pith. sign in

arxiv: 2605.18425 · v1 · pith:CFOWCKVDnew · submitted 2026-05-18 · 💻 cs.LG · math.ST· stat.TH

Generative Adversarial Learning from Deterministic Processes

Joris C. K\"uhl , Hanno Gottschalk This is my paper

Pith reviewed 2026-05-20 12:49 UTC · model grok-4.3

classification 💻 cs.LG math.STstat.TH
keywords generative adversarial networksinvariant distributionchaotic dynamical systemstime seriesJensen-Shannon divergenceergodicitymixing propertiesphysical AI
0
0 comments X

The pith

An infinite-dimensional generative adversarial model can learn the invariant distribution of a sufficiently chaotic dynamical system from a single deterministic time series.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that generative adversarial learning does not require independent and identically distributed samples when the underlying data comes from a chaotic dynamical system. A single long trajectory suffices if the system has strong enough mixing or ergodic properties to make one orbit visit states densely and reveal the unique invariant distribution. The authors introduce an infinite-dimensional version of GAN training and prove that the learned distribution converges to the true invariant one at explicit rates measured by Jensen-Shannon divergence. This supplies a statistical-learning explanation for why such methods work on non-random physical data such as turbulence measurements.

Core claim

It is possible, using an infinite-dimensional model of generative adversarial learning, to learn the invariant distribution of a sufficiently chaotic dynamical system from a single deterministically evolving time series of its states or measurements thereof, and to give explicit rates for the convergence to the solution in terms of the Jensen-Shannon divergence.

What carries the argument

Infinite-dimensional generative adversarial learning model that replaces the i.i.d. sampling assumption with the mixing or ergodic properties of a chaotic dynamical system so that one deterministic trajectory densely samples the invariant measure.

If this is right

  • Training no longer requires an ensemble of independent realizations; one sufficiently long deterministic orbit is enough.
  • Convergence occurs at explicit rates bounded in the Jensen-Shannon divergence.
  • The approach directly covers data generated by turbulent or other chaotic physical processes.
  • Generative models can succeed on measurements that are produced by deterministic evolution rather than random sampling.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the result holds, it suggests that practical GAN training on real sensor streams may succeed precisely because those streams are long enough to exploit hidden mixing.
  • The same single-trajectory argument could be tested on other generative architectures beyond the infinite-dimensional GAN model considered here.
  • Numerical checks on standard chaotic maps would give concrete numbers for the predicted convergence rates.

Load-bearing premise

The dynamical system must possess suitable mixing or ergodic properties so that a single trajectory densely samples the state space and the invariant distribution is unique and learnable.

What would settle it

Showing that the learned distribution fails to converge to the known invariant measure at the stated Jensen-Shannon rates when the method is applied to a concrete mixing system such as the logistic map at parameter 4 and trained on one long orbit would falsify the claim.

read the original abstract

Physical AI is being successfully applied to data which does not follow the traditional paradigm of independent and identically distributed (i.i.d.) samples. In fact, physical AI is often trained on data which is not random at all, and is instead derived from chaotic dynamical systems like turbulence. We aim to explain the empirical success of these methods using the example of generative adversarial networks (GANs), whose statistical learning theory under the i.i.d. assumption is generally well understood. We prove that it is possible, using an infinite-dimensional model of generative adversarial learning (GAL), to learn the invariant distribution of a sufficiently chaotic dynamical system from a single deterministically evolving time series of its states or measurements thereof, and give explicit rates for the convergence to the solution in terms of the Jensen-Shannon divergence.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript claims to prove that an infinite-dimensional model of generative adversarial learning (GAL) can recover the invariant distribution of a sufficiently chaotic dynamical system from a single deterministic time series (or measurements thereof), and supplies explicit convergence rates to this distribution measured in Jensen-Shannon divergence.

Significance. If the central derivation holds, the result would furnish a rigorous statistical-learning foundation for training GAN-style models on non-i.i.d. data generated by chaotic physical processes, thereby explaining observed empirical success in turbulence and related domains. The explicit rates and the replacement of the i.i.d. assumption by ergodic properties of a single orbit constitute the main technical contribution.

major comments (2)
  1. [§4 / main theorem] The main theorem (presumably Theorem 4.1 or the central result in §4): the stated explicit rates in Jensen-Shannon divergence are derived from the assumption that the system is 'sufficiently chaotic,' yet the manuscript supplies no quantitative mixing or correlation-decay bounds. The classical Birkhoff ergodic theorem yields only almost-sure convergence without rates; explicit finite rates require a uniform spectral gap or exponential mixing estimate that is not shown to follow from the given definition of sufficient chaos.
  2. [§3] Definition of the infinite-dimensional GAL model (likely §3): the reduction from the empirical occupation measure along a single orbit to the invariant measure is load-bearing for the rate claim, but the argument does not verify that the adversarial objective remains well-defined and the minimax gap contracts at the claimed speed when the data are deterministically generated rather than i.i.d.
minor comments (2)
  1. Notation for the function spaces and the precise statement of the Jensen-Shannon divergence in the infinite-dimensional setting could be made more explicit, perhaps with a short example of the embedding.
  2. A brief comparison table or remark contrasting the obtained rates with the classical i.i.d. GAN rates would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for identifying key points that require clarification to strengthen the rigor of our results. We address each major comment below and outline the revisions we will implement.

read point-by-point responses

  1. Referee: [§4 / main theorem] The main theorem (presumably Theorem 4.1 or the central result in §4): the stated explicit rates in Jensen-Shannon divergence are derived from the assumption that the system is 'sufficiently chaotic,' yet the manuscript supplies no quantitative mixing or correlation-decay bounds. The classical Birkhoff ergodic theorem yields only almost-sure convergence without rates; explicit finite rates require a uniform spectral gap or exponential mixing estimate that is not shown to follow from the given definition of sufficient chaos.

    Authors: We agree that explicit rates in Jensen-Shannon divergence require quantitative mixing conditions. Our definition of 'sufficiently chaotic' in Section 2 is meant to include systems admitting a spectral gap for the transfer operator, which implies exponential correlation decay. However, this implication is not derived explicitly in the current text. We will revise §4 to state the mixing assumption precisely, add a lemma deriving the JS rate from the spectral gap using standard ergodic theory bounds, and update the main theorem statement accordingly. This addresses the gap between the qualitative chaos assumption and the quantitative rates. revision: yes

  2. Referee: [§3] Definition of the infinite-dimensional GAL model (likely §3): the reduction from the empirical occupation measure along a single orbit to the invariant measure is load-bearing for the rate claim, but the argument does not verify that the adversarial objective remains well-defined and the minimax gap contracts at the claimed speed when the data are deterministically generated rather than i.i.d.

    Authors: The infinite-dimensional GAL formulation defines the objective over a function space (e.g., RKHS) that depends only on the measure class, not on i.i.d. sampling. The occupation measure converges weakly to the invariant measure under ergodicity, and we will add an explicit continuity argument in §3 showing that the JS divergence between the empirical and invariant measures controls the minimax gap via Lipschitz properties of the discriminator class. This will verify that the gap contracts at the claimed rate for deterministic orbits. The revision will include this verification step. revision: yes

Circularity Check

0 steps flagged

Derivation self-contained; no circular reductions identified

full rationale

The paper advances a mathematical proof that an infinite-dimensional generative adversarial learning model can recover the invariant measure of a chaotic dynamical system from a single deterministic orbit, with explicit Jensen-Shannon convergence rates. The central assumption of 'sufficiently chaotic' behavior (mixing or ergodicity) is introduced as an external hypothesis on the dynamical system rather than being defined in terms of the learned distribution or the GAL outputs. No load-bearing step reduces by construction to a fitted parameter, a self-citation chain, or a renaming of the target result; the argument is presented as deriving the rates from quantitative ergodic properties supplied by the chaos assumption. The derivation therefore remains independent of its own conclusion.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests primarily on the domain assumption of sufficient chaos in the dynamical system and the construction of the infinite-dimensional GAL model; these are not expanded in the abstract.

axioms (1)
  • domain assumption The dynamical system is sufficiently chaotic to possess a unique invariant distribution that can be learned from a single trajectory.
    This replaces the i.i.d. sampling assumption and enables the single time-series learning claim.

pith-pipeline@v0.9.0 · 5661 in / 1291 out tokens · 56776 ms · 2026-05-20T12:49:27.151848+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages · 1 internal anchor

  1. [1]

    Universal physics transformers: A framework for efficiently scaling neural operators

    B. Alkin, A. Fürst, S. Schmid, L. Gruber, M. Holzleitner, and J. Brandstetter. “Universal physics transformers: A framework for efficiently scaling neural operators.” In:Advances in Neural Information Processing Systems37 (2024)

    work page 2024

  2. [2]

    Wasserstein Generative Adversarial Networks

    M. Arjovsky, S. Chintala, and L. Bottou. “Wasserstein Generative Adversarial Networks.” In:Proceedings of the 34th International Conference on Machine Learning. International Conference on Machine Learning. PMLR, 2017

    work page 2017

  3. [3]

    V. I. Arnold and A. Avez.Ergodic Problems of Classical Mechanics. The Mathematical Physics Monograph Series. Benjamin, 1968

    work page 1968

  4. [4]

    A Convenient Infinite Di- mensional Framework for Generative Adversarial Learning

    H. Asatryan, H. Gottschalk, M. Lippert, and M. Rottmann. “A Convenient Infinite Di- mensional Framework for Generative Adversarial Learning.” In:Electronic Journal of Statistics17.1 (2023).issn: 1935-7524, 1935-7524. 34

    work page 2023

  5. [5]

    Simultaneous Approximation of a Smooth Function and Its Derivatives by Deep Neural Networks with Piecewise- Polynomial Activations

    D. Belomestny, A. Naumov, N. Puchkin, and S. Samsonov. “Simultaneous Approximation of a Smooth Function and Its Derivatives by Deep Neural Networks with Piecewise- Polynomial Activations.” In:Neural Networks161 (2023).issn: 0893-6080

    work page 2023

  6. [6]

    Some Theoretical Properties of GANS

    G. Biau, B. Cadre, M. Sangnier, and U. Tanielian. “Some Theoretical Properties of GANS.” In:The Annals of Statistics48.3 (2020).issn: 0090-5364, 2168-8966

    work page 2020

  7. [7]

    M. Bode, M. Gauding, D. Goeb, T. Falkenstein, and H. Pitsch. “Applying physics- informed enhanced super-resolution generative adversarial networks to turbulent pre- mixed combustion and engine-like flame kernel direct numerical simulation data.” In: Proceedings of the Combustion Institute39.4 (2023)

    work page 2023

  8. [8]

    Using physics-informed enhanced super-resolution generative adversarial net- worksforsubfiltermodelinginturbulentreactiveflows

    M. Bode, M. Gauding, Z. Lian, D. Denker, M. Davidovic, K. Kleinheinz, J. Jitsev, and H. Pitsch. “Using physics-informed enhanced super-resolution generative adversarial net- worksforsubfiltermodelinginturbulentreactiveflows.” In:Proceedings of the Combustion Institute38.2 (2021)

    work page 2021

  9. [9]

    R. E. Bowen, J.-R. Chazottes, and D. Ruelle.Equilibrium States and the Ergodic Theory of Anosov Diffeomorphisms. Vol. 470. Lecture Notes in Mathematics. Berlin, Heidelberg: Springer, 2008.isbn: 978-3-540-77605-5

    work page 2008

  10. [10]

    Optimal Concentration Inequalities for Dynamical Sys- tems

    J.-R. Chazottes and S. Gouëzel. “Optimal Concentration Inequalities for Dynamical Sys- tems.” In:Communications in Mathematical Physics316.3 (2012).issn: 1432-0916

    work page 2012

  11. [11]

    Exponential Inequalities for Dynamical Measures of Expanding Maps of the Interval

    P. Collet, S. Martinez, and B. Schmitt. “Exponential Inequalities for Dynamical Measures of Expanding Maps of the Interval.” In:Probability Theory and Related Fields123.3 (2002).issn: 1432-2064

    work page 2002

  12. [12]

    Deviation and Concentration Inequalities for Dynamical Systems with Subexponential Decay of Correlations

    C. Cuny, J. Dedecker, and F. Merlevède. “Deviation and Concentration Inequalities for Dynamical Systems with Subexponential Decay of Correlations.” In:Stochastics and Dy- namics23.03 (2023).issn: 0219-4937

    work page 2023

  13. [13]

    Subgaussian Concentration Inequalities for Geometrically Ergodic Markov Chains

    J. Dedecker and S. Gouëzel. “Subgaussian Concentration Inequalities for Geometrically Ergodic Markov Chains.” In:Electronic Communications in Probability20 (2015).issn: 1083-589X, 1083-589X

    work page 2015

  14. [14]

    Learning Transient Convective Heat Transfer with Geometry Aware World Models

    O. T. Doganay, A. Klawonn, M. Eigel, and H. Gottschalk. “Learning Transient Convective Heat Transfer with Geometry Aware World Models.” In:arXiv preprint arXiv:2601.22086 (2026)

    work page arXiv 2026

  15. [15]

    R. Douc, E. Moulines, P. Priouret, and P. Soulier.Markov Chains. Springer Series in Op- erations Research and Financial Engineering. Cham: Springer International Publishing, 2018.isbn: 978-3-319-97703-4

    work page 2018

  16. [16]

    Generative Modeling of Turbu- lence

    C. Drygala, B. Winhart, F. di Mare, and H. Gottschalk. “Generative Modeling of Turbu- lence.” In:Physics of Fluids34.3 (2022).issn: 1070-6631

    work page 2022

  17. [17]

    E. C. Ehrhardt, H. Gottschalk, and T. J. Riedlinger.Numerical and Statistical Analysis of NeuralODE with Runge-Kutta Time Integration. 2025.url:http://arxiv.org/abs/ 2503.10729(visited on 06/10/2025). Pre-published

    work page arXiv 2025

  18. [18]

    Eisner, B

    T. Eisner, B. Farkas, M. Haase, and R. Nagel.Operator Theoretic Aspects of Ergodic The- ory. Vol. 272. Graduate Texts in Mathematics. Cham: Springer International Publishing, 2015.isbn: 978-3-319-16897-5

    work page 2015

  19. [19]

    Super-resolution reconstruction of turbulent flows with machine learning

    K. Fukami, K. Fukagata, and K. Taira. “Super-resolution reconstruction of turbulent flows with machine learning.” In:Journal of Fluid Mechanics870 (2019).issn: 1469-7645. 35

    work page 2019

  20. [20]

    OntheDependenceoftheConvergenceRateintheStrongLawofLarge Numbers for Stationary Processes on the Rate of Decay of the Correlation Function

    V.F.Gaposhkin.“OntheDependenceoftheConvergenceRateintheStrongLawofLarge Numbers for Stationary Processes on the Rate of Decay of the Correlation Function.” In: Theory of Probability & Its Applications26.4 (1982).issn: 0040-585X

    work page 1982

  21. [21]

    Gilbarg and N

    D. Gilbarg and N. S. Trudinger.Elliptic Partial Differential Equations of Second Order. Vol. 224. Classics in Mathematics. Berlin, Heidelberg: Springer, 2001.isbn: 978-3-540- 41160-4

    work page 2001

  22. [22]

    Generative Adversarial Nets

    I.Goodfellow,J.Pouget-Abadie,M.Mirza,B.Xu,D.Warde-Farley,S.Ozair,A.Courville, and Y. Bengio. “Generative Adversarial Nets.” In:Advances in Neural Information Pro- cessing Systems. Vol. 27. Curran Associates, Inc., 2014

    work page 2014

  23. [23]

    A Quantitative McDiarmid’s In- equality for Geometrically Ergodic Markov Chains

    A. Havet, M. Lerasle, E. Moulines, and E. Vernet. “A Quantitative McDiarmid’s In- equality for Geometrically Ergodic Markov Chains.” In:Electronic Communications in Probability25 (2020).issn: 1083-589X, 1083-589X

    work page 2020

  24. [24]

    A Style-Based Generator Architecture for Generative Adversarial Networks

    T. Karras, S. Laine, and T. Aila. “A Style-Based Generator Architecture for Generative Adversarial Networks.” In:IEEE Transactions on Pattern Analysis and Machine Intelli- gence43.12 (2021).issn: 1939-3539

    work page 2021

  25. [25]

    Unsupervised deep learning for super-resolution reconstruction of turbulence

    H. Kim, J. Kim, S. Won, and C. Lee. “Unsupervised deep learning for super-resolution reconstruction of turbulence.” In:Journal of Fluid Mechanics910 (2021)

    work page 2021

  26. [26]

    Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers

    J. Kim and C. Lee. “Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers.” In:Journal of Computational Physics406 (2020).issn: 0021- 9991

    work page 2020

  27. [27]

    Creating Turbulent Flow Realizations with Genera- tive Adversarial Networks

    R. King, P. Graf, and M. Chertkov. “Creating Turbulent Flow Realizations with Genera- tive Adversarial Networks.” In:APS Division of Fluid Dynamics Meeting Abstracts. APS Meeting Abstracts. Nov. 2017, A31.008

    work page 2017

  28. [28]

    SPATE-GAN: Improved Genera- tive Modeling of Dynamic Spatio-Temporal Patterns with an Autoregressive Embedding Loss

    K. Klemmer, T. Xu, B. Acciaio, and D. B. Neill. “SPATE-GAN: Improved Genera- tive Modeling of Dynamic Spatio-Temporal Patterns with an Autoregressive Embedding Loss.” In:Proc. Thirty-Sixth AAAI Conference on Artificial Intelligence. 2022

    work page 2022

  29. [29]

    J. M. Lee.Introduction to Riemannian Manifolds. Vol. 176. Graduate Texts in Mathe- matics. Cham: Springer International Publishing, 2018.isbn: 978-3-319-91754-2

    work page 2018

  30. [30]

    Fourier neural operator for parametric partial differential equations

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, and A. Anandkumar. “Fourier neural operator for parametric partial differential equations.” In: International Conference on Learning Representations (ICLR). 2021

    work page 2021

  31. [31]

    How Well Generative Adversarial Networks Learn Distributions

    T. Liang. “How Well Generative Adversarial Networks Learn Distributions.” In:Journal of Machine Learning Research22.228 (2021).issn: 1533-7928

    work page 2021

  32. [32]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators

    L. Lu, P. Jin, G. Pang, Z. Zhang, and G. E. Karniadakis. “Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators.” In:Nature Machine Intelligence3.3 (2021)

    work page 2021

  33. [33]

    Young Towers for Product Systems

    S. Luzzatto and M. Ruziboev. “Young Towers for Product Systems.” In:Discrete and Continuous Dynamical Systems36.3 (2015).issn: 1078-0947

    work page 2015

  34. [34]

    DistributionLearningviaNeuralDifferential Equations: A Nonparametric Statistical Perspective

    Y.Marzouk,Z.(Ren,S.Wang,andJ.Zech.“DistributionLearningviaNeuralDifferential Equations: A Nonparametric Statistical Perspective.” In:Journal of Machine Learning Research25.232 (2024).issn: 1533-7928. 36

    work page 2024

  35. [35]

    On the Method of Bounded Differences

    C. McDiarmid. “On the Method of Bounded Differences.” In:Surveys in Combinatorics, 1989: Invited Papers at the Twelfth British Combinatorial Conference. Ed. by J. Siemons. London Mathematical Society Lecture Note Series. Cambridge: Cambridge University Press, 1989.isbn: 978-0-521-37823-9

    work page 1989

  36. [36]

    Conditional Generative Adversarial Nets

    M. Mirza and S. Osindero.Conditional Generative Adversarial Nets. 2014.url:http: //arxiv.org/abs/1411.1784(visited on 03/18/2025). Pre-published

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  37. [37]

    Learning tur- bulent flows with generative models for super resolution and sparse flow reconstruction

    V. Oommen, S. Khodakarami, A. Bora, Z. Wang, and G. E. Karniadakis. “Learning tur- bulent flows with generative models for super resolution and sparse flow reconstruction.” In:Nature Communications(2026)

    work page 2026

  38. [38]

    Rates of Con- vergence for Density Estimation with Generative Adversarial Networks

    N. Puchkin, S. Samsonov, D. Belomestny, E. Moulines, and A. Naumov. “Rates of Con- vergence for Density Estimation with Generative Adversarial Networks.” In:Journal of Machine Learning Research25.29 (2024).issn: 1533-7928

    work page 2024

  39. [39]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

    M. Raissi, P. Perdikaris, and G. E. Karniadakis. “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.” In:Journal of Computational physics378 (2019)

    work page 2019

  40. [40]

    When do World Models Successfully Learn Dynamical Systems?

    E. Ross, C. Drygala, L. Schwarz, S. Kaiser, F. di Mare, T. Breiten, and H. Gottschalk. “When do World Models Successfully Learn Dynamical Systems?” In:arXiv preprint arXiv:2507.04898(2025)

    work page arXiv 2025

  41. [41]

    Shalev-Shwartz and S

    S. Shalev-Shwartz and S. Ben-David.Understanding machine learning: From theory to algorithms. Cambridge university press, 2014

    work page 2014

  42. [42]

    Stylegan-v: A continuous video generator with the price, image quality and perks of stylegan2

    I. Skorokhodov, S. Tulyakov, and M. Elhoseiny. “Stylegan-v: A continuous video generator with the price, image quality and perks of stylegan2.” In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022

    work page 2022

  43. [43]

    A. W. Van Der Vaart and J. A. Wellner.Weak Convergence and Empirical Processes: With Applications to Statistics. Springer Series in Statistics. Cham: Springer International Publishing, 2023.isbn: 978-3-031-29038-1

    work page 2023

  44. [44]

    Van Handel.Probability in High Dimension:Fort Belvoir, VA: Defense Technical In- formation Center, 2014

    R. Van Handel.Probability in High Dimension:Fort Belvoir, VA: Defense Technical In- formation Center, 2014

    work page 2014

  45. [45]

    Vershynin.High-Dimensional Probability: An Introduction with Applications in Data Science

    R. Vershynin.High-Dimensional Probability: An Introduction with Applications in Data Science. Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge: Cam- bridge University Press, 2018.isbn: 978-1-108-41519-4

    work page 2018

  46. [46]

    Esr- gan: Enhanced super-resolution generative adversarial networks

    X. Wang, K. Yu, S. Wu, J. Gu, Y. Liu, C. Dong, Y. Qiao, and C. Change Loy. “Esr- gan: Enhanced super-resolution generative adversarial networks.” In:Proceedings of the European conference on computer vision (ECCV) workshops. 2018

    work page 2018

  47. [47]

    Recurrence Times and Rates of Mixing

    L.-S. Young. “Recurrence Times and Rates of Mixing.” In:Israel Journal of Mathematics 110.1 (1999).issn: 1565-8511

    work page 1999

  48. [48]

    Statistical Properties of Dynamical Systems with Some Hyperbolicity

    L.-S. Young. “Statistical Properties of Dynamical Systems with Some Hyperbolicity.” In: Annals of Mathematics147.3 (1998).issn: 0003-486X

    work page 1998

  49. [49]

    Zehnder.Lectures on Dynamical Systems: Hamiltonian Vector Fields and Symplectic Capacities

    E. Zehnder.Lectures on Dynamical Systems: Hamiltonian Vector Fields and Symplectic Capacities. European Mathematical Society, 2010.isbn: 978-3-03719-081-4. 37

    work page 2010