The paper develops a new way to characterize two important mathematical concepts, Hausdorff dimension and packing dimension, using ideas from information theory and coding. These dimensions are ways of measuring how "complex" or "spread out" a set is in space, and they generalize the familiar notion of geometric dimension. For example, a fractal like the Cantor set has a Hausdorff dimension strictly between 0 and 1, reflecting its intricate, self-similar structure. Traditionally, these dimensions are defined through coverings and measures, but the paper reframes them entirely in terms of how efficiently points in a set can be described by a coding scheme.
The key new object introduced is a "ball code," which is a systematic way of assigning binary string labels to closed balls in n-dimensional Euclidean space. Given such a code, any point in the space can potentially be described by a sequence of balls that zoom in on it, and one can measure the rate at which the length of these descriptions grows relative to the precision of the approximation. The main result is that the Hausdorff dimension of a set equals the minimum possible value, over all ball codes, of the worst-case lower description rate for points in the set. Packing dimension is recovered by looking at upper description rates instead. This mirrors a classical result in data compression theory due to Ryabko, now transplanted into continuous Euclidean geometry.
A further payoff of this framework is a clean derivation of the "point-to-set principle," a powerful tool developed by Jack Lutz and Neil Lutz that connects the dimension of a set to the computational complexity of its individual points. In the algorithmic version of the theory, one typically needs to encode the real line into sequences of symbols, which is technically cumbersome. The ball code approach avoids that encoding entirely and instead shows that replacing an optimal ball code with a computable approximation and storing it in an oracle recovers the point-to-set principle naturally. The result is a more direct and geometrically transparent foundation for studying fractal dimensions through the lens of computation and information.