Positive metric entropy for the standard map

The explanation

Picture a square whose edges wrap around, like an old video game screen. Leave by the right edge and you come back on the left. Leave by the top and you return at the bottom. Mathematicians call this surface a torus. The standard map moves every point of it by one simple rule. Pick a number k. Add k times sin(2πx) to the height y. Then add the new height to the position x, and wrap around.

Take k = 10 and the point x = 0.1, y = 0.2. Follow a second point a tiny distance to its right. After one step, the gap between them is about 72.6 times larger. After seven steps, it is about 10 billion times larger. That averages to about 27 times per step. But the growth is uneven. On the third step, the gap grew only about 3.3 times. That step began near x = 0.25, where the sine curve is flat at its peak. There the push barely stretches anything.

This mix is typical of large k. The paper describes strong local stretching combined with repeated losses of that stretchingSource 1. The long-run growth rate of such gaps, measured as a logarithm, is called the Lyapunov exponent. A positive exponent means tiny gaps grow exponentially, which is the usual signature of chaos. The map also keeps area unchanged: a patch of the square moves around but never shrinks or grows.

Metric entropy measures how much unpredictability the map creates, averaged over the whole area. A classical formula of Pesin turns it into the average Lyapunov exponent over the square. So the entropy is positive exactly when the exponent is positive on a region of positive area. Chaos on a set with no area does not count. The point x = 0, y = 0 shows why this matters. It never moves, and gaps near it stretch about 64.8 times per step in one direction. Yet one point has no area at all.

Much chaos of this zero-area kind was already known. In 1994 Duarte built such chaotic sets for large k, and also found islands where motion stays orderly. In 2012 Gorodetski found chaotic sets of full dimension for a large but possibly zero-length set of big k values. The paper notes that neither result gives positive areaSource 2. In 2024 Oliveira studied topological entropy, a count of distinct behaviours. Oliveira proved it positive for every k > 0 other than k = 2/π, a value that result leaves out. That still says nothing about entropy measured by area. Other work added noise, coupling or small changes to the map. In a form stated by Berger and Turaev, Sinai's conjecture asks for positive area entropy on a set of k values of positive total length.

This paper proves that conjecture for the exact map, in a stronger formSource 3. There is a threshold k₀ such that every k at or above k₀ has positive metric entropySource 4. It also finds, for each such k, a region of positive area that the map explores as one indivisible whole. That region splits into finitely many pieces that the map visits in turnSource 5. On each piece, a fixed number of steps behaves as randomly as repeated coin tossing.

The one obstacle is cancellation. Stretching built up over a long time can be undone later. The paper shows this is rare. Two long stretches that each grow near the top rate seldom cancel when joined. The chance falls exponentially with their lengthSource 6. A second estimate controls fast-growing stretches without assuming that distant times act independentlySource 7. The proof then argues by contradictionSource 8. Suppose arbitrarily large k had zero exponent almost everywhere. The two estimates make a bookkeeping identity across many time scales impossible.

The theorem has clear limits. The contradiction shows that k₀ exists, but the paper does not give a value for itSource 9. Nothing is said about smaller k. The paper also makes no claim that the exponent is positive almost everywhereSource 10. So it does not rule out orderly regions, and it gives no uniform lower bound on entropy.

How sure is this? A computer proof checker called Lean has checked formal versions of three statements. They are the main entropy theorem, positive Lyapunov exponents on positive area, and the coin-tossing region. The Lean scope note says no special assumption on k is used. No specialist review of the paper has been published.

What it doesn't show. This does not say how large k must be, or that the map is chaotic on the whole square.