Zero entropy does not guarantee a smooth positive-volume model

The explanation

Flip a fair coin once a second. Each flip brings one bit of new information, since heads and tails are equally likely. A coin that lands heads nine times in ten brings less, about 0.469 bits per flip. Now watch a clock hand that jumps by the same angle every second. Note only which quarter of the dial it sits in. A long record of quarters soon becomes predictable, and new readings add almost nothing. Kolmogorov–Sinai entropy is this long-run rate of new information per step. The coin machine scores 1 bit, and the clock hand scores 0.

Mathematicians study such systems in an abstract form. A measure-preserving transformation is a rule that moves the points of a space each step. It keeps the probability of every region unchanged. The abstract form forgets shape and distance. It only remembers which sets have which probability. Two systems count as the same, or measurably conjugate, when a relabeling of points turns one into the other. The relabeling may ignore sets of probability zero, and it need not be continuous.

The clock hand also has a smooth picture. Its dial is a compact manifold, a smooth bounded shape of finite dimension, like a circle or a sphere. The jump is a diffeomorphism, a smooth map whose reverse is also smooth. It keeps arc length, which is a smooth positive volume. That means a smooth way to measure size that gives every small region some weight. Call such a picture a smooth positive-volume model. The smooth realization problem asks which abstract systems have one.

Some limit was known. A smooth map on a compact shape cannot create information without bound. Ruelle's inequality from 1978 gives finite entropy from limits on how much the map stretches. So finite entropy is necessary. The open question was whether it is enough once the kept measure must be a smooth positive density.

Earlier results came close. In 1977, Lind and Thouvenot copied every finite-entropy ergodic system by continuous, area-preserving maps of a torus, a doughnut-shaped surface. Ergodic means the system cannot be split into two separate parts, each with some probability, that each stay put. Those maps need not be smooth. They also used smooth torus maps. A standard example of such a map sends (x, y) to (2x + y, x + y), wrapped around. That map stretches one direction by about 2.618 each step. With a suitable spread-out measure, it copies every system with entropy below its own, about 1.39 bits. In 2016, Quas and Soo extended this to more torus maps. The catch is the measure these copies keep. It may be singular, piled onto thin sets, instead of a smooth volume. In 1997, Katok and Thouvenot found an obstruction for pairs of commuting maps. That did not settle the question for one map.

The paper claims to prove that finite entropy is not enough. It goes further. It builds one ergodic system with zero entropy, like the clock hand, that has no smooth positive-volume modelSource 1. The model may use any finite dimension. Its shape may have a smooth edge. It need not be orientable, meaning it may lack a consistent sense of clockwiseSource 2. The system is fixed first, and only then is any smooth model tested. So a single example defeats every dimension at onceSource 3.

The obstacle is the relabeling. It keeps probabilities but may scramble distance and shape completely. So the proof needs a feature of smooth volume that survives any such scramblingSource 4. The paper finds a switching property. Cut the shape into small cells. Pick a cell by its volume, then two random points x and y inside it. Usually some third point z has a future that tracks x and a past that tracks y. On those pairs, the points z are never spread more than twice as thickly as the volumeSource 5. The number of cells may grow with time, but only at a rate that fades toward zero over the scales.

The abstract system is built to fail this test. Its points are endless strings of symbols, made of words. Each long word is a chain of shorter words, and each word carries a labelSource 6. Certain pairs of labels must pass a fixed compatibility test. Two independent histories rarely pass most of these testsSource 7. A smooth switch would force them to match one legal history, which does passSource 8. At well-chosen scales, these two conclusions clash.

The result needs the kept measure to be a smooth, strictly positive density. Foreman and Weiss discuss a looser version, where the measure only has the same zero-size sets as volume. The paper does not claim an obstruction for that versionSource 9.

How sure is this? A computer proof checker called Lean has checked a formal version of a weaker statement. It builds such a system with finite entropy, and no model in any finite dimension. Shapes with a smooth edge are included. The paper's stronger zero-entropy conclusion is outside that check. No specialist review has been published.

What it doesn't show. This does not rule out smooth copies whose kept measure only shares volume's zero-size sets.