fractal dimension

A line is one-dimensional, a square two, a cube three: dimensions are whole numbers, or so ordinary geometry says. But a coastline, a snowflake's edge, or the cross-section of a strange attractor is so infinitely crinkled that it seems to fill more space than a line yet less than a plane. Fractal dimension is the number, usually not a whole number, that measures exactly how thoroughly such a rough object fills the space it lives in.

The idea is to ask how the amount of detail scales with resolution. Cover the object with boxes of side epsilon and count how many, N(epsilon), are needed. For a smooth d-dimensional object N(epsilon) ~ epsilon^(-d); generalizing, the box-counting dimension is D = lim (as epsilon -> 0) of ln N(epsilon) / ln(1/epsilon), and this D need not be an integer. A classic example is the Cantor set (remove the middle third of a segment, forever): it has D = ln 2 / ln 3 = 0.6309..., more than a point but less than a line. For self-similar fractals made of m copies each scaled by a factor s, the same formula gives D = ln m / ln(1/s). Related but distinct definitions, namely Hausdorff dimension, correlation dimension, and information dimension, agree for clean self-similar sets but can differ subtly for real attractors.

Fractal dimension is what makes strange in strange attractor precise: it quantifies the self-similar, scale-invariant layering produced by the endless stretch-and-fold of chaotic dynamics. You meet it as the dimension of the Lorenz attractor (about 2.06), the Hénon attractor (about 1.26), coastlines, turbulence, and porous media. A non-integer dimension is the fingerprint of structure at every scale: no matter how far you zoom in, new detail appears. One caveat worth flagging: the several fractal dimensions (Hausdorff, box-counting, correlation) coincide for idealized self-similar constructions but are not identical in general, so a stated fractal dimension should specify which one is meant.