IB Mathematics

Internal Assessment

Calculus of Variations

- Would the function for the minimal surface area of a revolution be a good function to

model a soap film? -

Number of Pages: 16

Candidate #
Contents

1 Introduction 1

2 The theory of the investigation 1

2.1 Arc length Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2.2 Surface Area Of A Revolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2.3 Hyperbolic Trigonometric Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.4 Euler-Lagrange Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.5 Beltrami Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 The Problem Of Minimal Surface Area Of A Revolution 10

3.1 The Extrema Of A Surface Area Of A Revolution . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.2 Analysing The Obtained Equation with Soap Film . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4 Evaluation and Conclusion 16

References 17

Appendices 18

A Recordings of Breaking Distances αi 18

1 Introduction

As a person who has been always intrigued with the mathematical nature of objects within our life, my reason to pursue

further education within college is to explore and define the relationship between mathematics, computer science and

the world. One of the key components of computer science for me is its ability to mimic real life, given that it writes

its rules using the language of mathematics. To explore this idea, in this internal assessment I will be focusing on

exploring and mathematically defining the behavior of some material, as I hope to integrate my findings to strengthen

my understanding on how the real world can be virtually simulated. Therefore in this Internal Assessment, I will be

exploring a material which is one of its kind: soap films. The idea of exploring soap films came to me from my Single

Variable Calculus MOOC (Ghrist, 2016) that I took over the summer out of interest. The professor mentioned the

intriguing property of the cosh(x) function as we were exploring hyperbolic u−substitutions for integration, briefly

mentioning that it hides an interesting property in real life. After short research, I have found out that the intriguing

function cosh(x) is involved in various minimal shapes and surfaces. To see this in action, I will be looking at soap

films, an ideal candidate to test out any minimal surfaces for my exploration as it is a natural material that always is

a minimal of surfaces (Introduction to the calculus of variations, 2016, p.17-18). I have decided to infuse the idea

of soap films and minimal surfaces with a recent topic we have covered in class of rotating functions around an axis

to obtain interesting and complex shapes. Therefore in this Internal Assessment I seek to find whether the soap film

would replicate my results after I find the minimal surface area of a revolution in the hope of answering the question

”Would the function for the minimal surface area of a revolution be a good function to model a soap film?”. However,

finding the minimum of the surface area of a revolution is no easy task, and I will require a lot of theory before being

able to answer my exploration. Namely, I will first have to find the equation that will represent the surface area of a

revolution of a function in some domain and effectively find the minimal of this equation using the Beltrami Identity,

a special case of the Euler-Lagrange equation from Calculus of Variations which I require to understand first.

2 The theory of the investigation

2.1 Arc length Integral

I first require to understand what equation I will be truly minimising. Intuitively the equation must include the arc

length of the function as I will rotate it along an axis, similar to volume of revolution

1
 Figure 1: A graphical depiction of infinitesimal triangle dydx (not to scale)(Desmos Graphing Calculator, 2011)

To compute the surface area of a revolution of some function f (x) in the domain (xa , xb ) where xa < xb and

xa , xb ∈ R, I will consider the triangle of an infinitesimally small width dx and its corresponding infinitesimally

small height dy, to then compute its infinitesimally small hypotenuse dL which represents an infinitesimally small

function length by Pythagoras’ Theorem.

p
dL = dy 2 + dx2

And if I factor out dx from the square root I can obtain

s 2
dy
dL = + 1 dx (1)
dx

Then, if I use the definite integral definition to be the sum of all infinitesimals within the domain (xa , xb ), I can obtain

the exact length of the function as

s 2
Z xb
dy
L= + 1 dx
xa dx

2.2 Surface Area Of A Revolution

Whilst I have found a way to compute the arc length of a function, I will now have to find a way to rotate it in order

to obtain the surface area of a revolution. If I consider the volume of revolution integral V in the domain (xa , xb )

Z xb
V =π f (x)2 dx
xa

Which is indeed derived from the classical formula for the area of a circle πr2 , where the radius r can be represented

by an infinitesimal rectangle of width dx (imitating a very thin strip) and height f (x). If the infinitesimal rectangle is

rotated 360◦ around the x axis, it forms an infinitesimal cylinder of width dx (which in turn imitates a circle, hence

I use the formula for an area of a circle). The integral sums all such infinitesimal cylinder areas in the domain of its

bounds to obtain the volume.

Using similar intuition, I can use the formula 2πR to sum the surface areas of the infinitesimal cylinders instead.

However, in this particular case, the R within the function is not represented by f (x) only. If I use the same logic to

rotate a rectangle of width dx and height f (x) around the x axis to obtain the same cylinder, I obtain no information

2
on the surfaces. Instead, I need something that can represent this surface as I rotate. If I introduce the infinitesimal

arc length from equation 1 to be included after the height f (x) at some point, I am obtaining a new shape by the name

of a ”frustum” (a shape similar to a cone, but the cone’s tip is subtracted by a smaller similar cone). A frustum’s

surface area can be conveniently defined through its base radius multiplied by its arc length and 2π (Areas of Surface

of Revolution, 2019). Infinitesimally, I then can define R

s 
dy 2
R = f (x) + 1 dx
dx

Figure 2: An infinitesimal frustum with width dx (not to scale)(Geogebra 3D Calculator, 2011)

If I substitute my R into the circumference formula and sum all infinitesimal surface areas (like summing infinitesimal

circumferences of circles) in the interval (xa , xb ) by a definite integral, the surface area of a revolution is then

s 
xb
dy 2
Z
S = 2π f (x) + 1 dx (2)
xa dx

2.3 Hyperbolic Trigonometric Functions

Since the professor has mentioned that this property is specifically held within the hyperbolic trigonometric functions

as I was doing various u substitutions, it is wise define these functions and their identities, as I am likely to require

them in the future. The hyperbolic trigonometric functions are defined to be (Weisstein, n.d.)

ex − e−x
sinh(x) ≡
2
ex + e−x
cosh(x) ≡
2

The inverse of these functions are defined to be arsinh(x) and arcosh(x) respectively. Furthermore, the hyperbolic

functions result in the identity

cosh2 (x) − sinh2 (x) ≡ 1 (3)

3
Which is an identity similar to tan2 (x) − sec2 (x) ≡ 1, allowing an alternative for substitution for integrals originally

requiring tan(x), and can simplify some work which is the case for me later in the investigation. I will show the

validity of this identity by substituting the algebraic definitions of the functions, thus

2  x 2
ex + e−x e − e−x

=⇒ −
2 2
2x
e +2+e −2x e − 2 + e−2x
2x 4
=⇒ − = =1
4 4 4

Lastly, knowing the algebraic definitions of cosh(x) and sinh(x), it is also implied that cosh(x) is even and sinh(x)

is odd, that is:

e−x + ex
cosh(−x) = = cosh(x)
2

and
e−x − ex
sinh(−x) =
2
ex − e−x
=⇒ − = − sinh(x)
2

2.4 Euler-Lagrange Equation

The Euler-Lagrange Equation is an equation to find extrema (maximas and minimas) of equations I of the kind

Z xb
I= F (x, y, y 0 ) (5)
xa

The integrand contains what is defined to be a functional, which is a function with functions as its input. Indeed, the

surface area of a revolution formula is also an equation with an integrand that is explicitly dependent on y and y 0 ,

therefore the Euler-Lagrange is applicable for my investigation. The Euler-Lagrange equation is defined as

 
∂F d ∂F
− =0 (6)
∂y dx ∂y 0

However, this equation does not state the nature of the extrema it finds. The notation ∂ used within the equation

denotes a partial derivative, these allow me to analyse the rate of change of a multivariate function with respect to

a single variable, treating the rest like constants. For example, if I consider the equation F (x, y) = x3 y + y 3 x, its

partial derivative with respect to y is

∂F
= x3 + 3y 2 x
∂y

As I want to understand how the Euler-Lagrange equation works as it is a tool that is useful in this exploration, I will

see how it will help me find the minimum of a surface area of a revolution by analysing its derivation. However, in

4
order to do this, I will take the multivariate chain rule as a given. The multivariate chain rule states that for some

multivariate function F (x1 , x2 , . . . , xn ) with distinct variables x1 , x2 , . . . , xn , when differentiated with respect to

some variable such as t, the chain rule can be given by

n
dF X ∂F dxi
=
dt ∂xi dt
i=1

Similarly, if I am looking for the partial derivative instead

n
∂F X ∂F ∂xi
=
∂t ∂xi ∂t
i=1

The rigorous proofs of these generalisations, however, are well beyond the scope of this IA.

To understand how the Euler-Lagrange specifically applies to the equation of surface area of a revolution, I will

begin by denoting the function (The Euler-Lagrange Equation, or Euler’s Equation, n.d.)

Y (x) = y(x) + εη(x) (7)

Where y(x) is a twicely differentiable function whose domain is [xa , xb ] with points (xa , ya ) and (xb , yb ). For y(x)

I will assume that it is an extremum function in the interval [xa , xb ] (that is, it is either a function that maximises or

minimises of my function Y (x) in the domain [xa , xb ]).

η(x) is a twicely differentiable function whose domain is [xa , xb ] and has coordinates (xa , 0) and (xb , 0). This

allows me to add some arbitrary function to y(x) (variation) in order to get Y (x) in any arbitrary form whilst

ensuring my boundary conditions in (xa , ya ) and (xb , yb ) stay constant;

ε is some parameter which which allows me to configure my variation of η(x).

In order to intuitively understand equation 7, I have plotted Figure 3 to visualise it in terms of arc length:

Figure 3: diagram of Y (x), y(x) and εη(x) (Geogebra 3D Calculator, 2011)

My problem is then to find the extrema of the function Y (x). Since η(x) is some arbitrary function, it means that

5
Y (x) will represent a family of curves which can be resulted from adding some arbitrary function η(x) to an

extremum y(x).

Intuitively, a possible extremum function is a straight line in an arc length of the kind y = mx + b (as seen in figure

3), this minimises distance between two points. Moreover, if I were to choose the parameter ε to be approaching

smaller values (closer to 0), then function Y (x) will approach the extremum y(x) (that is, the variation added from

the arbitrary function εη(x) will be getting smaller). If I consider the equation Y (x) (equation 7) which represents

twicely differentiable family of curves in a restricted domain (namely my domain where I seek to minimise, that is,

the domain which I rotate around the x axis) and write it in the same as the equation of the kind 5, I obtain general

integral of a specific family of curves. Namely:

Z xb
I= F (x, Y, Y 0 )dx
xa

I know from the definitions of Y and Y 0 are dependent on x and ε. However, once the integral is computed with the

x boundaries, only the variable ε is left in the equation. This means that I is actually only dependent on the

parameter ε and its value, that is

Z xb
I[ε] = F (x, Y, Y 0 )dx
xa

And to find a candidate for the minimum, as I is only dependent on ε, I can set its derivative to 0, in order to compute

an extremum like in regular calculus. This allows me to find the extremum from the variational change added to y(x)

dI
=0
dε

However, recall from my assumption that y(x) is an extremum function, so I can then deduce that the solution of the

equation above would be when I set the variation ε = 0 (when my family of functions becomes an extremum i.e.

Y (x) = y(x)), giving me the equation

dI
=0
dε ε=0

substituting I
Z xb
d
F (x, Y, Y 0 ) dx = 0
dε xa ε=0

The boundaries of the integral are constants. Furthermore, the integrand consists of a multivariate function whose

variables can be reduced to x and ε only. The derivative, on the other hand, is with respect to ε. This implies that I

can apply Leibniz integral rule given as (Haile, 2020)

Z xb  Z xb
d ∂
f (x, ε)dx = f (x, ε)dx
dε xa xa ∂ε

6
Moreover, I know that x is not dependent on ε, so its derivative will simply be 0. However, Y (x) and Y 0 (x) are

explicitly dependent on ε, which implies after I evaluate the RHS

For x, as x is not dependent on ε For Y For Y 0

∂F ∂x ∂F ∂Y ∂F ∂Y 0
=0
∂x ∂ε ∂Y ∂ε ∂Y 0 ∂ε

Hence I obtain after applying the chain rule

xb
∂F ∂Y 0
Z  
dI ∂F ∂Y
= 0+ + dx = 0 (8)
dε xa ∂Y ∂ε ∂Y 0 ∂ε ε=0

∂Y 0
I can in fact obtain the expressions ∂Y
∂ε and ∂ε from equations Y 0 (x) and Y (x) which will allow me to simplify even

further. Partially differentiating equation 7 with respect to ε

∂Y ∂y ∂
= + (εη(x))
∂ε ∂ε ∂ε

I know that y(x) is not dependent on ε, so its derivative is 0. This implies

∂Y
= η(x) (9)
∂ε

I will differentiate with respect to x the equation 7 to obtain

Y 0 (x) = y 0 (x) + εη 0 (x)

I will also partially differentiate it with respect to ε

∂Y 0 ∂y 0 ∂
= + (εη 0 (x))
∂ε ∂ε ∂ε

∂Y 0
= η 0 (x) (10)
∂ε

Substituting equations 9 and 10 to equation 8 I obtain

Z xb  
dI ∂F ∂F 0
= η(x) + η (x) dx = 0 (11)
dε xa ∂Y ∂Y 0 ε=0

The equation can be simplified further to obtain it in the form of Euler-Lagrange. Using integration by parts with

7
respect to x where u = ∂F
∂Y 0 and v 0 = η 0 (x). Then I obtain that

 
d
0 ∂F
u = and v = η(x)
dx ∂Y 0
Z xb   xb Z xb  
∂F 0 ∂F d ∂F
=⇒ 0
η (x)dx = η(x) 0 − η(x)dx
xa ∂Y ∂Y x=xa xa dx ∂Y 0

However I know η(xa ) = 0 and η(xb ) = 0 as I have defined, implying that the evaluated expression is 0, therefore

Z xb Z xb  
∂F 0 d ∂F
η (x)dx = − η(x)dx
xa ∂Y 0 xa dx ∂Y 0

I will substitute this into equation 11 to obtain

Z xb    
dI ∂F d ∂F
= η(x) − η(x) dx = 0
dε xa ∂Y dx ∂Y 0 ε=0

Factorising η(x)
Z xb   
dI ∂F d ∂F
= − η(x) dx = 0
dε xa ∂Y dx ∂Y 0 ε=0

From the definition of Y (x), that is equation 7, if I evaluate at ε = 0 then Y (x) = y(x) and Y 0 (x) = y 0 (x), so my

new equation then becomes

Z xb   
dI ∂F d ∂F
= − η(x)dx = 0
dε xa ∂y dx ∂y 0

For equation 7 to represent a family of curves, I can’t let the arbitrary function η(x) = 0 otherwise variations would

also not exist. This implies that the inner bracket must be 0 instead, obtaining me the Euler-Lagrange equation

 
∂F d ∂F
− =0 (12)
∂y dx ∂y 0

Therefore, Euler-Lagrange allows me to find an extremum function defined as y(x) within the domain of the integral

limits. This is done by creating a family of curves resulting from adding some arbitrary function η(x) with

variational parameter ε to the extremum function y(x), obtaining Y (x). Differentiating the integral with the

integrand of the functional F (x, Y, Y 0 ) (which now represents a family of curves of functionals) with respect to ε

forces this functional to take shape of the extremum function y(x). This implies that the final result I obtain after

applying Euler-Lagrange to the surface area of a revolution formula will be the function y(x).

2.5 Beltrami Identity

The Beltrami Identity is a special case of the Euler-Lagrange equation when some functional in the integrand

F (x, y, y 0 ) is not explicitly dependent on x (by definition this is ∂F

∂x = 0). I can denote such functionals as F (y, y 0 ).

8
Since my surface area a revolution integrand is also not explicitly dependent on x in the integrand, thus I can make

use of the identity to find the minimal. I can obtain the Beltrami Identity by applying the Euler-Lagrange equation to

a general functional F (y, y 0 ) (Functionals leading to special cases, n.d.). Hence, I first find dF
dx through the

multivariate chain rule:

For y For y 0

∂F dy ∂F dy 0
∂y dx ∂y 0 dx

dF ∂F dy ∂F dy 0
=⇒ = + 0 (13)
dx ∂y dx ∂y dx
dy
I can apply the Euler-Lagrange equation to my functional F (y, y 0 ) if I multiply the Euler-Lagrange equation by dx

on both sides, therefore

 
dy ∂F dy d ∂F
− =0
dx ∂y dx dx ∂y 0
dy ∂F
Now substituting equation 13 to the Euler-Lagrange equation for dx ∂y I get

∂F dy 0
 
dF dy d ∂F
− 0 − =0
dx ∂y dx dx dx ∂y 0

I will factor out the negative to obtain

∂F dy 0
  
dF dy d ∂F
− 0
+ =0 (14)
dx ∂y dx dx dx ∂y 0

I can re-write one of the inner derivatives as the following

dy 0
 
d dy
=
dx dx dx

as y is a function which is only dependent on x. Using this information I can write the equation 14 as

    
dF ∂F d dy dy d ∂F
− + =0 (15)
dx ∂y 0 dx dx dx dx ∂y 0

 
dy ∂F
It became evident that the expression inside the bracket is actually the product rule of d
dx dx ∂y 0 , implying that we
dy
can simply our expression even further. We can confirm by letting u = dx and v = ∂y 0 ,
∂F
then

   
0 d dy 0 d ∂F
u = v =
dx dx dx ∂y 0

9
      
d dy ∂F ∂F d dy dy d ∂F
=⇒ = +
dx dx ∂y 0 ∂y 0 dx dx dx dx ∂y 0

Hence for equation 15 I can re-express it as

 
dF d dy ∂F
− =0
dx dx dx ∂y 0

I notice that the expression above is in fact an expression that was differentiated with respect to x all throughout,

therefore I can simplify by writing

 
d dy ∂F
F− =0
dx dx ∂y 0

And I can even further simplify if I integrate everything with respect to x

Z   Z
d dy ∂F
F− dx = 0dx
dx dx ∂y 0

Which after integration I obtain that for some arbitrary constant C:

dy ∂F
F− =C (16)
dx ∂y 0

Which is the Beltrami Identity, reduced version of the Euler-Lagrange equation that I can use since my surface area

of a revolution integrand is not dependent on x, granting me a simpler way to solve my question.

3 The Problem Of Minimal Surface Area Of A Revolution

3.1 The Extrema Of A Surface Area Of A Revolution

Indeed, due to my problem, I am looking to minimise the surface area of a revolution equation 2, therefore I am

looking to apply the Euler-Lagrange equation. Indeed, the surface area of a revolution equation is not dependent on

x, hence I could write it as s  2

0 dy
F (y, y ) = y(x) 1+
dx

Hence I will use the Beltrami identity in order to compute its extrema as planned. I first compute
s 
 2
∂F ∂ dy
0
= y(x) 0  1+ 
∂y ∂y dx

dy
∂F
=⇒ = y(x) r dx 
∂y 0 dy
2
1 + dx

10
Applying the Beltrami Identity (equation 16) I obtain

 2
dy
s  2
dy dx
y(x) 1 + − y(x) r 2 = C
dx 
dy
1+ dx

q
dy 2
1+( dx )
Factorising y(x) and multiplying the left expression inside the bracket by q
dy 2
to express it as a single fraction
1+( dx )

  2  2 
dy dy
 1+ dx  dx
y(x)   2 − r  2  = C

r
dy dy
1 + dx 1 + dx

y(x)
=⇒ r  2 = C
dy
1 + dx

dy
In fact, this is a hidden differential equation if I make dx the subject that I have to solve If I want to obtain a clear

result of my findings. Raising everything to 2 I get

y 2 (x) 2
 2 = C
dy
1 + dx

dy
And hence rearranging for dx
dy 2
 
2 2
y (x) = C + C
dx
p
y 2 (x) − C 2 dy
=
C dx

The ± is absorbed by the constant C. This is a separable differential equation, therefore

dx dy
=p
C y (x) − C 2
2

Z Z
dx dy
= p (17)
C y (x) − C 2
2

Whilst the right hand side could be solved with u substitution by letting y(x) = C tan(u) to obtain an integrand of

sec(u), the solution is non-trivial and complex. Instead, I will use my defined hyperbolic trigonometric identity for a

u substitution (equation 3). Let

y(x) = C cosh(u) (18)

11
 dy
=⇒ = C sinh(u)
du

Substituting in the integral and factoring out C in the denominator

Z
C sinh(u)
q du
C cosh2 (u) − 1

Using my identity (equation 3) for the denominator and simplifying I obtain

Z
C sinh(u)
q du
2
C sinh (u)

Z
du = u

Solving for u from my substitution in equation 18

 
y(x)
arcosh =u
C

Thus I obtain the solution

Z  
dy y(x)
p = arcosh
y 2 (x) − C 2 C

Substituting the integral and solving equation 17

 
x y(x)
+ h = arcosh
C C

Where h is the constant of integration. Finally rearranging for y(x) I obtain

x 
C cosh + h = y(x)
C

Therefore the above equation is the extrema function y(x) from equation 7 for my particular example of the surface

area of a revolution.

3.2 Analysing The Obtained Equation with Soap Film

Well, but what does this really mean? What does the cosh(x) really tell me for my soap film? Since the cosh(x)

function is even, I can force my equation C cosh Cx + h to be even too if I let h = 0, that is, I will force the

symmetry to occur at x = 0 to make the investigation and interpretation easier. Then, if I was to restrict the domain

with some number α where α ∈ R such that the domain is [−α, α], the left side ”end” (where x = −α) must be

12
identical with the right side ”end” (where x = α) from cosh(α) = cosh(−α), therefore I can assume these ”ends”

also have identical (same radius) frustums of infinitesimal width after rotating the function 360◦ along the x axis.

The C cosh Cx function in the domain [−α, α], then, should be hypothetically mimicked by the soap film if I was to

sandwich it between two pair of identical real life rings. That is, these rings will force the soap film to take the same

properties of C cosh Cx as these rings will help replicate the fact that cosh(α) = cosh(−α). Therefore, in order to

investigate this hypothesis, I have cut multiple water pipes of different radius with a saw blade to get approximately the

same thickness, and precisely the same radius pairs of rings to investigate. I was able to create a total of 8 rings of 4

distinct radii, which I will denote as ri where i ∈ {1, 2, 3, 4}. Specifically, these radii are r1 = 1.35cm, r2 = 1.10cm,

r3 = 0.85cm and r4 = 0.65cm. To investigate the equation, however, I have selected r1 as its radii is the biggest thus

the easiest to see. The outcome of the investigation to test the hypothesis can be seen in Figure 4 below, a graph with

a photo which I took as I investigated and compared with the function C cosh Cx :

Figure 4: Real life catenoid picture on an axis with function C cosh x

(Desmos Graphing Calculator, 2011)


C

The constant C for the image above was found using trial and error, in particular, for the measured distance α ≈ 0.87,

it is C ≈ 0.9. As hypothesised, the soap film is the rotated function C cosh( Cx ) for 360◦ around the x axis in the

interval [−α, α], representing a ”catenoid”, the 3D rotated variant of the function. However, whilst I have found

an expression for a specific value of α, I am looking to apply it to my soap film for all α, as I seek to uncover its

mathematical behaviour for all cases of α. To do this algebraically, I have noticed that I can apply my soap film to

this equation from the α that I have denoted. I know that for the distance x = α I will always have y = 1.35 as seen

in figure 4, as the height is y is bounded by the radius of the rings at that point. Hence, for my soap, the following

equation holds true

α
1.35 = C cosh (20)
C

13
Now reducing the equation to only α and C, I can begin analysing for what values of α and C I can obtain an answer

of 1.35 by graphing equation 20 so I can effectively solve for my C for any α value

Figure 5: Graph of 1.35 = C cosh( C

α
) (Desmos Graphing Calculator, 2011)

This shows me all the possible value of α with its C values, and the first thing I notice is that we do not have solutions

for all α ∈ R and I have a maximum attainable α value which I want to find and test out in real life. I first denote that

for any ri its corresponding α maximum coordinates to be (αi , Cαi ). Thus, in the figure above, Cα1 was denoted to

be the corresponding C value for α = α1 of the equation. Notice that from the definition of the equation, at C = 0 the

graph becomes discontinuous, as C 6= 0. Moreover, the above figure shows me that for |α| < α1 I have 2 solutions,

and for |α| > α1 I have no solutions, therefore finding the value of α1 may be crucial in finding the breaking distance

for my soap films, as solutions do not exist for the radius after the distance |α| > α1 . To find this value, I will implicitly

differentiate with respect to α as the gradient of the tangent is undefined at α1 . Let u = C and v = cosh( C
α
), then

dC  α  C − α dC
u0 = and v 0 = sinh · dα
dα C C2

Finalising the product rule

 α  C − α dC dC α
dα
sinh · · C + · cosh =0
C C2 dα C

Re-arranging to make dC
dα the subject

dC α  α  dC α α
− sinh + cosh = − sinh
dα C C dα C C

α


dC − sinh C
= α α

α


dα − C sinh C + cosh C

If I set the denominator 0, I will find all points where the gradients are undefined (or m → ∞ where m is the gradient

of the tangent). Hence

α α
−α sinh + C cosh =0 (21)
C C

Well, but what does this mean? This is a very, very interesting result. I know that the catenoid equation must be equal

to the radius ri of the rings. However, radius, which is a constant and the boundary condition of my equation, does not

14
affect the derivative. This means that the derivative of the catenoid equation must apply to all catenoids with different

radii, that is, it must show the solution to all possible maximum distances x = α of different ri . Consider Figure 6

with different ri = C cosh C α

and equation 21 to visualise this relationship:

Figure 6: Graphs of ri = C cosh( C

α
) and −α sinh( C
α α
) + C cosh( C ) = 0 (Desmos Graphing Calculator, 2011)

Finding the intersections between equation 21 and the catenoid equation is not solvable by hand, as isolating α or C

is not possible. Solving for the 2 using technology for all ri grants me the solutions for αi in the table below. To test

whether the hypothesised αi breaking distance is the same, I have recorded the breaking distances of my soap film

in real life using a high-res camera and slow motion. The recorded distance was denoted to be βi (By measuring the

distance between two rings and dividing by 2 on the breaking frame. See Appendix A for the recordings).
|αi −βi |
i ri αi βi αi · 100
1 1.35 0.894704 0.90 0.59%
2 1.10 0.729018 0.75 2.88%
3 0.85 0.563332 0.55 2.37%
4 0.65 0.430783 0.45 4.46%

Table 1: Recorded results of breaking point distance (βi ), hypothesised breaking distance (αi ) and percentage error

The evaluation of these results can be found in section 4. Whilst I was able to find an equation to solve for all

αi and the breaking distance, I have yet to understand why for any distance |α| < αi I have 2 solutions. Since the

Euler-Lagrange does not show the nature of the extrema, it is possible that one solution is a minima, whilst the other

is a maxima. I can test out for which solution of C resembles my soap film, as I know that soap film mimics minimal

surfaces, which will allow me to negate one of the solutions. To do this, I will consider a arbitrary distance such

as α = 0.5, and analyse how different C compute the equation y(x) = C cosh Cx (Introduction to the calculus of

variations, 2016, p.45-46). Using technology, I find that the following are the possible solutions for C when α = 0.5:

C ' 0.187988, 1.24854

I graph the function with these C values (graphs of y = 0.187988 cosh( 0.187988
x
) and y = 1.24854 cosh( 1.24854
x
)) in

15
the domain of [−0.5, 0.5] as I have chosen from α = 0.5, to see how my soap film shape differs with both C

Figure 7: Graph of y = 0.187988 cosh( 0.187988

x
) and y = 1.24854 cosh( 1.24854
x
) in the domain [−0.5, 0.5]
(Desmos Graphing Calculator, 2011)

It is now possible to see from Figure 7 that for larger C, the function arc length is smaller than the other C solution

function. Moreover, the shape with C = 1.24854 largely corresponds with the shape in my photo if it is rotated 360◦

around the x axis! Thus, I can deduce that for values of C > Cαi it is a minimal.

4 Evaluation and Conclusion

The function that I’ve found to express soap films between rings and find the breaking distance shown in table 1

seems to be accurate through real life testing! However, deviations in results in the table can occur from factors such

as thickness of the rings that I have used, the surrounding air pressure or even the concentration of the soap solution.

For a more accurate model in computers, these would also have to be mathematically defined. Moreover, given that

I’ve used a ruler to both measure the radius and the distance between the rings, I was prone to parallax error, as well

as limited accuracy of measurement to 1 decimal point. Despite this, however, my equation to solve for αi for any ri

seems to be a rather good approximation for the breaking distance βi in room conditions. This means that I can use

equation 20 whilst restricting the value of my C to be C ≥ Cαi for some distance α to then compute C and thus find a

function that would represent the soap film for that particular value of α, and finally rotate that function 360◦ around

the x axis to obtain the final simulation.

However, equation 21 also caught my eye as it seems that its relationship is almost linear as seen in figure 6, therefore

it is worth considering and checking whether all αi of their respective ri follow some linear equation and thus a

ratio. This can be useful in the future for optimisation of my model, as the computer will simply have to downscale

and upscale the breaking distance for some constant, without having to compute each solution for catenoid using the

equation of the derivative which can take significantly more processing power.

The main tool of this investigation which I had to learn and use, the Euler-Lagrange equation, undoubtedly brings many

opportunities for me to further define even more mathematical relationships for modeling. One of such mathematical

relationships is the newly introduced Ray Tracing technology in graphics. This technology is defined through Snell’s

law, which is also derived from the Euler-Lagrange equation and Fermat’s principle (light always seeks minimal path).

16
References

Areas of surface of revolution [Online]. (2019). Retrieved 2020-11-19, from https://math.libretexts

.org/Courses/University of California Davis/UCD Mat 21B%3A Integral Calculus/

6%3A Applications of Definite Integrals/6.4%3A Areas of Surfaces of Revolution (The

LibreTexts libraries are Powered by MindTouch® and are supported by the Department of Education Open

Textbook Pilot Project, the UC Davis Office of the Provost, the UC Davis Library, the California State

University Affordable Learning Solutions Program, and Merlot. We also acknowledge previous National

Science Foundation support under grant numbers 1246120, 1525057, and 1413739. Unless otherwise noted,

LibreTexts content is licensed by CC BY-NC-SA 3.0. Legal.)

Desmos graphing calculator [Online]. (2011). Retrieved 2020-08-12, from https://www.desmos.com/

calculator

The Euler-Lagrange Equation, or Euler’s Equation [Online]. (n.d.). Retrieved 2020-08-04, from https://www.ucl

.ac.uk/~ucahmto/latex html/chapter2 latex2html/node5.html#tex2html2

Functionals leading to special cases [Online]. (n.d.). Retrieved 2020-08-04, from https://www.ucl.ac.uk/

~ucahmto/latex html/chapter2 latex2html/node8.html

Geogebra 3d calculator [Online]. (2011). Retrieved 2020-08-04, from https://www.geogebra.org/3d?lang=en

Ghrist, R. (2016). Single Variable Calculus | University of Pennsylvania [MOOC]. Retrieved 2020-08-01, from

https://repository.upenn.edu/discrete calculus/index.2.html

Haile, P. (2020, February). Differentiating an Integral [Online]. Retrieved 2020-08-03, from http://www.econ

.yale.edu/%7epah29/409web/leibniz.pdf

Introduction to the calculus of variations [PDF]. (2016). Open University. Retrieved 2020-08-01,

from https://www.open.edu/openlearn/ocw/pluginfile.php/1118521/mod resource/content/

3/Introduction%20to%20the%20calculus%20of%20variations ms327.pdf

Weisstein, E. W. (n.d.). Hyperbolic Functions [Online]. Retrieved 2020-08-06, from https://mathworld.wolfram

.com/HyperbolicFunctions.html (Publisher: Wolfram Research, Inc.)

17
Appendices

A Recordings of Breaking Distances αi

For the availability, privacy, and security of these videos, the recordings were uploaded and are available as unlisted.

These were uploaded using an anonymous YouTube account.

Recording using r1 rings for α1

https://youtu.be/u1CGKfzeXLw

Recording using r2 rings for α2

https://youtu.be/A8n7OSjcINY

Recording using r3 rings for α3

https://youtu.be/gj7ul-aR6Rg

Recording using r4 rings for α4

https://youtu.be/4wBY77QKvWs

18