Sinai's Positive Metric Entropy Conjecture for the Chirikov Standard Map

Sinai's Positive Metric Entropy Conjecture for the Chirikov Standard Map

OpenAI's claim

For every sufficiently large positive parameter, the torus standard sine map is claimed to have positive entropy for normalized area, giving a full parameter tail. The cited manuscript argues for that result; its relation to this problem is specified below.

Mathematical statement of the claimed result

Mathematical claim 1.

On $\mathbb T^2=\mathbb R^2/\mathbb Z^2$ with normalized Lebesgue area $m$, define the standard map \[ f_k(x,y)=(x+y+k\sin(2\pi x),\ y+k\sin(2\pi x))\pmod{\mathbb Z^2}. \] OpenAI claims some $k_0>0$ with $h_m(f_k)>0$ for every real $k\ge k_0$. The entropy is Kolmogorov--Sinai entropy for the fixed invariant area measure; equivalently the largest Lyapunov exponent takes positive values on a subset with area greater than zero. This is not only a topological-entropy or zero-area hyperbolic-set assertion.

Outline of the argument

Manuscript 1 (family 146, 2026-09-23): Logarithmic matrix-product growth is treated as a stationary time semimetric whose expected shortfall is monitored across scales. Transverse graph families from finite Dirichlet problems give a weighted product estimate across fast bridges without assuming independent orbit segments; the multiscale argument forces positive average derivative growth.

Formal verification

For family 146, the scope page documents the following selections. Selections include positive entropy throughout a sufficiently large parameter tail, positive Lyapunov growth on positive area, and a hyperbolic ergodic component with finitely many cyclic Bernoulli pieces. The documented statement covers the main claim of manuscript 1. Its selected declaration also covers the clause used for this problem. Comparator files: StandardMapComponents.lean, StandardMapEntropy.lean, StandardMapLyapunov.lean. Declarations: OAI.StandardMapEntropy.main_components, OAI.StandardMapEntropy.main_entropy, OAI.StandardMapEntropy.main_entropy_complete.

This site has not executed the Lean files or checked their dependencies. These descriptions report the scope of the linked declarations, rather than a new verification of the proofs.

Status caveat

OpenAI attributes these results to an internal model. Its release notice cautions that results lacking full formalization can contain mistakes. This site has not independently assessed the proofs; a solved label records the scope of the reported claim.

Sources

• OpenAI, Positive Metric Entropy for the Standard Map at Large Parameters, 2026-09-23; citation key OAI:Positive-Metric-Entropy-for-the-Standard-Map-at-Large-Parameters-September-23-2026. PDF (GitHub); PDF (direct); Manuscript page and citation; Latest version; Statement source: main:entropy.
• Family 146: Lean scope.
• Comparator StandardMapComponents.json.
• Comparator StandardMapComponents.lean.
• Comparator StandardMapEntropy.json.
• Comparator StandardMapEntropy.lean.
• Comparator StandardMapLyapunov.json.
• Comparator StandardMapLyapunov.lean.
• OpenAI catalogue entry.

Sinai's Positive Metric Entropy Conjecture for the Chirikov Standard Map

Definitions and mathematical statement

On $\mathbb T^2={\mathbb{R}}^2/{\mathbb{Z}}^2$, define the area-preserving standard map \[ f_k(x,y)=\left(x+y+k\sin(2\pi x),\ y+k\sin(2\pi x)\right)\pmod{{\mathbb{Z}}^2}. \] Let $m$ be normalized Lebesgue area. Its metric entropy $h_m(f_k)$ measures the asymptotic information rate of measurable partitions under iteration, taking the supremum over finite partitions. The conjectured assertion is \[ \operatorname{Leb}\{k\in{\mathbb{R}}:h_m(f_k)>0\}>0. \] For this smooth area-preserving map, the entropy formula identifies positivity with a positive-$m$-measure set of points having positive maximal Lyapunov exponent, where that exponent is the long-time exponential growth rate of tangent vectors. Positive topological entropy on a zero-area invariant set is not sufficient. The parameter normalization here is exactly the one in the catalogue.

Formulation and concept references: [S1].

Short English statement

For a positive-measure set of parameters, does the standard map exhibit chaotic tangent growth on a positive-area set of trajectories?

Research attempt

Target and source audit. The measure in this question is the fixed area measure $m$, not a measure supported on a horseshoe. Both the phase-space and the parameter-space positivity requirements must be met. The primary paper [E1] constructs positive metric entropy by conservative perturbations, including perturbations near standard maps. Its perturbations are not required to stay in the one-parameter family $f_k$. The more recent standard-map result [E2] concerns topological entropy. Neither statement supplies the displayed target. This attempt gives explicit tangent estimates and identifies the precise infinite-time measure estimate that those estimates fail to provide.

Normalization, elementary dynamics, and a zero-entropy endpoint. Put $a(x)=2\pi k\cos(2\pi x)$. Direct differentiation gives \[ Df_k(x,y)=\begin{pmatrix}1+a(x)&1\\a(x)&1\end{pmatrix},\qquad \det Df_k=1. \] The inverse is obtained from $x=X-Y$, then $y=Y-k\sin(2\pi(X-Y))$. Thus the formula indeed defines a smooth area-preserving diffeomorphism on the torus. The translation $C(x,y)=(x+1/2,y)$ satisfies $C^{-1}f_kC=f_{-k}$, so entropy is an even function of the parameter. For $k=0$, \[ f_0^N(x,y)=(x+Ny,y),\qquad Df_0^N=\begin{pmatrix}1&N\\0&1\end{pmatrix}. \] The derivative grows at most linearly. Every Lyapunov exponent is zero, and the smooth area entropy formula stated in the problem gives $h_m(f_0)=0$. Consequently a proof cannot consist of an argument claiming positivity for all parameters without an exception.

The points $(0,0)$ and $(1/2,0)$ are fixed. Their derivative traces are $2+2\pi k$ and $2-2\pi k$. For $k>0$ the first is hyperbolic. For $0<k<2/\pi$ the second is elliptic at the linear level. Interchanging their roles handles negative $k$. This proves existence of exponentially growing directions on a periodic orbit for every nonzero parameter. A finite orbit has area zero, so even this exact exponential growth contributes nothing to the integral of the positive exponent against $m$.

A quantitative cone calculation. Use the integral linear torus coordinates $(q,r)=(x,x-y)$. In these coordinates the map is \[ F_k(q,r)=(2q-r+k\sin(2\pi q),q),\qquad DF_k=\begin{pmatrix}A(q)&-1\\1&0\end{pmatrix},\quad A(q)=2+2\pi k\cos(2\pi q). \] Fix $K>2$ and let $G=\{|A(q)|\geq K\}$. On the cone $\mathcal C=\{(u,v):|v|\leq|u|\}$, with the maximum norm, \[ |Au-v|\geq (K-1)|u|, \qquad |u|\leq\frac{|Au-v|}{K-1}. \] Thus $DF_k$ maps $\mathcal C$ into its interior and expands each vector in that cone by at least $K-1$. The sign of $A$ is irrelevant to this estimate. If all the first $N$ iterates of a point lie in $G$, induction gives \[ \|DF_k^N(q,r)(1,0)\|_\infty\geq(K-1)^N. \] The coordinate change is fixed and invertible, so it does not alter Lyapunov exponents.

This yields a clean sufficient condition. If a measurable invariant set $E\subset\bigcap_{j\in\mathbb Z}F_k^{-j}G$ has $m(E)>0$, then almost every point of $E$ has maximal exponent at least $\log(K-1)$, by Oseledets' theorem and the displayed expansion. The entropy formula would give \[ h_m(f_k)\geq m(E)\log(K-1)>0. \] The external ergodic theorems here apply because the map and its inverse are smooth on a compact manifold. The missing assertion is existence of such a positive-area invariant set, not the cone estimate.

Why the critical strip cannot just be discarded. Write $b=2\pi|k|$. When $b\geq2(K+2)$, the bad strip $B=\{|2+b\cos(2\pi q)|<K\}$ has area \[ \varepsilon_k=\frac1\pi\left[ \arccos\!\left(\frac{-K-2}{b}\right) -\arccos\!\left(\frac{K-2}{b}\right)\right] \leq\frac{4K}{\pi\sqrt3\,b}. \] Indeed both arguments belong to $[-1/2,1/2]$, where the absolute derivative of $\arccos$ is at most $2/\sqrt3$, and their separation is $2K/b$. The sign change for negative $k$ is absorbed by translation in $q$. Since area is preserved, the union bound gives \[ m\left(\bigcap_{j=0}^{N-1}F_k^{-j}G\right) \geq 1-N\varepsilon_k. \] For a large fixed parameter this certifies exponential tangent growth over a long finite window for most initial points. It becomes vacuous when $N\varepsilon_k\geq1$. Taking $k$ large after fixing $N$ does not prove an infinite-time assertion at any fixed $k$. Interchanging those quantifiers is exactly the error an apparent elementary proof would make.

Even a small frequency of visits to $B$ is insufficient without control of their effect on directions. An expanding direction can rotate into a contracting direction. For example, with \[ D=\begin{pmatrix}\lambda&0\\0&\lambda^{-1}\end{pmatrix},\qquad R=\begin{pmatrix}0&-1\\1&0\end{pmatrix},\quad\lambda>1, \] one has $DRDR=-I$. Each factor $DR$ has operator norm $\lambda$, while their product has norm one. This is not a model orbit of the standard map and is not a counterexample to the conjecture. It is an exact countercheck to the algebraic inference that multiplying large one-step norms gives a lower bound on the norm of the product.

The direction of the subadditivity bound. Let \[ a_N=\int_{\mathbb T^2}\log\|Df_k^N(z)\|\,dm(z) \] for the Euclidean operator norm. The chain rule and invariance of $m$ imply $a_{N+M}\leq a_N+a_M$. The subadditive ergodic theorem therefore identifies the integral of the maximal exponent with \[ \int\lambda_+(z)\,dm(z)=\inf_{N\geq1}\frac{a_N}{N}. \] Here determinant one makes the maximal exponent nonnegative. A computed positive value of $a_N/N$, even if rigorously enclosed for a large $N$, is an upper bound on this infimum. It is not a lower bound. To prove positivity by this route one needs a uniform positive lower bound for all sufficiently large $N$, or a separate mechanism controlling cancellations at successive scales. Sampling finite-time exponents and observing stable positive numbers does not supply that uniform estimate.

There is also a parameter issue after phase-space positivity is established. A proof for one special $k$ would be meaningful progress, but would not by itself prove positive Lebesgue measure of such parameters. Metric entropy is not automatically lower semicontinuous in the required form. A positive entropy perturbation in the space of all area-preserving diffeomorphisms can leave the standard-map curve. Its existence cannot be converted into a set of parameters by projecting a neighborhood onto that curve.

A possible route and the first unproved step. One could try to supplement the good-strip cone estimate with a return map that groups a visit to $B$ together with its recovery interval. A sufficient construction would give a positive-area invariant domain, an induced map with integrable return time, and a measurable cone field expanded by a factor larger than one on average over those returns. The recovery estimates must track the direction, not just the orbit's distance from the critical curves. The elementary bounds above supply neither a positive-area domain nor integrability and positive mean expansion for this induced system. They also do not make those properties persist for a positive-measure parameter set.

Verification and outcome. The saved diagnostic checks determinant and trace identities, the cancellation matrix product, and sample cone inequalities exactly for integer data. It also evaluates the critical-strip area formula numerically at selected parameters. These finite checks test constants and signs, not asymptotic entropy. The result of this attempt is a precise sufficient invariant-set criterion and an explanation of why predominantly expanding one-step derivatives fail to verify it. The required positive-measure set of parameters has not been constructed.

Sources

• S1 A. Gorodetski and V. Kaloshin, "Conservative Homoclinic Bifurcations and Some Applications," Proc. Steklov Inst. Math. 267 $2009$, 76-90. DOI:10.1134/S0081543809040063. https://www.math.uci.edu/~asgor/Psim076.pdf. Formulation source.
• E1 Pierre Berger and Dimitry Turaev, On Herman's Positive Entropy Conjecture. https://arxiv.org/abs/1704.02473. Primary perturbation result; perturbations are not confined to the standard-map parameter family.
• E2 Fernando Oliveira, Positive topological entropy for the Standard Map $2024$. https://arxiv.org/abs/2404.02209. Primary topological-entropy result; it does not establish positivity for Lebesgue metric entropy.