Measuring the Unmeasurable

Dimension without coordinates

Instead of asking how many coordinates an object needs, we can ask a more physical question:

When I shrink an object, how many smaller copies are needed to rebuild the original?

Suppose every small copy has linear size $1/s$ of the original. Then

$N=s^D.$

The three symbols have very concrete meanings:

• $N$ — number of copies. How many smaller self-similar pieces make the whole?
• $s$ — scale denominator. Each copy is $1/s$ as large in length as the original. Thus $s=2$ means “half-size”; $s=3$ means “one-third-size.”
• $D$ — dimension. This is the exponent that tells us how the number of pieces grows when the measuring scale becomes finer.

SEE THE EXPONENT

One shrink factor, three familiar dimensions

In every row, each smaller piece has half the linear size of the original, so s = 2. What changes is how many directions the object extends in.

LINE2 half-size pieces

N = 2s = 22 = 2¹ → D = 1

Halving happens along one direction: length.

SQUARE4 half-size pieces

N = 4s = 24 = 2² → D = 2

Halving happens independently along two directions: width and height.

CUBE8 half-size pieces

N = 8s = 28 = 2³ → D = 3

Halving happens along three directions: width, height and depth.

THE PATTERN

1 direction → 2 pieces

2 directions → 2 × 2 = 4 pieces

3 directions → 2 × 2 × 2 = 8 pieces

So the exponent in N = s^D counts how many independent directions are being scaled. Fractals keep the same scaling idea—but the exponent need not be a whole number.

Start with familiar objects

Take a line segment of length 1. Cut it into two equal pieces. Each piece is half as long, so $s=2$, and we need $N=2$ copies:

$2=2^D \quad\Longrightarrow\quad D=1.$

That is exactly what we expect: a line is one-dimensional.

Now take a square and halve both its width and height. Four half-size squares are needed to tile the original, so $N=4$ and $s=2$:

$4=2^D \quad\Longrightarrow\quad D=2.$

A square is two-dimensional. Notice why the number of pieces jumped from 2 to 4: shrinking happens independently in two directions.

For a cube, halve width, height, and depth. Eight half-size cubes rebuild it:

$8=2^D \quad\Longrightarrow\quad D=3.$

So a cube is three-dimensional.

Object	Size of each copy	$s$	Copies $N$	Equation	$D$
Line	$1/2$	2	2	$2=2^D$	1
Square	$1/2$	2	4	$4=2^D$	2
Cube	$1/2$	2	8	$8=2^D$	3

This gives a lovely pattern:

$\text{line: }2=2^1,\qquad \text{square: }4=2^2,\qquad \text{cube: }8=2^3.$

The exponent is behaving exactly like dimension. Solving the scaling law for that exponent gives

$\boxed{D=\frac{\log N}{\log s}}.$

For ordinary Euclidean objects this simply rediscovers $1,2,3$. The surprise comes when a self-similar object needs, say, three half-size copies. Then $3=2^D$, so $D=\log 3/\log 2\approx1.585$. There is no whole-number dimension that works — and that is precisely where fractal dimension begins.

Three landmarks

NOW BREAK THE INTEGER PATTERN

The same scaling law gives fractional dimensions

Nothing new is being invented here. We keep asking exactly the same question: how many copies, and how much smaller?

Cantor set

2 copies, each ⅓ size

2 = 3^D → D ≈ 0.631

Less than a line: most of the interval disappears.

Koch curve

4 copies, each ⅓ size

4 = 3^D → D ≈ 1.262

More than a line, less than a filled surface.

Sierpiński triangle

3 copies, each ½ size

3 = 2^D → D ≈ 1.585

Too substantial for a curve, too full of holes for a surface.

Ordinary geometryD = 1, 2, 3same question →Fractal geometryD = 0.631, 1.262, 1.585…

Object	Copies $N$	Scale	Similarity dimension
Cantor set	2	$1/3$	$\log 2/\log 3\approx0.631$
Koch curve	4	$1/3$	$\log 4/\log 3\approx1.262$
Sierpiński triangle	3	$1/2$	$\log 3/\log 2\approx1.585$

A warning worth learning early

“Fractal dimension” is not one universal formula. Similarity dimension works cleanly for non-overlapping exactly self-similar constructions. Box-counting dimension asks how the number of boxes needed to cover a set grows as boxes shrink. Hausdorff dimension is the deeper measure-theoretic notion. They often agree for the classic examples in this fair, but not always.

MEASURE IT YOURSELF

Box-counting dimension: put a grid on a fractal

Shrink the boxes. Count how many touch the Sierpiński triangle. If N(ε) grows like (1/ε)^D, the slope on a log-log plot estimates D.

Grid 4 × 48 × 816 × 1632 × 32

1/ε	N(ε)	log(1/ε)	log N

The finite-stage picture makes this an estimate. As the construction and grid are refined, the slope approaches log 3 / log 2 ≈ 1.585.

The coastline paradox: measurement has a scale

A perfectly smooth curve has a stable length as the ruler becomes shorter. A rough coast behaves differently. A long ruler jumps across coves and peninsulas; a short ruler follows more of them. So asking “How long is the coast?” is incomplete until we also say “measured with what ruler?”

RULER LAB

How long is a coastline?

Measure the same jagged coast with a shorter ruler. A finer ruler slips into bays and around headlands that a coarse ruler skips, so the measured length grows.

Ruler scale

Richardson scaling: if N(ε) ∝ ε^−D, then measured length L(ε)=N(ε)ε ∝ ε^1−D. For a rough curve with D>1, smaller rulers report a longer coast.

If boxes or ruler steps of size $\varepsilon$ are needed in number $N(\varepsilon)\propto\varepsilon^{-D}$, then the measured curve length behaves like

$L(\varepsilon)=N(\varepsilon)\varepsilon\propto\varepsilon^{1-D}.$

For a smooth curve $D=1$, the exponent is zero and the length stabilizes. For a fractal-like curve with $D>1$, decreasing $\varepsilon$ can make the measured length grow. This is the coastline paradox associated with Lewis Fry Richardson and later popularized by Mandelbrot: the paradox is not that geography has no meaning, but that roughness makes length scale-dependent.

Why the logarithms?

If the count behaves approximately like $N(\varepsilon)\approx C(1/\varepsilon)^D$, taking logarithms turns a power law into an almost straight line:

$\log N(\varepsilon) \approx \log C + D\log(1/\varepsilon).$

So dimension is the slope. This is a recurring mathematical trick: logarithms turn multiplicative scaling into geometry you can see on a graph.

Now the dimension labels elsewhere in the fair are no longer mysterious decorations: they are compressed statements about how geometry scales. In the previous chapter’s Fractal Microscope, those compressed scaling statements become a visible experiment: zoom, count copies, and ask what quantity keeps returning.