One observable whose time shifts describe an entire smooth system

← All digests

Family 144 · Dynamical systems and ergodic theory
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-21

1. The question

Imagine observing a deterministic system through one numerical measurement. Record the measurement yesterday, today, tomorrow, and at every other integer time. Could these shifted measurements form an orthonormal basis for every square-integrable observation of the system, after subtracting its average?

The basis requirement has two parts. Different time shifts must be uncorrelated and have unit variance. Yet their linear combinations must also approximate every centered observation, not merely some small collection of signals.

This is a concrete way to formulate the probability-preserving version of Banach’s simple Lebesgue-spectrum problem. Family 144 claims such a system can be realized by an infinitely differentiable, volume-preserving map of the three-dimensional torus. The invariant probability is ordinary smooth volume, rather than a specially chosen measure supported on a complicated subset.

The claimed example would combine strong statistical decorrelation with zero measure-theoretic entropy and zero Lyapunov exponents. Those properties measure different aspects of dynamics, so understanding their coexistence is part of the interest.

2. Background and essential definitions

From trajectories to an operator

Write $\mathbb T=\mathbb R/\mathbb Z$. The three-torus $\mathbb T^3$ is a unit cube with opposite faces identified. Let $\mu$ be its normalized volume, so $\mu(\mathbb T^3)=1$.

A volume-preserving diffeomorphism $T$ is a smooth invertible map with smooth inverse that preserves this measure. An observable is a measurable function $g$ on the space. The square-integrable observables form the complex Hilbert space $L^2(\mu)$, with

$$\langle g,h\rangle=\int g\overline h\,d\mu, \qquad \|g\|_2^2=\int |g|^2\,d\mu.$$

The Koopman operator describes how an observable changes after one step:

$$U_Tg=g\circ T.$$

It is unitary because $T$ preserves volume and is invertible. Instead of following points, spectral theory studies this linear operator on functions.

Constants are always fixed by $U_T$. They must be separated off when asking for a purely Lebesgue model. The relevant space is

$$L^2_0(\mu)=\left\{g\in L^2(\mu):\int g\,d\mu=0\right\}.$$

What simple Lebesgue spectrum means

On the unit circle $S^1\subset\mathbb C$, let $m$ be normalized arc length. The model operator is multiplication by the circle coordinate:

$$(Mg)(w)=w\,g(w),\qquad g\in L^2(S^1,m).$$

Its powers take the constant function one to $w^n$, for every integer $n$. These functions form the usual Fourier orthonormal basis.

To say that $U_T$ has simple Lebesgue spectrum on $L^2_0(\mu)$ means that an inner-product-preserving change of Hilbert-space coordinates identifies it with this one-copy model. “Lebesgue” specifies arc-length spectral type; “simple” means multiplicity one.

Equivalently, there is a function $f$ for which

$$\langle f\circ T^n,f\circ T^k\rangle= \begin{cases}1,&n=k,\\0,&n\ne k,\end{cases}$$

and the closed linear span of all these shifts is the entire mean-zero space. Indices run through positive and negative integers: this is a bilateral orbit.

Uncorrelated measurements need not be independent random variables. The basis statement is about Hilbert-space geometry and linear approximation, not a claim that repeated observations are independent coin tosses.

3. Development history and previous partial results

The manuscript distinguishes its probability-preserving question from the historical real-line formulation recorded by Ulam and attributed to Banach. Rokhlin explicitly asked for ergodic probability-preserving automorphisms with simple, or at least finite-multiplicity, Lebesgue spectrum. The setting and multiplicity are part of the problem, not optional details.

Earlier constructions achieved pieces of the desired picture. Helson and Parry constructed cocycles whose associated unitary operators have Lebesgue spectrum. A cocycle adds an auxiliary motion over a base transformation; the resulting operator need not already be the complete centered Koopman operator sought here. Mathew and Nadkarni constructed a probability-preserving transformation with a Lebesgue component of multiplicity two.

Guenais related a cocycle version of the question to flat polynomials and obtained a countable abelian group action with simple spectrum of mixed discrete and Haar type. Such results show why both the full space and the precise acting group matter.

Within smooth dynamics, Fayad constructed smooth volume-preserving torus transformations with simple spectrum. That alone did not identify the spectral type as Lebesgue. A separate infinite-measure construction discussed by el Abdalaoui likewise does not directly solve the probability-space problem.

Continuous-time flows provide another important comparison. Prikhod’ko constructed finite-measure flows with simple Lebesgue spectrum. Fayad, Forni, and Kanigowski, and later Abdedou, Fayad, and Kessi, obtained smooth or analytic flows with countably infinite Lebesgue multiplicity in specific torus settings. Sampling a simple Lebesgue flow at a fixed nonzero time does not preserve simplicity: the real frequency line wraps around the circle repeatedly, producing countably many spectral copies.

Finally, successive smooth coordinate changes have a long history in the approximation-by-conjugacy method of Anosov and Katok. The new manuscript uses explicit transitions between smooth integrable maps and supplies its own quantitative spectral control. Its claimed advance is the conjunction of all the desired features: a single discrete-time transformation, finite invariant volume, smooth realization, and multiplicity one on the whole centered space.

4. What the new manuscripts claim

The family contains one manuscript. Its main theorem asserts the existence of a $C^\infty$ volume-preserving diffeomorphism

$$T:\mathbb T^3\longrightarrow\mathbb T^3$$

and a real-valued $f\in L^2_0(\mathbb T^3,\mu)$ such that

$$\{f\circ T^n:n\in\mathbb Z\}$$

is an orthonormal basis of the complex mean-zero Hilbert space. Completeness uses complex linear combinations even though the generator itself is real. The measure is standard normalized volume.

The observable $f$ is only asserted to be square-integrable. The theorem does not claim that the generating observation is smooth or bounded merely because the underlying map is smooth.

The manuscript derives ergodicity and ordinary mixing. Ergodicity means every invariant measurable set has measure zero or one. Mixing means, for fixed measurable sets $A,B$,

$$\mu(A\cap T^{-n}B)\longrightarrow\mu(A)\mu(B) \qquad(|n|\to\infty).$$

It also states zero Kolmogorov–Sinai entropy and three vanishing Lyapunov exponents at almost every point. The former says that the asymptotic information produced per step by any finite measurable observation partition is zero. Lyapunov exponents measure exponential rates of infinitesimal stretching along typical trajectories; all three rates are claimed to vanish.

Mixing of every finite order is an additional corollary using the separate claim in family 145. It does not follow merely by replacing two observables with several in the spectral calculation. The ordinary mixing, entropy, and Lyapunov conclusions do not depend on that companion theorem.

5. Why the advance matters

Provisional importance: 5/5. The proposed construction answers a central existence problem in spectral ergodic theory in a strong smooth setting. It would identify a deterministic volume-preserving system whose entire centered operator theory is one copy of the basic Fourier shift.

The requirement of completeness is crucial. Finding one observation with flat spectral density would give a Lebesgue component, but other observations could live in additional spectral components. The claimed theorem eliminates that hidden remainder.

The example also sharpens the distinction between statistical and geometric notions of complexity. Long-separated observations can decorrelate even when typical infinitesimal separations have no positive exponential growth rate. Zero entropy likewise does not mean that every orbit repeats or that every correlation persists.

6. The main idea

Begin with maps whose spectra can be calculated

The approximating maps are integrable twists. In suitable volume-preserving coordinates $(\theta,u,a)$, they take the form

$$S(\theta,u,a)=(\theta+\varphi(u),u,a).$$

The angle $\theta$ rotates at a speed depending on $u$. A finite angular Fourier signal has pieces of the form $A(u,a)e^{2\pi i m\theta}$. On an arc where the speed varies monotonically, the spectral density of such a piece can be read from how its spatial mass is distributed over rotation speeds.

The paper groups these finite signals according to which target observation they are meant to encode. These labels keep different approximation tasks distinguishable during subsequent changes of coordinates.

Change coordinates while retaining information

A carefully chosen smooth passage replaces a twist by a nearby one whose angular coordinate plays a different role. Oscillatory-integral estimates describe where the old signals move among the new angular frequencies.

The third coordinate supplies a small variation in speed. Large integer parameters separate the frequency bands associated with different labels while leaving the earlier approximation as accurate as needed. At each stage only finitely many signals and error requirements are involved; a parameter can then be chosen sufficiently large for that stage.

This is more than convergence of maps alone. The construction must preserve useful spectral information while the coordinate systems and frequencies change substantially.

Encode labels, then flatten the density

The manuscript introduces small, controlled biases in the spectral densities belonging to the labels. Many such biases form a block. Together they make it possible to predict which target label a signal carries, while spending only a small amount of a separate spectral error budget.

The total density is kept close enough to uniform that whitening, a spectral rescaling that makes it exactly uniform, changes the observable little in physical $L^2$ norm. Flat density gives orthogonality of its distinct time shifts. Accurate prediction of labels makes linear combinations of those shifts approximate the selected target observations.

A diagonal construction repeatedly revisits a countable dense collection of centered observables. Errors shrink while the maps and their inverses converge smoothly. In the claimed limit, orthogonality survives and every member of the dense collection lies in the closed orbit span. The latter step supplies completeness, which is the part that a single flat signal would not establish by itself.

Read the dynamical consequences from the model

In the circle model, correlations become Fourier coefficients of integrable functions. Their decay gives ordinary mixing by the Riemann–Lebesgue lemma. Finite spectral multiplicity then invokes Rokhlin’s entropy theorem to obtain zero measure entropy.

For a smooth map preserving volume, Pesin’s entropy formula relates entropy to the positive Lyapunov exponents. Zero entropy rules out positive exponents almost everywhere, and volume preservation makes their sum zero. The manuscript concludes that every exponent vanishes. These are consequences of the claimed construction combined with the cited general theorems, rather than extra properties imposed separately at every approximation stage.

7. Scope, limitations, and open questions

The theorem is an existence statement on $\mathbb T^3$. It does not provide a short closed formula, establish real analyticity, or claim realization on every compact manifold or in every dimension.

Its spectral assertion excludes constants and includes all remaining square-integrable observables. The generator’s low regularity is compatible with a smooth underlying map. No independence or Gaussian distribution is asserted for its time shifts.

Zero measure-theoretic entropy refers to the specified invariant volume. It should not be silently replaced by a statement about topological entropy, which measures a different form of orbit complexity. Likewise, vanishing Lyapunov exponents do not rule out slower-than-exponential separation.

The stronger multiple-mixing corollary explicitly uses the separate theorem of family 145. Keeping this dependency visible distinguishes the principal construction from conclusions that require another manuscript.

8. Sources and further reading

Primary historical sources include Helson and Parry, Cocycles and spectra, Arkiv för Matematik 16 (1978), 195–206; Mathew and Nadkarni, A measure preserving transformation whose spectrum has Lebesgue component of multiplicity two, Bulletin of the London Mathematical Society 16 (1984), 402–406; and Fayad, Partially mixing and locally rank one smooth transformations and flows on the torus, Journal of the London Mathematical Society 64 (2001), 637–654.

For the distinct infinite-measure setting, see el Abdalaoui’s Spectral ergodic Banach problem and flat polynomials. Prikhod’ko, On ergodic flows with simple Lebesgue spectrum, Sbornik: Mathematics 211 (2020), 594–615, concerns continuous time. Optional background on the Koopman operator explains why spectral questions about functions encode properties of a nonlinear dynamical system.

Source manuscripts

• M1. OpenAI, A smooth three-torus diffeomorphism with simple Lebesgue spectrum (September 23, 2026).

Sources consulted on 21 October 2026 at repository revision 1581d4c. Download the manuscript citations.