self-similarity
Look closely at a fern frond and you notice each little leaflet is a miniature copy of the whole frond; zoom into a coastline on a map and the bays and headlands look much the same whether you view a thousand kilometers or ten; a bolt of lightning branches, and each branch branches again the same way. In all these the part resembles the whole. That property — a shape built from smaller copies of itself — is self-similarity, and it is the defining heartbeat of fractal geometry.
Precisely, a set is self-similar if it can be broken into pieces, each of which is a scaled-down copy of the entire set (possibly also rotated or reflected). The cleanest examples are mathematical fractals built by repeating a rule forever. The Cantor set: take the segment from 0 to 1, delete the open middle third, then delete the middle third of each remaining piece, and repeat endlessly; the leftover dust is made of two copies of itself, each shrunk by a factor of 3. The Sierpinski triangle: from a filled triangle remove the central upside-down triangle, then do the same to each of the three smaller triangles, forever; it consists of three copies of itself at half scale. This exact 'copies of copies' recipe lets you compute a fractal dimension from how many copies N appear at each scale factor s, via the relation N = s raised to the dimension.
Self-similarity is why fractals capture the roughness of the real world — clouds, mountains, blood vessels, river networks — far better than the smooth shapes of classical geometry, and it powers image compression, antenna design, and computer-generated landscapes. Two honest distinctions matter. First, EXACT self-similarity (the part is a perfect rescaled copy) is rare outside pure mathematics; nature shows STATISTICAL self-similarity — the part merely resembles the whole in its statistics, not pixel for pixel. Second, self-similarity is not the same as mere symmetry: a circle is highly symmetric but not self-similar, because zooming in reveals something new (a nearly straight line), whereas zooming into a true fractal reveals the same intricacy again, without end.
Build the Cantor set. Start with [0, 1]. Step 1: remove the middle third, leaving [0, 1/3] and [2/3, 1] — two copies of the original, each scaled by 1/3. Step 2: remove each of their middle thirds, leaving four pieces, each scaled by 1/9. Notice the pattern: at every scale factor 1/3, the number of copies is 2. Feeding N = 2 and s = 3 into N = s^d gives 2 = 3^d, so the dimension d = log 2 / log 3, about 0.63 — not a whole number, the hallmark of a fractal.
Counting copies per scale factor turns self-similarity into a number: the fractal dimension.
Real-world objects are at best statistically self-similar, and only across a limited range of scales — a coastline stops looking like a coastline once you reach grains of sand. Exact, endless self-similarity belongs to mathematical fractals, not to nature.