Brownian motion spreads particles with Gaussian steps. The distribution widens monotonically; entropy increases monotonically. Replace Gaussian steps with geometrically shrinking discrete steps — fixed-size jumps that decay by a ratio r at each step — and the story changes. The resulting distributions are Bernoulli convolutions: bounded, fractal-structured probability measures whose properties depend sensitively on the contraction ratio.
Note (2026-11-08): the abstract says that the coarse-grained Shannon entropy of the distributions from random walks with shrinking steps (Bernoulli convolutions) has a local maximum at the dyadic ratio 1/2, from the competition between diffusive spreading and emergent fine structure, and that this is a general property of systems driven by non-Gaussian discrete noise whose dynamics near stable fixed points are autoregressive; it also discusses implications for protocell self-replication and vesicle proliferation. The essay's description of the regimes (low entropy below 1/2, and above 1/2 a distribution 'confined to a narrower support') and its remark about maximum information content at the transition from fractal to smooth are not in the abstract, and the narrower-support statement looks inconsistent with steps of size r^k, whose total span grows with r (my own check, not from the paper). The essay also drops the words 'coarse-grained' and 'local'.
Near the dyadic ratio r = 1/2, the Shannon entropy of the particle distribution peaks (arXiv:2603.00317). This is surprising: the entropy is not monotone in the contraction ratio. Two competing effects cross at r = 1/2: diffusive spreading (which increases entropy) and emergent fine structure (which decreases it). The peak marks their balance point.
The structural observation: entropy can be non-monotone in a natural parameter of the process because the parameter controls two opposing effects simultaneously. The contraction ratio is not a simple “more disorder” knob — it governs both the spread of the distribution and the regularity of its internal structure. At r < 1/2, the fine structure dominates and entropy is low. At r > 1/2, the distribution becomes more uniform but is confined to a narrower support. The peak at the dyadic point is where the trade-off is sharpest — maximum information content at the transition between fractal and smooth.