosh2 + (1 cos2 ) sinh2

(d)2
(cosh cos )4

+2

sinh2 (1 cos2 ) + cosh2 cos2 2 cosh cos + 1

(d)2 + f (, )dd .
(cosh cos )4
(B.33)

Now using cosh2 sinh2 = 1, produces

(dx)2 + (dy)2 = 2

1 2 cos cosh + cosh2 1 + cos2

(d)2
(cosh2 2 cosh cos + cos2 )2

+2

1 + cos2 + cosh2 2 cos cosh 1

(d)2 + f (, )dd .
(cosh2 2 cosh cos + cos2 )2

(B.34)

Both numerators cancel out with a factor of the denominator leaving us with the same
result for both scale factors:

(dx)2 + (dy)2 =



2
2
2
(d)
+
(d)
+ f (, )dd ,
cosh 2 cosh cos + cos2
2

or
(dx)2 + (dy)2 =



2
(d)2 + (d)2 + f (, )dd .
(cosh cos )2

72

(B.35)

(B.36)

The cross-terms produced in (B.31) are

f (, )dd = 2

(1 cos cosh )( sinh sin )

( sinh sin )(cos cosh 1)
dd + 2
dd .
(cosh cos )2
(cosh cos )2
(B.37)

From the numerators it can be seen that the cross-terms cancel out leaving
f (, )dd = 0 .

(B.38)

An orthogonal basis has been obtained with scale factors equal to

h2 = h2 =

2
.
(cosh cos )2

(B.39)

.
cosh cos

(B.40)

The squares then cancel to give

h = h =

B.4

Modified Bipolar Coordinates

Sarojini and Rahman use a modified bipolar coordinate system in their work on the asymmetric Indian Drum [SR]. This gives a set of concentric circles if r is constant and and
( z) approaches infinity. This is analagous to the right-handed drum with r and
ordinary polar coordinates. The transform is given by
r = 2e ,

(B.41)

r
.
2

(B.42)

or
= ln

Substituted in to the cartesian orthodox bipolar transforms in equations (B.1) produces

x=

4 r
r

4 + r cos

If this is multiplied by

x=

4r
4r

y=

r
4

sin
.
+ r cos

(B.43)

it tidies up to

r2 42
,
r2 + 42 4r cos

y=

r2

4r sin
.
+ 42 4r cos

(B.44)

The scale factor is modified by the change in variable. The Laplacian of unmodified bipolar
coordinates is given by
 2

1

2
2 = 2
+
.
(B.45)
h
2 2
73

Now a substitution must be made for the differential. Equation (B.41) is used to get
r

=
=r .

r
r

(B.46)

Using the chain rule

2

=
2






r
=
r
,
r
r
r

(B.47)

2
2
=
r
.
+
r
2
r
r2

(B.48)

The Laplacian is now

1
= 2
h

r2
= 2
h

or

2
r
+
r
+
r2
r 2
2

2
1
1 2
+
+
r2 r r r2 2

(B.49)

(B.50)

which gives the ordinary polar coordinates Laplacian as the bracket with a scale factor. If
equation (B.40) is substituted into (B.50) the result is
2

r
4

cos
2

2
r

2
1
1 2
+
+
r2 r r r2 2


.

(B.51)

When used with polar coordinate identities this gives a modified bipolar coordinate scale
factor of

h=

r2
4

+ r cos

74

(B.52)

Appendix C

Maple Programs
Included here are the Maple 12 programs used to solve the eigenvalue equation, draw
the graphs and optimize the parameters for minimum error. The limits within which the
different harmonics are searched for in the optimization programmes need to be altered if
the 1-density case is solved.

C.1

Solution and Graphs of the 2-Density Drum

Setting the parameters

> n := 0; sigma := 2.84; a := .385; b := 1; k := a/b; Amp := 1;

The 2-density eigenvalue equation

> c := sigma*BesselJ(n-1, sigma*k*x)/BesselJ(n, sigma*k*x)-(BesselY(n, x)*BesselJ(n-1,
k*x)-BesselJ(n, x)*BesselY(n-1, k*x))/(BesselY(n, x)*BesselJ(n, k*x)-BesselJ(n, x)*BesselY(n,
k*x));

Plot equation to locate roots

> plot(c, x = 0 .. 13, -100 .. 100);

Solve equation for chosen root

> d := fsolve(c = 0, x = 1 .. 2);

Radial plot of outer region

> e := Amp*BesselJ(n, sigma*k*d)*(BesselY(n, d)*BesselJ(n, d*r/b)-BesselJ(n, d)*BesselY(n,
d*r/b))/(BesselY(n, d)*BesselJ(n, k*d)-BesselJ(n, d)*BesselY(n, k*d));
> plot(e, r = 0 .. b, -1 .. 5);

Radial plot of inner region

75

> f := Amp*BesselJ(n, sigma*k*d*r/a);

> plot(f, r = 0 .. b, -1 .. 5);

Radial plot of drum membrane

> g := plot(e, r = a .. b, -1 .. 1); h := plot(f, r = 0 .. a, -1 .. 1); with(plots);
> display(g, h);

3D plot of drum membrane

> i := plot3d([r, theta, e*cos(n*theta)], r = a .. b, theta = 0 .. 2*Pi, coords = cylindrical);
l := plot3d([r, theta, f*cos(n*theta)], r = 0 .. a, theta = 0 .. 2*Pi, coords = cylindrical,
colour=black); m := plot3d([r, theta, f*cos(n*theta)], r = 0 .. a, theta = 0 .. 2*Pi, coords
= cylindrical, style=wireframe,colour=grey);
> display(i, l, m);

C.2

Optimizing the 2-Density Parameters

Error calculation
> Error4 := proc (sigma, k) local x01, x02, x03, x11, x12, x13, x21, x22, x31, x41;
x01 := fsolve(sigma*BesselJ(-1, sigma*k*y)/BesselJ(0, sigma*k*y)-(BesselY(0, y)*BesselJ(1, k*y)-BesselJ(0, y)*BesselY(-1, k*y))/(BesselY(0, y)*BesselJ(0, k*y)-BesselJ(0, y)*BesselY(0,
k*y)) = 0);
x02 := fsolve(sigma*BesselJ(-1, sigma*k*y)/BesselJ(0, sigma*k*y)-(BesselY(0, y)*BesselJ(1, k*y)-BesselJ(0, y)*BesselY(-1, k*y))/(BesselY(0, y)*BesselJ(0, k*y)-BesselJ(0, y)*BesselY(0,
k*y)) = 0, y = 2*x01 .. 4*x01);
x03 := fsolve(sigma*BesselJ(-1, sigma*k*y)/BesselJ(0, sigma*k*y)-(BesselY(0, y)*BesselJ(1, k*y)-BesselJ(0, y)*BesselY(-1, k*y))/(BesselY(0, y)*BesselJ(0, k*y)-BesselJ(0, y)*BesselY(0,
k*y)) = 0, y = 5*x01 .. 7*x01);
x11 := fsolve(sigma*BesselJ(0, sigma*k*y)/BesselJ(1, sigma*k*y)-(BesselY(1, y)*BesselJ(0,
k*y)-BesselJ(1, y)*BesselY(0, k*y))/(BesselY(1, y)*BesselJ(1, k*y)-BesselJ(1, y)*BesselY(1,
k*y)) = 0, y = x01 .. 3*x01);
x12 := fsolve(sigma*BesselJ(0, sigma*k*y)/BesselJ(1, sigma*k*y)-(BesselY(1, y)*BesselJ(0,
k*y)-BesselJ(1, y)*BesselY(0, k*y))/(BesselY(1, y)*BesselJ(1, k*y)-BesselJ(1, y)*BesselY(1,
k*y)) = 0, y = 3*x01 .. 5*x01);
x13 := fsolve(sigma*BesselJ(0, sigma*k*y)/BesselJ(1, sigma*k*y)-(BesselY(1, y)*BesselJ(0,
k*y)-BesselJ(1, y)*BesselY(0, k*y))/(BesselY(1, y)*BesselJ(1, k*y)-BesselJ(1, y)*BesselY(1,
k*y)) = 0, y = 4*x01 .. 6*x01);
x21 := fsolve(sigma*BesselJ(1, sigma*k*y)/BesselJ(2, sigma*k*y)-(BesselY(2, y)*BesselJ(1,
k*y)-BesselJ(2, y)*BesselY(1, k*y))/(BesselY(2, y)*BesselJ(2, k*y)-BesselJ(2, y)*BesselY(2,
k*y)) = 0, y = 2*x01 .. 4*x01);
x22 := fsolve(sigma*BesselJ(1, sigma*k*y)/BesselJ(2, sigma*k*y)-(BesselY(2, y)*BesselJ(1,
k*y)-BesselJ(2, y)*BesselY(1, k*y))/(BesselY(2, y)*BesselJ(2, k*y)-BesselJ(2, y)*BesselY(2,
k*y)) = 0, y = 4*x01 .. 6*x01);
x31 := fsolve(sigma*BesselJ(2, sigma*k*y)/BesselJ(3, sigma*k*y)-(BesselY(3, y)*BesselJ(2,
k*y)-BesselJ(3, y)*BesselY(2, k*y))/(BesselY(3, y)*BesselJ(3, k*y)-BesselJ(3, y)*BesselY(3,
k*y)) = 0, y = 3*x01 .. 5*x01);

76

x41 := fsolve(sigma*BesselJ(3, sigma*k*y)/BesselJ(4, sigma*k*y)-(BesselY(4, y)*BesselJ(3,

k*y)-BesselJ(4, y)*BesselY(3, k*y))/(BesselY(4, y)*BesselJ(4, k*y)-BesselJ(4, y)*BesselY(4,
k*y)) = 0, y = 4*x01 .. 6*x01); return (1/9)*(x02/x01-3)2 +(1/4)*(x11/x01-2)2 +(1/16)*(x12/x014)2 +(1/25)*(x13/x01-5)2 +(1/9)*(x21/x01-3)2 +(1/25)*(x22/x01-5)2 +(1/16)*(x31/x01-4)2 +
(1/25)*(x41/x01-5)2 end proc;

Parameter input
> Error4(2.84, .385);

C.3

Optimizing the 3-Density Parameters

Parameter input
> (sigmaValues, tauValues, kValues, qValues) := ({1.04, 1.05, 1.06}, {2.81, 2.82, 2.83},
{.37, .38, .39}, {.38, .39, .40});

Producing every permutation of parameters

> V := combinat[cartprod]([sigma;Values, tau;Values, kValues, qValues]);
> T := table();
> for i while not V[finished] do T[i] := V[nextvalue]() end do;
> comb := [seq(T[j], j = 1 .. i-1)];

Error calculation
> Err1 := proc (sigma, tau, k, q) local x01, x02, x03, x11, x12, x13, x21, x22, x31, x41,
beta, beta1, beta2, beta3, beta4, TheError;
beta := (BesselY(-1, sigma*k*q*y)*BesselJ(0, sigma*tau*k*q*y)-tau*BesselJ(-1, sigma*tau*k*q*y)*
BesselY(0, sigma*k*q*y))/(tau*BesselJ(-1, sigma*tau*k*q*y)*BesselJ(0, sigma*k*q*y)BesselJ(-1, sigma*k*q*y)*BesselJ(0, sigma*tau*k*q*y));
beta1 := (BesselY(0, sigma*k*q*y)*BesselJ(1, sigma*tau*k*q*y)-tau*BesselJ(0, sigma*tau*k*q*y)*
BesselY(1, sigma*k*q*y))/(tau*BesselJ(0, sigma*tau*k*q*y)*BesselJ(1, sigma*k*q*y)BesselJ(0, sigma*k*q*y)*BesselJ(1, sigma*tau*k*q*y));
beta2 := (BesselY(1, tau*k*q*y)*BesselJ(2, sigma*tau*k*q*y)-tau*BesselJ(1, sigma*tau*k*q*y)*
BesselY(2, sigma*k*q*y))/(tau*BesselJ(1, sigma*tau*k*q*y)*BesselJ(2, tau*k*q*y)-BesselJ(1,
sigma*k*q*y)*BesselJ(2, sigma*tau*k*q*y));
beta3 := (BesselY(2, sigma*k*q*y)*BesselJ(3, sigma*tau*k*q*y)-tau*BesselJ(2, sigma*tau*k*q*y)*
BesselY(3, sigma*k*q*y))/(tau*BesselJ(2, sigma*tau*k*q*y)*BesselJ(3, sigma*k*q*y)BesselJ(2, sigma*k*q*y)*BesselJ(3, sigma*tau*k*q*y));
beta4 := (BesselY(3, sigma*k*q*y)*BesselJ(4, sigma*tau*k*q*y)-tau*BesselJ(3, sigma*tau*k*q*y)*
BesselY(4, sigma*k*q*y))/(tau*BesselJ(3, sigma*tau*k*q*y)*BesselJ(4, sigma*k*q*y)BesselJ(3, sigma*k*q*y)*BesselJ(4, sigma*tau*k*q*y));
x01 := fsolve(sigma*(beta*BesselJ(-1, sigma*k*y)+BesselY(-1, sigma*k*y))/(beta*BesselJ(0,
sigma*k*y)+BesselY(0, sigma*k*y))-(BesselY(0, y)*BesselJ(-1, k*y)-BesselJ(0, y)*BesselY(1, k*y))/ (BesselY(0, y)*BesselJ(0, k*y)-BesselJ(0, y)*BesselY(0, k*y)), y = 0 .. 2);
x02 := fsolve(sigma*(beta*BesselJ(-1, sigma*k*y)+BesselY(-1, sigma*k*y))/(beta*BesselJ(0,
sigma*k*y)+ BesselY(0, sigma*k*y))-(BesselY(0, y)*BesselJ(-1, k*y)-BesselJ(0, y)*BesselY(1,k*y))/(BesselY(0,y)*BesselJ(0, k*y)-BesselJ(0,y)*BesselY(0,k*y)), y = 2*x01 ..4*x01);
77

x03 := fsolve(sigma*(beta*BesselJ(-1, sigma*k*y)+BesselY(-1, sigma*k*y))/(beta*BesselJ(0,

sigma*k*y)+BesselY(0, sigma*k*y))-(BesselY(0, y)*BesselJ(-1,k*y)-BesselJ(0,y)*BesselY(1,k*y))/ (BesselY(0,y)*BesselJ(0,k*y)-BesselJ(0,y)*BesselY(0,k*y)), y=4.5*x01 ..5.5*x01);
x11 := fsolve(sigma*(beta1*BesselJ(0, sigma*k*y)+BesselY(0, sigma*k*y))/(beta1*BesselJ(1,
sigma*k*y)+BesselY(1, sigma*k*y))-(BesselY(1, y)*BesselJ(0, k*y)-BesselJ(1, y)*BesselY(0,
k*y))/(BesselY(1, y)*BesselJ(1, k*y)-BesselJ(1, y)*BesselY(1, k*y)), y = x01 .. 3*x01);
x12 := fsolve(sigma*(beta1*BesselJ(0, sigma*k*y)+BesselY(0, sigma*k*y))/(beta1*BesselJ(1,
sigma*k*y)+BesselY(1, sigma*k*y))-(BesselY(1, y)*BesselJ(0, k*y)-BesselJ(1, y)*BesselY(0,
k*y))/ (BesselY(1, y)*BesselJ(1, k*y)-BesselJ(1, y)*BesselY(1, k*y)), y = 3*x01 .. 5*x01);
x13 := fsolve(sigma*(beta1*BesselJ(0, sigma*k*y)+BesselY(0, sigma*k*y))/(beta1*BesselJ(1,
sigma*k*y)+BesselY(1, sigma*k*y))-(BesselY(1, y)*BesselJ(0, k*y)-BesselJ(1, y)*BesselY(0,
k*y))/ (BesselY(1, y)*BesselJ(1, k*y)-BesselJ(1, y)*BesselY(1, k*y)), y = 4*x01 .. 6*x01);
x21 := fsolve(sigma*(beta2*BesselJ(1, sigma*k*y)+BesselY(1, sigma*k*y))/(beta2*BesselJ(2,
sigma*k*y)+BesselY(2, sigma*k*y))-(BesselY(2, y)*BesselJ(1, k*y)-BesselJ(2, y)*BesselY(1,
k*y))/(BesselY(2, y)*BesselJ(2, k*y)-BesselJ(2, y)*BesselY(2, k*y)), y = 2*x01 .. 4*x01);
x22 := fsolve(sigma*(beta2*BesselJ(1, sigma*k*y)+BesselY(1, sigma*k*y))/(beta2*BesselJ(2,
sigma*k*y)+BesselY(2, sigma*k*y))-(BesselY(2, y)*BesselJ(1, k*y)-BesselJ(2, y)*BesselY(1,
k*y))/ (BesselY(2, y)*BesselJ(2, k*y)-BesselJ(2, y)*BesselY(2, k*y)), y = 4*x01 .. 6*x01);
x31 := fsolve(sigma*(beta3*BesselJ(2, sigma*k*y)+BesselY(2, sigma*k*y))/(beta3*BesselJ(3,
sigma*k*y)+BesselY(3, sigma*k*y))-(BesselY(3, y)*BesselJ(2, k*y)-BesselJ(3, y)*BesselY(2,
k*y))/ (BesselY(3, y)*BesselJ(3, k*y)-BesselJ(3, y)*BesselY(3, k*y)), y = 3*x01 .. 5*x01);
x41 := fsolve(sigma*(beta4*BesselJ(3, sigma*k*y)+BesselY(3, sigma*k*y))/(beta4*BesselJ(4,
sigma*k*y)+ BesselY(4, sigma*k*y))-(BesselY(4, y)*BesselJ(3, k*y)-BesselJ(4, y)*BesselY(3,
k*y))/ (BesselY(4, y)*BesselJ(4, k*y)-BesselJ(4, y)*BesselY(4, k*y)), y = 4*x01 .. 6*x01);
TheError := (1/9)*(x02/x01-3)2 +(1/4)*(x11/x01-2)2 +(1/16)*(x12/x01-4)2 +(1/25)*(x03/x015)2 +(1/9)*(x21/x01-3)2 +(1/25)*(x22/x01-5)2 +(1/16)*(x31/x01-4)2 +(1/25)*(x41/x01-5)2 ;
if (0 <= TheError < 1) = true then return TheError else return undefined
end if end proc;

Error values for each permutation

> Error := map(@(Err1, op), comb);

Minimum error for set

> MinError := min[defined](Error);

Return parameters corresponding to minimum

> member(MinError, Error, p);
> ParametersMin := comb[p];

78

C.4

Minization in the Variational Method

This program find the value of x22 .

Define Parameters
> := 0; n := 2; k := 0.4; := 3.125;
> scale := 1+4*2 *s2 +4**s*cos()+6*2 *s2 *cos(2*);
Eliminating Parameters using the Boundary Conditions
> b2 := b4 ;
> a2 := b2 + b4 k 2 + c2 ln(k) + c4 k 2 ln(k) a4 k 2 a6 k 4 a8 k 6 ;
> c2 := 2*(b2 + b4 k 2 + c4 k 2 ln(k) a4 k 2 a6 k 4 a8 k 6 ) + 4 a4 k 2 + 6 a6 k 4
+ 8 a8 k 6 + 2 b4 4 b4 k 2 c4 (4 k 2 ln(k) + k 2 );
Eliminating Parameter using Orthogonality
> b4 := 0.03814950200 a8 0.1615789455 a6 0.5178547004 a4 0.5112898471 c4 ;
Functions for 1 , 2
> 1 := (a2 s2 + a4 s4 + a6 s6 + a8 s8 ) cos(n );
> 2 := (b2 s2 + b4 s4 + c2 s2 ln(s) + c4 s4 ln(s)) cos(n );
The equation for x2
> d := int(s*((diff(1 , s))2 +(diff(1 , ))2 /s2 ), [s = 0 .. k, = 0 .. 2*Pi]);
> e := int(s*((diff(2 , s))2 +(diff(2 , ))2 /s2 ), [s = k .. 1, = 0 .. 2*Pi]);
> f := 2 *(int(s*12 *scale, [s = 0 .. k, = 0 .. 2*Pi]));
> g := int(s*22 *scale, [s = k .. 1, = 0 .. 2*Pi]);
> h := (d+e)/(f+g);
Minimization of x2
> with(Optimization); Normalizer := simplify; Minimize(h, iterationlimit = 10000);
The value of x
> sqrt(26.8591158221655206);

79

Appendix D

CD Contents
Filename: perfectharmonicity.wav
This file contains the sound created if all the harmonics were in integer multiples.
Filename: ordinarydrum.wav
This is the sound of the harmonics produced by the ordinary drum, up to the 6th
harmonic.
Filename: 2density6th.wav
This file contains the sound produced by the best agreement with perfect harmonicity
for the 2-density model if the 03 root is placed at the 6th harmonic.
Filename: perfect3density6th.wav
This file contains the sound produced by the best agreement with perfect harmonicity
for the 3-density model if the 03 root is placed at the 6th harmonic.
Filename: perfect3density5th.wav
This is the sound produced by the best agreement with perfect harmonicity for the
3-density model if the 03 root is placed at the 5th harmonic. There is a notable
improvement over the 6th harmonic case.
Filename: experimental.wav
This file contains the sound created by the harmonics that were experimentally found
by Ramakrishna and Sondhi.
Filename: model3densitylimited.wav
This is the sound produced by the best agreement of the 3-density model with the
experimental values for the right-handed tabla.
Filename: model3densityminimal.wav
This is the sound produced by the best agreement with the experimental values for
the 3-density model when the relative density and radial ratio of the syahi and
membrane are within the physical limits of the right-handed tabla.

80