Why Fractals Are Everywhere in Archaeology

The Archaeology Mystery
17 followers
Dec 13, 2024
What do any of these things have to do with fractals?
If you've ever heard or seen anything about chaos theory you're probably at least somewhat
familiar with the concept of fractals. But it's probably not at all obvious what on earth fractals
could possibly have to do with archaeology. Well in fact fractals are relevant to several
aspects of archaeology and I'm going to talk about some of those today. You might be familiar
with fractals because of some of the startling images that are associated with them such as
Mandelbrot sets. But more generally fractals are something that results from simple deterministic
processes that create patterns that exhibit self-similarity at different scales. Fractals
are a feature of complex dynamical systems as found in what's popularly known as chaos theory.
Lots of things in both nature and culture have fractal properties.
For example a jagged coastline has infinite length and appears very similar in shape as
we change the scale right down to the point where the water touches the edges of pebbles
and grains of sand on the edge of the beach. If we try to measure the length of a coastline
such as the length of the main British Isles seen here the result depends on the precision
of the instrument we use. Using something like rulers to try to measure it on a map
is likely to result in a fairly low estimate of the coastline's length.
Following the coastline with a measuring wheel like this one would lead to a much greater length.
And if we followed every nook and cranny or got down on our hands and knees and tried to trace
every pebble at the water's edge we'd get a longer length still. In principle the coastline's
length approaches infinity as we use instruments with greater and greater precision and as we try
to follow the coast with greater and greater detail. The first person to recognize that
coastlines had fractal properties, in fact he invented the term fractal, was Benoit Mandelbrot,
inventor of the Mandelbrot set that you see here. Coastlines aren't the only natural phenomena that
have fractal properties. There's the winding and braiding of river channels, the branching of
broccoli florets and tree branches. Fluctuations in the weather have fractal properties,
as do fluctuations in the population of humans and animals, and even fluctuations in stock market
indices. As we'll see shortly we can also find fractals in human settlement patterns.
These are just some of the ways in which fractals can be relevant to archaeology.
When we think of things in our environment we tend to think in terms of two dimensions,
like an infinitely thin sheet of paper. When that paper has thickness or turns into a board
it has a third dimension. But what if I told you that this rather spiky looking shape,
known as a Cesaro fractal, has 1.5 dimensions? What? Even more weirdly if we connect three of
these fractals together to create an object that looks like it's enclosing some space,
it turns out to have zero volume but infinite area. To demonstrate let's look at a very simple
kind of fractal that's related to the one just described called a Koch curve. It starts with
the line segment that we divide into three equal parts. We remove the middle third and replace it
with two angled pieces that are of the same size as the piece we just removed. We now repeat this
process on each of the four line segments that are left and then repeat it again on each of the
16 line segments that result from that step. Repeating this process indefinitely results
in a Koch curve that happens to have 1.26 dimensions. And if instead of starting out
with a line segment we start with an equilateral triangle the result is a fractal called a Koch
snowflake. It also has 1.26 dimensions. So what does this fractional dimension business really
mean? Well a shape like the Koch curve cannot be one-dimensional even though it has a start and
end just like a line segment. Unlike a line segment its length is infinite and it can't
be two-dimensional either because even though it looks kind of two-dimensional it actually
encloses zero area. Consequently it has to have a dimension somewhere between one and two.
And notice by the way that it looks a little bit like a coastline.
Similarly if we look at this other fractal called a Menger sponge it looks kind of three-dimensional
but it actually can't be because despite appearances it actually has no volume.
At the same time it can't be two-dimensional because its surface area is infinite.
Consequently it has 2.7 dimensions. Effectively what the fractal dimension is measuring is how
much the degree of detail changes when we change the scale or how many self-similar
copies do we get when we go from one scale to a smaller scale. As it turns out that coastline of
the British Isles has about 1.2 dimensions while the more jagged coastlines of Norway and Greenland
have about 1.3 dimensions. And Israel's much smoother coastline would have something closer
to 1.1 dimensions. These differences among coastlines are due to variations in such
things as bedrock geology, the spacing of rivers that drain into the sea, and the effect of wave
action on erosion. Okay we're now five minutes into this video and I still haven't told you why
any of this matters. I think a good place to start is this article by Brown, Witchey, and
Leibovich which they published in 2005. One of the archaeologically important applications of
fractals that they talk about is fragmentation. One of the inescapable truths of archaeological
fieldwork is that almost none of the remains that we recover are whole artifacts or even whole bones
or whole seeds. Instead we deal with fragments. In fact in the case of pottery it's so rare for us
to find whole vessels that our usual unit of analysis is the sherd or pottery fragment.
These sherds and other kinds of fragments can sometimes be quite large and other times they
are a little bit smaller or perhaps very small. Potentially getting down to micro flakes and small
crumbs of pottery less than a millimeter in diameter. For the most part this necessity to
deal with fragments is a little bit frustrating. In the case of pottery archaeologists have had to
learn to be quite creative in how to reconstruct a whole vessel on the basis of its constituent
sherds. Whether they do that physically by actually joining the sherds together or do it on paper
or by using digital tools. However archaeologists have also learned how to yield useful information
from fragments. For example refitting flakes and their fragments to cores can tell us a lot about
the reduction sequence of lithics. While studying the size distribution of some kinds of fragments
can tell us things about site formation processes such as trampling. High traffic surfaces that have
experienced trampling by humans or their animals are likely to have on average smaller fragments.
While middens where people disposed of their trash directly are more likely to have large
fragments. You might well ask what any of this has to do with fractals but as it turns out
fragmentation has aspects that are fractal like and this is particularly true of those size
distributions. You might think that the sensible way to approach the problem of distinguishing
say a trampled surface from a midden deposit would be just to measure the maximum length of
all the pot sherds found in it or perhaps the surface area of the exterior surface of the sherds
with the idea that the average for the trampled surface would be lower than for the midden deposit.
Intuitively this may make sense but it's actually wrong. The real average size for any fragmented
remains approaches being infinitely small. In a sense that average is meaningless and that's
because it's so dependent on the size of the smallest artifacts we decide to measure. As
already noted we could in principle measure sherds down to the size of the little crumbs like the
ones in this little vial. However in practice nobody does that. Instead they might measure
every sherd bigger than two centimeters or bigger than four centimeters. The point is the lower cut
off for what deserves to be measured is completely arbitrary yet has a tremendous impact on the value
of statistics like mean and standard deviation. Since mean size won't really work then for
characterizing the degree of fragmentation in our assemblage we can turn to the fractal dimension
instead. Just as with those weird fractal shapes we saw earlier the degree of fragmentation in an
assemblage has fractal properties and therefore fractal dimension. So how do we measure that?
One option is to take our collection of fragments whether those are pottery sherds, lithic flakes,
bone fragments or something else and run them through a stack of nested sieves.
For each size category corresponding with the sieve aperture for one of the sieves
we then measure how much pottery or lithics is in that sieve. In Brown et al's article
they applied this method to an assemblage of obsidian flakes. Realizing that each sieve was
catching flakes of the aperture diameter or larger they counted how many flakes were on each sieve
but then used the cumulative distribution of flakes across the range of sieve sizes.
So in the table you see here n is the number of flakes found on each sieve
while n greater than s is the cumulative number of flakes for that sieve size or larger. They
then transformed the data into algorithms here using the natural logarithms for both size and
the cumulative number of flakes and plotted those values on a graph. They then fitted a straight
line to the dots in the graph using regression and then we can measure the slope of this line
by dividing the rise by the run. In this case because the line is declining to the right
it's a negative value. So it's a slope of minus 1.37 which corresponds to a fractal dimension of
d equals positive 1.37. We would expect this value of d to vary depending on whether each
flake tended to break into several pieces or many pieces on each breakage event. Trampling
of a flake for example is likely to crush the flake into many pieces. When archaeologists study
settlement systems of the past they're often interested in the relationships between sites
of different sizes like small villages larger towns and big cities in an urban landscape.
Quite often they try to classify sites mainly on the basis of site size.
On the assumption that bigger sites had more people probably more services and economic power
and sometimes more political power. And sometimes they look for spatial relationships among sites
that might indicate some kind of political or economic hierarchy. For example they might look
for crystalline settlement lattices. This is a model of settlement whereby a large city is
surrounded by six smaller cities each in turn surrounded by smaller towns and those are
surrounded by smaller villages. However we need to ask ourselves whether it's realistic to classify
sites into neat little packages like towns villages and cities. In many settlement landscapes
there's a continuum among sites in terms of their size population services and economic clout.
And the landscape can look really patchy with some areas quite densely populated with clusters
of settlements and other parts having much more scattered settlement. As with other phenomena
that involve changes in scale it's possible that we have fractal relationships among these
settlements. When archaeologists study ancient settlement patterns especially for societies that
had urbanism they often think in terms of the rank size relationships among the sites.
In fact a German physicist named Felix Auerbach had observed way back in 1913 that when you look
at the population sizes in modern countries the second largest city was usually about half the
size of the largest city. The third largest city was a third that size. The fourth largest city
was a quarter that size and so on. Over the next 20 years it was found that this relationship
applied to lots of other things besides cities and their rank in a settlement system and it
became known as Zipf's rule after the American linguist George Zipf who applied it to the
frequency of words in the English language during the 1930s. Going back to our settlement pattern
example it is typical to show this relationship by using logarithmic scales on both the site size
axis and the rank axis. This turns the relationship into a straight line. Notice too that the
conventional way of showing this graph is with size on the y-axis rather than the x-axis as we
saw previously. That means the slope is run over rise instead of rise over run. We can also represent
that relationship by an equation PR equals P1 over R. For the moment ignore K or assume that it equals
1. P1 is the size of the biggest site and R equals the rank. So for the second biggest site R equals
2 and its size is P1 divided by 2. In other words it's half the size of the biggest site.
Since it's nice to work with straight lines it's usual to work with the logarithms of those values
and translating that equation results in log PR equals log P1 minus K log R. For the original
applications of Zipf's rule it was assumed that K equals 1. As it turns out the basic version of
Zipf's law does not apply to all settlement patterns including a lot of archaeological ones
because the second rank site is either a little more or a little less than half the size of the
biggest site. In other words the value of K is not 1. Among archaeologists Laxton and Kavanaugh
were probably the first to notice that this relationship is a fractal one as K is just the
reciprocal of the fractal dimension D. In other words D equals 1 over K. Bill Kavanaugh applied
this kind of fractal analysis to archaeological sites that he had found in his survey of Laconia
in the Peloponnesian region of Greece. During that survey he had found dozens of sites dating
from the Neolithic period onwards and he constructed rank size graphs for each period
like this one for the early Bronze Age. Notice here that instead of showing the site size in
hectares he shows it as a proportion of the size of the largest site. As often happens with
archaeological survey data the dots don't actually make a straight line but kind of drop off rapidly
at the lower right part of the graph. That fall off among the smallest sites is no doubt due to
the fact that a lot of sites are missing from the survey simply because of sampling issues
and the fact that these small sites are often very hard to detect.
But we can fit a line to the larger sites that are probably not subject to these sampling issues
and on the basis of the slope of that line Kavanaugh was able to infer a fractal dimension
of about 1.3. I prefer to use a line that passes exactly through the largest site
and this results in a fractal dimension of around 1.67. This fractal dimension might seem pretty
esoteric but it actually tells us something about the relationship among sites in a settlement
system. For the classic modern cases that Auerbach studied where K equals 1 therefore D also equals 1
it describes a typical settlement structure in a nation-state with one very large industrial center
and a number of smaller cities each in turn surrounded by small cities towns and villages.
In the case of the early bronze age settlement system of Laconia however
the fractal dimension of somewhere between 1.3 and 1.7 tells us something very different. Namely
that the settlement system was less hierarchical with several of the larger sites being of
approximately equal size. By contrast for the classical Greek and Hellenistic periods Kavanaugh
found that the fractal dimension was less than 0.7. This suggests that the second third and fourth
rank sites are unusually small and that the largest settlement dominates the entire settlement
system. The fact that some rank size distributions can have fractal properties reminds us that
fractals might be relevant to other aspects of settlement patterns as well. So it should come
as no surprise that the spatial distribution and not just the size distribution of settlements
can also have fractal properties. In fact we can measure the fractal dimension of a settlement
system by using the box counting method. We begin by imposing a grid of large boxes over the map of
the area. Count how many squares have at least one site in them and then subdivide the squares
into smaller and smaller squares at each stage counting how many squares contain sites. In this
case I began with squares that were 16 units on a side and then eight and then four and then two
and then one. It doesn't matter what the units are because fractals are scale free. Once we have our
data with s being the size of the boxes a being the number of boxes that contain sites we can
calculate the logarithms of those two values. When we then plot a graph showing the relationship
between those two logarithms we typically get a straight line. We can find this line either by a
regression analysis or we can estimate it by i. The next step is to find the slope of the line
which we can get by dividing the rise by the run. The rise is negative because the line is actually
going down to the right and here it's minus 0.85 while the run is 0.6. The fractal dimension is just
the negative of the ratio of these two values which in this case is 1.42. Note that this indicates
that the settlement pattern is neither one-dimensional nor two-dimensional but actually
1.4 dimensional indicating that the settlement pattern is somewhere in between linear and grid
like since lines are one-dimensional. In this case the fact that most of the sites are arranged along
rivers and canals probably contributes to it being closer to a linear distribution than to a grid
like one. A common problem in archaeology is to determine the uses of ancient artifacts including
lithics. A good approach to this is to look for damage on the tool that resulted from use
and compare this to damage on tools that we've used experimentally. We can compare the character
and distribution of such things as polish and striations and scratches that occur on both the
using low power or high power microscopy. Here for example you can see the kind of polish that
occurs when the contact material is antler while this polish comes from using a flint sickle to
cut through reeds. However a downside of both low and high power microscopy is that they require
subjective assessments of the wear and polish that can be seen in the image. Several newer methods
measure surface topography or texture allowing quantitative assessment of surface wear and
polish. In this example of confocal microscopy used on the edge of a tool that was used to scrape hides
we get quantitative measurements over a small area of the tool edge.
This can facilitate statistical comparisons against other tools but these data can also be
used for a fractal analysis. A cross-section through the scratches and striations produces
a graph that looks very similar to a jagged post line or to those stock market fluctuations I
showed earlier. The tiny scratches have still tinier scratches within them showing self-similarity
at different scales. Among those who have applied fractal analysis to use wear studies are James
as in this article from 2009. Taking highly magnified profiles along short sections near
the use edge of the tool we can fit line segments to the profile much as though we were measuring
the coast of Britain with a ruler and then repeat this process with shorter line segments
and then yet again with ones that are shorter still. At each of these scales we record how
much detail these line segments are capturing. For example we can record how many of the line
segments have valleys in them meaning portions of the profile that go below the line segment.
Ideally we'd like to do this with many profiles or at least several of them in order to get a
decent sample size but this profile will serve for the purposes of demonstration.
For each segment size I counted how many segments had at least one valley in them.
Then I took the logarithms of those values and plotted them on a graph.
As with the previous examples you've seen we can then fit a line to this graph
and calculate its slope. In this case the slope was minus 0.76 so the fractal dimension is 0.76.
We would follow this process on profiles taken on tools that were used with different contact
materials in different ranges of motion to see if different combinations of contact material
and types of motion resulted in different fractal dimensions. Decorative patterns on
things like rugs and pottery can also have fractal properties although the self-similarity
may be restricted to only two or three changes of scale. Textiles tend to provide good examples of
this because weaving lends itself to having small geometrical shapes nested within larger shapes.
But decorated pottery is often also a good example whether the decoration consists of incision or
impression or painted designs. And we can measure the fractal dimension of these designs in the
same way that we measured the fractal dimension of settlement patterns using the box counting method.
Although ideally we'd want to use smaller boxes than the smallest size I use here.
Using boxes that are 16 units per side then 8 and then 4 and then 2 and then 1.
We count how many boxes include the pattern. In this case we'll count boxes that have at least
part of a dot in them. And we compile our data with s being the box size and a the number of
boxes that contain at least a part of a dot. We take the logarithms of those two values,
plot the logarithms against each other and then we try to fit a line to the dots.
In this case it doesn't make a perfect straight line but we fit it as best we can.
Preferably by regression. The slope of this line is minus 1.44 meaning that the fractal dimension
is 1.44. By coincidence this is pretty similar to the fractal dimension we had for that settlement
pattern. Just a little bit closer to being linear than to being two-dimensional. Possibly that's
because the dots are grouped into sinusoidal bands that are somewhat linear in their organization.
You might still be wondering what's the point of measuring this fractal dimension?
Well it's possible that two different potting communities making pots with substantially
similar designs might still make those designs in ways that gave them different fractal dimensions.
This might allow us to identify the products of those two different potting communities.
To summarize, fractal geometry describes many of the things that archaeologists deal with
because they're complex, rough or irregular. I only talked about a few of them here but they
include the size distributions of broken fragments, the texture of rough surfaces
even when that roughness is only visible under high magnification, the outlines of tool edges
and site size distributions. They also include the spatial aspect of settlement patterns,
the organization of decoration on things like pottery and fluctuations in things like population
density. Even though we can't measure those fluctuations directly very often, we can often
study them using agent-based modeling. Well thank you for making it to the end of this video.
I hope you've learned that there are several ways in which fractals might help us understand
such things as site formation processes, settlement patterns and even usewear on lithics.
If you'd like to learn more about these topics please check out the references I placed at the
end of the video and if you'd like to be updated when I make new videos please click on the
subscribe button down below. Thank you and stay safe.
