# The Structure Behind Trigonometric Functions

This entry is dedicated to my friend Bud in Milwaukee.

Sine, cosine, tangent, etc. are actually abbreviations for combinations of complex exponential functions:

$\text{sin}(\theta) = \frac{1}{2i} (e^{i\theta} - e^{-i\theta})$

$\text{cos}(\theta) = \frac{1}{2} (e^{i\theta} + e^{-i\theta})$

$\text{tan}(\theta) = \frac{ e^{i\theta} - e^{-i\theta} }{ i (e^{i\theta} + e^{-i\theta}) }$

It’s just easier to write out something like “cos”, instead of the complex exponential form. However, it is handy to know the complex exponential forms to derive trig identities on the fly. So how exactly are trig functions made of complex exponential functions?

In a previous entry, I briefly discuss how the exponential function relates angle to Cartesian coordinates. When you put an imaginary number in the exponential function it results in a complex number (a number with a real and an imaginary component). If you treat the real component like an $x$ coordinate, and the imaginary component like a $y$ coordinate, then any complex number can represent a point  on a 2-D plane.

To visualize what’s actually going on in terms of geometry, see the figure below. Imagine a point (pictured in red) that can travel along a unit circle (a circle with radius 1). You can track the point’s angular position $\varphi$ along the circle, and you can also track its position in terms of $x$ and $y$. There is a relationship between the two ways to track position, and this relationship can be expressed as:
$e^{i\varphi} = x + iy$

At this point, you may see that $x = \text{cos}(\varphi)$ and $y = \text{sin}(\varphi)$.
By definition, cosine is adjacent over hypotenuse $h$
$\text{cos}(\varphi) = x/h$
and sine is opposite over hypotenuse
$\text{sin}(\varphi) = y/h$
and since radius $= h = 1$, you get:
$x = \text{cos}(\varphi)$ and $y = \text{sin}(\varphi)$

So $e^{i\varphi} = x + iy$ is really:
$e^{i\varphi} = \text{cos}(\varphi) + i \text{sin}(\varphi)$

From this equation, you can get the equations listed at the beginning by adding or subtracting complex conjugates. For example:
Take the complex conjugate (which just means to change all the $i$ to $-i$).
$e^{-i\varphi} = \text{cos}(\varphi) - i \text{sin}(\varphi)$
Now add it to the original to get:
$e^{i\varphi} + e^{-i\varphi} = \text{cos}(\varphi) + i \text{sin}(\varphi) + \text{cos}(\varphi) - i \text{sin}(\varphi)$
$e^{i\varphi} + e^{-i\varphi} = 2 \text{cos}(\varphi)$
Now divide by $2$ and you have the identity of cosine shown at the beginning!
$\text{cos}(\varphi) = \frac{1}{2} (e^{i\varphi} + e^{-i\varphi})$

So how do we know that something like $e^{i\varphi}$ breaks up into a complex number like $x + iy$? One way is through Taylor expansions. I won’t get into how Taylor expansions are derived, since it requires calculus. The only thing you need to know here is that a function can be expressed as an infinitely long polynomial. In the case of the exponential function:
$e^{z} = 1 + z + \frac{1}{2}z^2 + \frac{1}{6}z^3 + \frac{1}{24}z^4 + ...$
or to be complete, we can write it as a summation:
$e^{z} = \sum\limits_{n=0}^{\infty} \frac{1}{n!} z^n$

So if we replace the variable $z$ with $i \varphi$ we get:
$e^{i\varphi} = 1 + i\varphi + \frac{1}{2}(i\varphi)^2 + \frac{1}{6}(i\varphi)^3 + \frac{1}{24}(i\varphi)^4 + ...$
The terms in the polynomial with odd exponents will be imaginary, and the terms with even exponents will be real. Recall that:
$i^0 = 1$
$i^1 = i$
$i^2 = -1$
$i^3 = -i$
$i^4 = 1$
So then we get:
$e^{i\varphi} = 1 + i\varphi - \frac{1}{2}\varphi^2 - \frac{1}{6}i\varphi^3 + \frac{1}{24}\varphi^4 + ...$
We can then group all the real terms together and all the imaginary terms together:
$e^{i\varphi} = \left(1 - \frac{1}{2}\varphi^2 + \frac{1}{24}\varphi^4 -... \right) + i \left( \varphi - \frac{1}{6} \varphi^3 + ... \right)$
or to be complete:
$e^{i\varphi} = \sum\limits_{n=0}^{\infty} \frac{(-1)^n}{(2n)!} \varphi^{2n} + i \sum\limits_{n=0}^{\infty} \frac{(-1)^n}{(2n+1)!} \varphi^{2n+1}$
So you can see that $e^{i\varphi}$ indeed breaks up into a complex number. By definition, this is equivalent to:
$e^{i\varphi} = \text{cos}(\varphi) + i \text{sin}(\varphi)$

It is also worth noting that this gives us the Taylor series for sine and cosine:
$\text{sin}(\varphi) = \sum\limits_{n=0}^{\infty} \frac{(-1)^n}{(2n+1)!} \varphi^{2n+1}$

$\text{cos}(\varphi) = \sum\limits_{n=0}^{\infty} \frac{(-1)^n}{(2n)!} \varphi^{2n}$

For an example, let’s derive a trig identity using the complex exponential forms.
Let’s show that:
$\text{sin}(\theta) \text{cos}(\theta) = \frac{1}{2} \text{sin}(2 \theta)$

First write the complex exponential forms:
$\text{sin}(\theta) = \frac{1}{2i} (e^{i\theta} - e^{-i\theta})$

$\text{cos}(\theta) = \frac{1}{2} (e^{i\theta} + e^{-i\theta})$
Then multiply them:
$\frac{1}{2i} (e^{i\theta} - e^{-i\theta}) \cdot \frac{1}{2} (e^{i\theta} + e^{-i\theta})$
$= \frac{1}{2} \cdot \frac{1}{2i} (e^{2 i \theta} + 1 - 1 - e^{-2 i \theta})$
$= \frac{1}{2} \cdot \frac{1}{2i} (e^{2 i \theta} - e^{-2 i \theta})$
This is the same exponential form as the sine function with a couple adjustments (a factor of $1/2$ out front and a factor of $2$ inside the function):
$= \frac{1}{2} \text{sin} (2 \theta)$

Look up some other trig identities and try to derive them using complex exponentials.

# More Buffon’s Needle

In my previous blog entry, I discussed Buffon’s needle (a method of measuring $\pi$ using random numbers). I tried to recreate what would occur in physical reality without using trig functions in the measurement of $\pi$. The tricky part was trying to make a uniform probability distribution along the angle $\theta$, which is what you expect with physical needles being dropped. In this entry, I will do something more exact, but less physical.

In reality, the probability density in $\theta$ is uniform due to there being no preferential direction for a needle to point. To simulate this with uniform random numbers without using trigonometric functions (to avoid just measuring python’s $\pi$) is tricky. The way I did this was assign uniform random numbers (uniform probability density) to a projection of a needle, either perpendicular or parallel to the lines (with a probability of 0.5 assigned to each case). This results in a probability distribution in $\theta$ that isn’t quite uniform, but it’s close enough to approximate $\pi$ after a large number of needles are dropped:
$2 \frac{N_T}{N_C} = \pi$
where $N_T$ and $N_C$ are the total number of needles dropped and number of needles crossing a line, respectively.

If instead of assigning a uniform probability to either projection fairly, what if we assign it to just one? This was shown in the previous post. As the number of needles dropped increases, the ratio $2 \frac{N_T}{N_C}$ approaches some number around $2.5$$2.6$, shown in the figure below ($\pi$ shown in red). Why is this the case?

The reason why has to do with a uniform probability density along a projection ($q$) does not translate into a uniform probability density along $\theta$, see figure below. This means needles have a preferred direction. It turns out this can still be used to extract a measurement of $\pi$!

The uniform probability density is assigned to the projection $q$:
$\rho_q = 1/L$, for $0 \leq q \leq L$, and $\rho_q=0$ elsewhere.
To get this in terms of $\theta$, some substitutions are made in the probability density integral:
$\int_0^L \rho_q dq = 1$
The variables $q$ and $\theta$ are related by:
$\text{cos}(\theta) = q/L$
so by taking the derivative we get the substitution:
$dq = -L\text{sin}(\theta) d \theta$

So now the probability density integral can be written as:
$\int_0^{\pi/2} \rho_q L \text{sin}(\theta) d \theta$
which is also:
$= \int_0^{\pi/2} \rho_{\theta} d \theta$
So it gives the form of the probability density for $\theta$ in terms of the one for $q$:
$\rho_{\theta} = \rho_q L\text{sin}(\theta)$

Plug in $\rho_q = 1/L$ to get:
$\rho_{\theta} = \text{sin}(\theta)$
and it is obviously NOT uniform, it’s just the sine function!
Plotting a histogram of the number of needles per angle $\theta$ will give the shape of the sine function:

Following the same steps as in the previous entry, we can find out why when a uniform probability is assigned to one projection, $2 N_T/N_C$ approaches $2.5$ instead of $\pi$.
For simplicity $W=L$, and again for a needle to be crossing a line it has to be below the upper bound:
$x \leq \frac{L}{2}\text{sin}(\theta)$
and the probability of a needle crossing a line is:
$P = \int_0^{\frac{L}{2}\text{sin}(\theta)} \int_0^{\pi/2} \rho_x \rho_{\theta} d \theta d x$
Plugging in the probability densities: $\rho_x = 2/L$ and $\rho_\theta = \text{sin}(\theta)$:
$P = \int_0^{\pi/2} \text{sin}^2(\theta) d \theta = \pi/4$

$P = \frac{N_C}{N_T}$ when the number of needles dropped is large, so by combining these equations:
$2 \frac{N_T}{N_C} = \frac{8}{\pi} = 2.52$
This is the number that is being approached!
But wait, $\pi$ is there, so we can extract it by:
$4 \frac{N_C}{N_T} = \pi$

By plotting this ratio as a function of number of needles dropped, we see it approaching $\pi$. See the figure below.

Even though this isn’t how needles behave in reality, it’s a simpler way to measure $\pi$ by using random numbers in a simulation. The code used to make these plots is shown below.

By the last drop:
$2 \frac{N_T}{N_C} = 2.54$
$4 \frac{N_C}{N_T} = 3.14$

import os
import sys
import math
import numpy
import pylab
import random

CL = 0
NN = 0

PIa = []
PIb = []
NNa = []
PIc = []
ANGLES = []

for ii in range(100000):

PPy = random.uniform(0.0,1.0) #A-end y position (lines separated by unit distance)

RRx = random.uniform(0.0,1.0) #x projection
RRy = numpy.sqrt(1.0 - RRx**2) #y projection

QQy = PPy + RRy #B-end y position
QQyf = math.floor(QQy)
PPyf = math.floor(PPy)

NN = NN + 1
if (QQyf-PPyf > 0.0): #if crossing line
CL = CL + 1

NNa.append(NN)
PIc.append(math.pi)
RAT = (1.0*CL)/(1.0*NN)
if (CL==0):
PIa.append(0)
else:
PIa.append(2.0/RAT)

PIb.append(4*RAT)

ANGLES.append(abs(math.atan(RRy/(RRx+0.0001)))/math.pi)

print PIa[-1]
print PIb[-1]

pylab.figure(1)
pylab.xlabel("NT")
pylab.ylabel("2 NT/NC")
pylab.plot(NNa,PIc,'r')
pylab.plot(NNa,PIa,'k')
pylab.savefig('DPI1.png')

pylab.figure(2)
pylab.hist(ANGLES, bins=10)
pylab.xlabel("Acute angle in factors of pi")
pylab.ylabel("Needles dropped")
pylab.savefig('DPI2.png')

pylab.figure(3)
pylab.xlabel("NT")
pylab.ylabel("4 NC/NT")
pylab.plot(NNa,PIc,'r')
pylab.plot(NNa,PIb,'k')
pylab.savefig('DPI3.png')

pylab.show()


# Lego Loop Limits

How can we make a circular loop out of lego bricks? For simplicity, we’ll use 2×1 bricks. The dimensions of a 2×1 brick are shown below in millimetres.

We’ll define some variables using these dimensions:
Clearance per brick end: $\epsilon = 0.1$ mm
Distance between pegs: $m = 8.0$ mm
Width: $w = m - 2 \epsilon = 7.8$ mm
Length: $l = 2m - 2 \epsilon = 15.8$ mm

If bricks are assembled in a “running bond” style, two adjacent bricks can rotate maximally by using all the available clearance space. This means that if we build a wall long enough, it can be wrapped around into a loop. The question now is, what is the minimum number of bricks we need to do this? (The figure below exaggerates this rotation.)

For any regular polygon with $n$ sides, angles $A$ and $B$ are already known (see next figure below).
Center angle $A$ is just a circle ($2\pi$) divided by the number of sides.
$A=2\pi/n$
Any triangle’s angles add up to $\pi$.
$\pi = A + 2B$
Combining these equations, inner angle $B$ is then defined in terms of the number of sides:
$B = \pi/2 - \pi/n$

Angle $B$ is related to angle $\theta$ (shown in the figure below) by:
$\pi = B + \pi/4 + \theta$
(The space between $B$ and $\theta$ is $\pi/4$ since the peg is centered such that it’s equidistant from the lengthwise edges and the nearest width-wise edge of the brick.)
Angle $\theta$ can be expressed in terms of brick width $w$ and clearance $\epsilon$.
In the figure below, the distance between the red point and blue point is $w/\sqrt{2}$, and the shortest distance from the blue point to the dashed line of symmetry is $w/2 + \epsilon$. So the angle $\theta$ can be expressed as:
$\text{sin}\theta = \frac{w/2+\epsilon}{w/\sqrt{2}}$
or
$\theta = \text{arcsin}(1/\sqrt{2} + \sqrt{2} \epsilon/w)$

Angle $B$ is then:
$B = 3\pi/4 - \text{arcsin}(1/\sqrt{2} + \sqrt{2} \epsilon/w)$

Plugging the earlier polygon formula into this equation gives a formula for $n$ in terms of brick width and clearance:
$n = \frac{\pi}{\text{arcsin}(1/\sqrt{2} + \sqrt{2} \epsilon/w)-\pi/4}$

Plugging in the dimensions, we get a minimum number of sides to make a loop:
$n \approx 121$
and since at least 3 layers are needed to create a secure running bond, the minimum number of bricks needed is 363.
This has a corresponding angle:
$\theta = 0.81136$
and acute angle between bricks:
$\delta = 2(\theta - \pi/4) = 0.051928$

Trying this in real life, I was able to get a loop with 110 sides (330 bricks).

$n_R = 110$
$\theta_R = 0.81396$
$\delta_R = 0.057120$
The theoretical calculation done above only assumes a perfectly rigid lego brick.
So the difference in $n$ may be accounted for by tolerances on the lego brick (stretching and squashing) and asymmetric angles, allowing for tighter inner angles.