A discussion on the theory behind fractals, several different examples and applications of fractals in modern day life. This discusses the Coastline Paradox, Image Compression and uses within Creative Media
2. OVERVIEW
• Introduction
• Examples of Fractals
• Cantor Set
• Mandelbrot Set
• Sierpinksi Triangle
• Koch Snowflake
• Hausdorff Dimension
• Relation to Chaos Theory
• Area filling Fractals
• Applications of Fractals
• How to measure the length of a coastline
• Image compression
• Fractals and Stocks
• Use in creative media
4. WHAT ARE
FRACTALS
• Infinitely complex pattern
• Each part has the same
statistical character as the
whole.
• They are useful in modelling
things that have repeating
patterns
• Snowflakes
• Brownian Motion
• Biological systems
• Describe chaotic phenomena
• Crystal growth
• Galaxy Formation
6. FORMAL DEFINITION OF A FRACTAL
• The fractal dimension needs to exceed the topological dimension
• Topological dimension - coordinates needed to describe a point
• Fractal dimension - a measure of roughness
• Mandelbrot 1975
8. CANTOR SET
• Late 1800’s
• Infinite segments
𝑁𝑛 = 2𝑛
𝑁 = lim
𝑛 → ∞
𝑁𝑛 = ∞
• Zero Length
𝐿𝑛 =
2
3
𝑛
𝐿 = lim
𝑛→ ∞
𝐿𝑛 = 0
9. HAUSDORFF
DIMENSION
• One example of a fractal
dimension.
• Also called the similarity
dimension
• If after each step a fractal
makes 𝑚 copies of itself scaled
down by a factor of 𝑟 then we
have:
𝑚 = 𝑟𝑑
𝑑 =
ln 𝑚
ln 𝑟
10. CANTOR SET
• 𝑚 = 2 copies of the
previous step
• Scaled down by a
factor of 𝑟 = 3 so
𝑑 =
ln 2
ln 3
≈ 0.63
11. MANDELBROT SET
• Created from the logistics function:
𝑓 𝑧 = 𝑧2 + 𝑐
• The Mandelbrot set is the values for c for
which this function does not diverge
(after infinite steps) starting from 𝑧 = 0.
• Not perfectly self-similar
• Hausdorff Dimension of boundary
𝑑 = 2
12. SECTIONS OF THE
MANDELBROT SET
• Different areas have different
properties
• Main Cardioid body – set of
values c for which map has
attracting fixed point
• Period Bulb – values c for
attracting period 2 cycle
• Smaller bulb- larger cycle
13. JULIA SETS
• We defined the Mandelbrot by the logistics function. This is one
(very famous) example of a Julia Set
• The general definition is any complex, non-constant function f(z)
that is holomorphic and maps the Riemann sphere onto itself.
• This turns out to be all nonconstant complex rational functions
such that:
𝑓 𝑧 =
𝑝 𝑧
𝑞(𝑧)
where p(z) and q(z) are complex polynomials.
14. FUN JULIA
SETS
𝑓 𝑧 = 𝑧2 + 𝑐
• We know certain values c that can
generate interesting graphs.
−0.42 + 0.6𝑖
0.355 + 0.355𝑖
0.37 + 0.1𝑖
−0.54 + 0.54𝑖
• Obviously pictures are a little more
engaging
• We’ll show you where you can
generate your own later
18. KOCH SNOWFLAKE
• Finite Area – fits inside a circle
𝐴 =
8
5
𝑎0
• Same pattern on all sides
- Actually 3 fractals
- Focus on one side only
19. KOCH SNOWFLAKE
• Infinite Length 𝐿𝑛 =
4
3
𝑛
𝐿 = lim
𝑛→∞
𝐿𝑛 = ∞
• Look at one side – contains 4 copies scaled by 3
𝑑 =
ln 𝑚
ln 𝑟
≈ 1.26
20. RELATION TO CHAOS
THEORY
• Mandelbrot’s book - “Fractals and
Chaos the Mandelbrot Set and
Beyond”
• Mandelbrot set is linked to logistics
function.
• The areas “coloured-in” represent
stable periods. The end of the
cardioid is the split from one periodic
orbit to two.
• Chaos is present along the x-axis
afterwards.
27. HOW TO
MEASURE A
COASTLINE
• Should be easy?
• Length given as:
• 3,171km from the OSI
• 6,226km from Heritage Council
• 6,347km from US Defense
Mapping Agency
• Who is right?
• Depends on the line you draw
28. COASTLINE OF
BRITAIN
• Smaller the line the more accurate
the picture
• But no part is perfectly straight.
How small do you go?
• This is essentially a fractal
30. DIFFERENT METHODS
• There are a few different ways
• Manually
• Python
• We’ll start with the manual method. Programmes used were:
• Google Earth
• GIMP
• Mouse Clickr
34. VALUE OF THE DIMENSION
• This gives us a slope of m = -0.21 ± 0.06.
Our dimension is then:
• d = 1 - m = 1.21 ± 0.06
• Convert to grey image and then count the
boxes by the difference in two pixels. This
becomes dependent on the threshold you
set.
35. RESULT OF
GREY-SCALE
METHOD
• What we found was the result of our
dimension was entirely based on the
threshold we set. We plotted the result
against the threshold and obtained this
graph.
• Whilst this is interesting data, it shows that
we can’t be reliant on this image for getting
the dimension.
36. BASEMAP MODULE
• Allows you to plot areas of a map.
This is what we obtained with a
Mercator projection.
• Useful because it’s already black
and white. Threshold shouldn’t
matter anymore.
37. BASEMAP
MODULE
• If we plot the same thing
as before we get:
• Value obtained is 1.35
• Hutzler, S. (2013). Fractal
Ireland. Science Spin, 58, 19-20
• McCartney M., Abernethy G., and
Gault L. (2010). The Divider
Dimension of the Irish Coast. Irish
Geography, 43, 277-284.
40. IMAGE
COMPRESSION
• A way of reducing file size of digital
images
• Lossy v Lossless
• Fractal image compression involves
splitting an image into grids
• Create a function that goes from
compression to final image
• Associate small grids in the range
with larger grids in the domain
43. FRACTALS AND
THE STOCK
MARKET
• Efficient Market Hypothesis (EMH)
• Says that stocks always trade at their
fair value on exchanges
• Flaw is it doesn’t predict anomalies
• Fractal Market Hypothesis (FMH)
• Attempts to incorporate volatility and
chaos into the model
• In markets, stock prices act like
fractals. They appear to move in
replicating geometric patterns
• https://www.investopedia.com/terms/f/fractal-
markets-hypothesis-fmh.asp
44. CREATIVE
WORKS
• Have been used on Jackson Pollock’s
artwork to identify real works from
imitations with a 93% success rate.
• David Foster Wallace admitted that
the first draft of Infinite Jest was
based on a fractal design.
• Alexis Wajsbrot has made great use
of fractals in both Dr. Strange and
Mary Poppins Returns.
• Italian Mosaics 13th Century
• Medieval Churches
45. RECAP
General Fractals
• Cantor Set
• Mandelbrot Set
• Julia Sets
• Sierpinksi Triangle
• Koch Snowflake
Area Filling Fractals
• Peano Curve
• Hilbert Curve
Applications
• Coastline
• Image Compression
• Stock Trading
• Creative Works
46. IMAGE SOURCES
• Coastline
Lake Mead (ukga.com)
• Romanesco Broccoli
Scientific American - PDPhoto.org
• Fern
Wikipedia Commons - António Miguel de Campos
• Lightning
GeoGebra
• Mandelbrot Period
https://fractal.institute/hidden-numbers-and-basic-
mathematics-in-the-mandelbrot-set/
• Sierpinksi Antenna
V. A. Sankar Ponnapalli and P. V. Y. Jayasree
• Mandelbrot Chaos
Georg-Johann Lay. Public Domain
• Gorilla
Cincinnati Zoo
V. A. Sankar Ponnapalli and P. V. Y. Jayasree (March 31st 2018). Fractal Array Antennas and Applications, Emerging Microwave Technologies in Industrial, Agricultural, Medical and Food Processing, Kok Yeow You, IntechOpen
Take time on this slide and try to explain the concept slowly. Make sure to reinforce what was said before to allow people to make the connection.
Should be easy? Just draw a line.
Well not quite. It depends on who you ask.
Who’s right?
The answer is that is depends on how large of a line you draw and how far you zoom in.
We can see that when you vary the size of the line you get a more accurate reflection of the length of the coastline
The problem is that no stretch of coastline is going to be perfectly straight so you can continue doing this endlessly.
In this way, coastlines act like fractals and we can find the rate at which the length grows in comparison to the scale we are working at.
We need some figure to describe this rate.
There are a few different ways of achieving this. We can manually count the number of boxes on a picture with successively smaller grids or we can use python to automate some/all the work. We did both.
First let’s take a look at the manual method. This involved getting as high a res image from Google Earth as possible. It was then imported into GIMP and a grid was put over the image. The boxes were marked if the coastline crossed through them and a was used to count mouse clicks as we clicked through all the boxes.
Now let’s look at another way we can do it. We can convert an image to grey-scale and then count the boxes based on how different two pixels are in colour. This is dependent on the threshold you set. We used the same image for consistency and the converted it to grey like so:
But of course, there is a python module for everything. The one we utilised was the basemap module. It allows you to plot areas of a map. This is what we obtained with a Mercator projection.
What was useful about this is that we could set it to transparent and the “grey” version would just be either fully black or fully white. What we’d expect then is that the dimension we calculate is independent of the threshold
This is exactly what we observe. If we do the same method as before and plot the dimension against the threshold value, we see we get the same dimension value for each threshold.
This turns out to be 1.35 which is higher than what we had before but not too bad
There was a group in Trinity that did it the same manual method we did at got 1.20 and there was a group in University of Ulster who used dividers and got 1.23
Okay, so we have a value and it conforms with previous values. Well, what’s the point? Aside from the absolute riveting fun that is staring at maps for 2 hours, why do we do it.
The dimension is a measure of how rough a coastline is. We can compare this against other countries.
We can see the slope of the individual countries
West coast of Ireland v East Coast
South Africa smooth
Norway not very smooth
A lossy compression for digital images
Uses an iterated function system. Basically a set of functions that will create the image.
You separate the image into Ri blocks of size s x s
Then for each Ri search for blocks Di that are size 2s x 2s that are similar to Ri
Pick too little and it won’t look like your image. Pick too many and it’s computationally intensive
Now if people remember back to the first few lectures we were told to keep this entertaining so…
So, did you notice the Gorilla? Also, bonus marks for anyone that recognises this particular gorilla.
First thing we do is reduce the image by just averaging neighbouring pixels. This reduces the amount of computation that we need to do
The right hand image is the reduced image after being compressed and decompressed. We can see visual artificing on it. The way it’s being done is quite crude as we’re compressing and decompressing each individual RGB channel separately.
So if we convert the image to greyscale and then reduce it we can do the same process for different levels of compression and see how close it is to the original image.
The current “Main” market hypothesis is the efficient market hypothesis which says that stocks always trade at their fair value of exchanges
Gamestop and Bitcoin would probably beg to differ
Clearly this isn’t a perfect model but it’s main flaw is predicting anomalies
Alternative hypothesis is the fractal market hypothesis. This attempt to incorporate volatility and chaos into the model
In markets, stock prices act like fractals. They appear to move in replicating geometric patterns. The issue is you don’t know what time frame they’re repeating at.
There’s more information at that link.
Read
Clearly the ideas of fractals have been something that humans have found great fascination in for a lot of human history even if it’s only been mathematically rigorous for the past few decades
What we have found visually appealing is in fact mathematically complex and can used in a variety of situations.
We now invite people who have any questions and we’ll do a brief recap of what we’ve covered in case it jogs your mind.