Pith. sign in

REVIEW 3 major objections 5 minor 51 references

Wasserstein Geometry of Information Loss in Nonlinear Dynamical Systems

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A data-driven score, intrinsic stochasticity, certifies when time-delay reconstructions fail to be deterministic closures.

desk verdict Useful finite-resolution closure diagnostic; the advertised scaling law overreaches because it depends on an unproved finite-sheet decomposition. read the letter →

arxiv 2601.22814 v2 pith:2ZZHHCAY submitted 2026-01-30 stat.CO

Yiting Duan , Zhikun Zhang , Yi Guo This is my paper · ORCID
classification stat.CO MSC 37M1037D2549Q22
keywords time-delayembeddingTakens'theoreminformationlossintrinsicstochasticitydeterministicclosureWassersteindistanceSRBmeasurenon-injectivereconstruction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Time-delay embedding is only as good as the map that builds it: when the map is not injective, distinct latent states collapse to the same reconstructed state, and the observed future is a distribution rather than a point. The paper claims that this ambiguity is not an artefact of noise or parameter choice but an intrinsic property of the system, and it introduces a single number — intrinsic stochasticity E*_n — that measures it as the minimal expected distance from the n-step conditional future kernel to its best deterministic prediction. Lower E*_n means the reconstruction is closer to deterministic closure, and in numerical experiments the reconstructions with lower E*_n are exactly the ones whose EDMD and DIM rollout predictions are more accurate. The authors further derive a geometric mechanism: information loss scales as the product of the probability mass on outlier branches and the branch separation, which is set by a competition between dynamical stretching and observation curvature. A sympathetic reader would care because the score is computed from data alone and could serve as a practical certificate for selecting measurement functions and embeddings without governing equations.

What carries the argument

The central object is the intrinsic stochasticity E*_n (Definition 3.1), defined for each horizon n as the integral over reconstructed states x of the optimal-transport risk m_n(x) = inf_y ℓ_n(x,y), where ℓ_n is the Wasserstein-1 cost between y and the conditional future kernel K_n(x,·) = (F∘T^n)_#ν_x obtained by disintegrating the invariant measure along the fibres of F. The infimum is attained at the geometric median Ψ*_n(x), selected over the conditional mean because the mean may land in a low-density gap between branches. The supporting machinery is the Pesin-block branch model: under non-uniform hyperbolicity and SRB product structure, ν_x decomposes into finitely many atoms (Eq. 13), b

What would settle it

Take a smooth non-injective reconstruction with a known continuous fibre — for instance a two-dimensional projection of a three-dimensional flow with an interval preimage carrying positive SRB mass — estimate the pointwise cost with the k-nearest-neighbour cloud, and check whether the tightness ratio QLB concentrates at an O(1) constant as in the Rössler validation. If the cost is spread continuously across the fibre rather than concentrated on finitely many separated branches, the scaling law E*_n ~ b_n Δ_n fails. Alternatively, search two reconstructions with E*_n lower but rollout error hig

Watch

Extended reading notes

Core claim

The paper's central claim is that non-injectivity of the time-delay map produces an intrinsically stochastic closed evolution on the reconstructed space, and that this stochasticity is quantifiable by E*_n = ∫_X inf_y ℓ_n(x,y) dµ(x), where ℓ_n(x,y) is the expected distance under the conditional kernel K_n(x,·) to y. When E*_n = 0 the kernel is a Dirac mass almost everywhere and the reconstruction is a deterministic closure; when it is positive, the best deterministic predictor is the geometric median Ψ*_n(x), and the residual cost is irreducible information loss. On Pesin blocks the paper argues that the conditional measure decomposes into finitely many branches, so the loss scales as the pr

Load-bearing premise

The scaling law rests on the assumption that the conditional measure along the preimage fibre decomposes into finitely many discrete branches with a controlled minimum separation on Pesin blocks; if the fibres are continuous or the branch count is unbounded, the product formula for information loss is not established.

Editorial extensions

If this is right

  • E*_n supplies a data-only certificate of deterministic closure: a vanishing score means the n-step future is uniquely determined by the reconstruction, generalising Takens' injectivity condition to a continuous metric.
  • Reconstructions with lower estimated E*_n should be preferred as inputs to downstream models; the paper demonstrates this ordering for EDMD and DIM rollout errors on Rössler data.
  • The score is sensitive to finite resolution even when a map is analytically invertible, so it ranks embeddings by practical predictability rather than mathematical injectivity.
  • Signal transformations can be audited: log-transforming the intermittent NYC measles series lowers E*_n, matching the known difficulty of predicting the raw series, and the framework lets one choose the better measurement.
  • The stretching–curvature scaling law identifies where in state space multi-valued evolution concentrates, so it can locate the SRB-relevant folding regions that actually degrade reconstruction.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the score is as reliable as the experiments suggest, it could be used offline as an automatic sensor-placement and embedding-parameter criterion in control and network science — a use the authors mention as future work.
  • The Wasserstein framing suggests a natural extension to stochastic or noisy systems: one could separate intrinsic ambiguity from observational noise by comparing E*_n against a noise baseline, which the paper does not do.
  • A testable extension is to use E*_n as a loss or regulariser when training delay autoencoders or measurement maps, directly optimising the reconstruction for closure rather than only scoring fixed embeddings.
  • The apparent paradox that diffeomorphic embeddings like (z2,z3,˙z2) score poorly suggests E*_n measures finite-horizon, finite-resolution closure rather than global invertibility; whether that is a feature or a limitation depends on the downstream goal.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper defines intrinsic stochasticity E*_n as the µ-average of the minimal Wasserstein-1 distance from the n-step conditional kernel K_n(x,·) to a point mass (the geometric median), and proposes a k-NN/Weiszfeld estimator for it. The authors argue that E*_n is an almost-everywhere, data-driven certificate of deterministic closure for time-delay reconstructions, and that information loss from non-injectivity is governed by a geometric competition between hyperbolic stretching and observation curvature, summarized by the scaling relation E*_n ~ ∫ b_n(x) Δ_n(x) dµ(x) (Eq. 23). The paper reports numerical experiments on Rössler, a double pendulum, measles incidence data, and downstream EDMD/DIM rollout tasks, with the code publicly available.

Significance. If the central claims held, the paper would provide a principled, observable-based diagnostic for time-delay reconstruction quality, with clear practical value for measurement selection. The estimator is simple and reproducible, and the downstream correlations with EDMD/DIM performance are plausible and worth reporting. However, the theoretical mechanism is currently an unproven structural ansatz, and the numerical validation is performed largely in the regime where the ansatz is assumed. The central claim is therefore not yet established at the level of a rigorous certificate, although the estimator may still function as a heuristic diagnostic. I regard the paper as a promising but not yet fully supported contribution.

major comments (3)
  1. [§3.2.1, Eq. (13)] The finite-sheet decomposition ν_x = Σ_{i=1}^N p_i(x) δ_{z_i(x)} is assumed, not proved. The preceding text only states that conditional SRB measures along unstable leaves are absolutely continuous; a smooth non-injective map can have continua or fractal fibres, and compactness plus local product structure does not imply atomicity. The branch weights in Eq. (17), the separation bound in Eq. (21), and the scaling law in Eq. (23) all depend on Eq. (13). If this is a phenomenological ansatz, its scope must be stated explicitly; if it is a theorem, a proof or precise citation is required. As written, the advertised 'physical law' is unsupported for the Rössler, double-pendulum, and measles applications.
  2. [§3.2.3, Fig. 5] The validation of the lower bound is restricted to 110 of 400 query points, selected for bimodality and strong stretching. Those are exactly the points where the finite-sheet branch model is assumed to hold, so the test is self-consistent rather than a test of Eq. (13). Moreover, Eq. (23) is a definitional consequence of the branch model: once the kernel is a finite sum of Dirac masses, E*_n is the averaged median cost and is bounded by the dominant-branch risk. Without an independent specification of b_n(x) and Δ_n(x), the scaling relation is not falsifiable. Please report the behavior on all 400 points and provide a criterion for detecting where the branch model fails.
  3. [§3.4, Table 1] The central claim is that lower E*_n identifies reconstructions closer to deterministic closure. Table 1 is in tension with this claim: the diffeomorphic embeddings (z2,z3, ˙z2) and (z1,z3, ˙z1) receive E*_n ≈ 4.7, while non-diffeomorphic embeddings such as (z1, ˙z1, ˙z2) receive E*_n ≈ 0.05. Under the definition in Eq. (11), a diffeomorphism gives ν_x = δ, hence K_n(x,·) = δ and E*_n = 0. The nonzero values therefore come from the finite-k estimator, not from the theoretical certificate. The radius diagnostics in Eq. (31) and Table 2 confirm that the estimator conflates finite-neighborhood sampling, chaotic divergence over the horizon, and non-injectivity. This is a load-bearing issue for the paper's advertised interpretation; please clarify what exactly is being certified and provide a decomposition separating sampling resolution from intrinsic loss.
minor comments (5)
  1. [§3.1, Definition 3.1] The loss ℓ_n(x,y) is used in Eq. (10) but its explicit form appears only later in Eq. (24). Please define ℓ_n(x,y) = ∫ dX(y,y') K_n(x,dy') at the definition.
  2. [§3.3, Eq. (25)] Consistency of the uniform KNN empirical kernel is cited to Stone (1977), which is a regression paper. A more precise statement for conditional distribution estimation under mixing/ergodic assumptions is needed.
  3. [§3.5, measles paragraph] The text says 'after the log transformation, the value of E* consistently increases', but the surrounding discussion and Figure 6 indicate that the log transform improves reconstruction, so presumably E* decreases. Please correct.
  4. [§3.2.3, Eq. (24)] The upper bound U_n(x) ≥ m_n(x) is tautological by definition of m_n as the infimum. Calling it a validated upper bound is only meaningful because Q_UB is reported; please state explicitly that the nontrivial part is the tightness, not the inequality itself.
  5. [Algorithm 1] There is no guidance on choosing the Theiler window w, neighborhood size k, or query count N_Q, although results appear sensitive to these, especially k and w. A sensitivity analysis or practical recommendation would strengthen the paper.

Circularity Check

2 steps flagged · score 6.0 of 10

The advertised scaling law is largely a restatement of the median-cost definition once the atomic branch ansatz is imposed, and its numerical lower-bound validation calibrates the mass constant on the same selected subset, making the 'prediction' partly self-consistent by construction.

  1. self definitional [§3.2.3, Eqs. (13)-(14) and Eq. (23)]
    "Pushing the measure forward by n steps, the transition kernel splits into discrete branches: K(n)(x,·) = Σ_{i=1}^N p_i(x)δ_{y_i(n)(x)} ... Combining the mass decomposition (Eq. 17) and the separation geometry (Eq. 21), we derive a unified physical law ... E∗_n ∼ ∫_A b_n(x)·∆_n(x)dµ(x)."

    Under the assumed decomposition (13)-(14), K_n(x,·) is a finite atomic measure, so the pointwise Fréchet-median cost in Definition 3.1 is, by definition, a function of exactly the same branch probabilities p_i(x) and inter-branch distances that appear on the right-hand side of Eq. (23). In the two-branch case the identity is exact: m_n(x) = (1-p_i0(x))Δ_n(x). Thus Eq. (23) is not a new scaling law derived from the dynamics; it is the definition of E*_n rewritten in branch coordinates. The only genuinely dynamical content is the lower bound (21), which is separate. Consequently the advertised claim that information loss 'scales with the product of geometric separation and probability mass' is true by construction once Eq. (13) is assumed, and unproved if that finite-sheet ansatz fails.

  2. fitted input called prediction [§3.2.3, lower-bound validation paragraph]
    "Using the 95% quantile of the branch-weight ratio r = wmax/w2 gives r95∗ = 4.689 and hence a folding-mass lower bound b0 = (1+r95∗)−1 = 0.176. On this subset we observe P( ˆm(w)_n ≥ b0∆) = 1.00 with correlation ρcorr = 0.998..."

    The constant b0 is calibrated from the empirical branch-weight ratios r = wmax/w2 on the same 110-point subset that is later used to test the inequality. In the two-branch atomic model the Fréchet cost is m_n ≈ bΔ with b = 1/(1+r), so choosing r as the 95th quantile makes m_n ≥ b0Δ hold for about 95% of the selected points by construction; the reported 100% is a self-consistency check rather than an out-of-sample validation. The subset is further restricted to pushed-forward clouds that are 'genuinely bimodal' with positive silhouette scores and strong stretching, i.e., exactly the regime where the atomic branch ansatz (13) is assumed. Thus the numerical support for Eq. (23) does not independently confirm the scaling law outside its defining assumptions.

full rationale

The main diagnostic E*_n itself — the Wasserstein-1 distance from the conditional future kernel to the closest Dirac — is a well-defined data-driven quantity, and the downstream correlations with EDMD and DIM rollout errors (Figs. 7-8) are genuinely external: they compare E*_n(z1) vs E*_n(z3) against independently measured prediction error. Those results do not depend on the scaling law, so the paper has real empirical content. However, the central theoretical claim advertised in the abstract — that irreducible information loss 'scales with the product of the geometric separation and the probability mass' — is Eq. (23), and that relation is essentially the definition of the Fréchet-median cost once the finite-branch decomposition (13)-(14) is imposed. The numerical lower-bound verification in the same section is weakened because b0 is fitted from branch weights on the same selected bimodal subset, making the observed P=1.00 and high correlation partly forced. No load-bearing self-citation or imported uniqueness theorem was found; references to prior work by the authors are not used to justify the central mechanism. Overall: partial circularity in the scaling-law derivation and its validation, while the core diagnostic retains independent downstream support. Score 6.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central derivation is built on unproven structural assumptions about SRB measures and Pesin blocks, plus user-chosen computational parameters (k, n, Theiler window). The scaling-law constant b0 is fitted to a selected subset during validation. No new physical entities are introduced; E*_n is a data-defined statistic.

free parameters (4)
  • k (KNN neighborhood count) = 50
    E*_n,k depends on the finite neighborhood scale; the paper acknowledges this by subscripting with k and shows high scores for diffeomorphic embeddings arise from large-radius neighborhoods. The choice k=50 is user-set, not data-driven or derived.
  • Theiler exclusion window w = not specified
    Algorithm 1 requires a Theiler window; its size is not reported, so the empirical kernel and E*_n,k depend on an unreported choice.
  • push-forward horizon n (p in experiments) = 20 for synthetic and DIM; varied in double-pendulum and measles
    E*_n is horizon-specific; experiments choose n=20 for Rössler and DIM, and grid n for other datasets. The diagnostic's ranking can depend on the horizon.
  • b0 = (1 + r95*)^(-1) = 0.176 (from r95* = 4.689)
    In §3.2.3, the 95% quantile of the branch-weight ratio r = w_max/w_2 is estimated on the selected subset and used to define the lower-bound constant b0. This is a data-fitted constant in the validation of the scaling law.
assumptions (5)
  • standard math Takens' theorem and generic embedding; F is a smooth map from a compact C^r manifold to R^m
    Used in §2.1 to justify that embeddings are generic; the paper then argues fixed measurement functions can be exceptional.
  • domain assumption The attractor carries an ergodic SRB physical invariant measure ν, and Birkhoff averages converge along typical trajectories
    Invoked in §2.1 and §3.2 to justify the measure-theoretic certificate and empirical sampling.
  • ad hoc to paper Hyperbolic visibility: non-uniform hyperbolic expansion with rate λ_u and C^2 bounded observation on Pesin blocks with local product structure
    Introduced in §3.2.2 to derive the separation lower bound Eq. (21); not established for the Rössler, double-pendulum, or measles systems.
  • ad hoc to paper The conditional fibre measure ν_x decomposes into a finite sum of Dirac masses supported on distinct sheets (Eq. 13)
    Central structural ansatz for the branch model and the scaling law; not proven for fractal attractors or non-injective maps with continuous fibres.
  • standard math KNN empirical measures converge weakly to K_n(x,·) under mixing/ergodic regularity as k→∞, k/N→0
    Used in §3.3 (citing Stone 1977) to justify the estimator; the finite-resolution bias is acknowledged and later shown to dominate scores.
invented entities (1)
  • Intrinsic stochasticity E*_n
    purpose: Scalar certificate of deterministic closure used as a diagnostic for time-delay reconstruction quality.
    It is a statistic defined from the data itself; the paper provides no falsifiable outside prediction that would confirm it independently of the same KNN clouds used to compute it. The only external support is the downstream EDMD/DIM correlation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Wasserstein Geometry of Information Loss in Nonlinear Dynamical Systems." pith.science (2026). https://pith.science/paper/2ZZHHCAY

@misc{pith2026260122814,
  author       = {Pith},
  title        = {Pith review of: Wasserstein Geometry of Information Loss in Nonlinear Dynamical Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2ZZHHCAY}},
  note         = {Machine review of arXiv:2601.22814}
}
abstract

Time-delay embedding is a powerful technique for reconstructing the dynamics of nonlinear systems. However, the reconstruction map is not always an embedding, a condition rarely verified in practice. When the reconstruction map is non-injective, multiple latent states may map to the same reconstructed state, leading to multi-valued $n$-step evolution. Consequently, the induced system no longer admits a deterministic closure, and the dispersion of future trajectories leads to ambiguity. In this work, we establish a measure-theoretic framework to quantify the ambiguity induced by multi-valued evolution and introduce intrinsic stochasticity to quantify the ambiguity over a finite horizon. For numerical implementation, we use the $k$-nearest-neighbor estimator to approximate intrinsic stochasticity under finite-resolution and finite-sampling settings. Numerical experiments on the synthetic and real-world datasets are consistent with the expectation: reconstructions closer to deterministic closure tend to produce lower scores, and deterministic predictors that take reconstructions with lower empirical closure scores as input are associated with lower rollout errors, suggesting that intrinsic stochasticity provides a new perspective for understanding failures of reconstruction and serves as a diagnostic for selecting reconstruction maps.

Figures

Figures reproduced from arXiv: 2601.22814 by the authors.

Figure 1
Figure 1. Performance of CCM on the Rössler system with dynamically-coupling state [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of our measure-theoretic pipeline. (b) Illustrate the case when the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (a) The stretch-and-fold phenomenon as increasing the lag value [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: A numerical validation of the dynamical separation and observation curvature [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Numerical Verification of the lower and upper bound via Rössler system with [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Application of Intrinsic Stochasticity for two real-world systems: a measles [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Pipelines for the multi-horizontal rollout prediction using EDMD. We select [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: figure 8 [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 8
Figure 8. Figure 8: Pipeline for the DIM method. This approach aims to learn the operator/vector [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 3 linked inside Pith

  1. [1]

    Investigating observability properties from data in nonlinear dynamics

    Luis A Aguirre and Christophe Letellier. Investigating observability properties from data in nonlinear dynamics. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 83(6):066209, 2011

    2011

  2. [2]

    Observability of multivariate differ- ential embeddings.Journal of Physics A: Mathematical and General, 38(28):6311, 2005

    Luis Antonio Aguirre and Christophe Letellier. Observability of multivariate differ- ential embeddings.Journal of Physics A: Mathematical and General, 38(28):6311, 2005

    2005

  3. [3]

    Springer, 2005

    Luigi Ambrosio, Nicola Gigli, and Giuseppe Savaré.Gradient flows: in metric spaces and in the space of probability measures. Springer, 2005

    2005

  4. [4]

    Discov- ering governing equations from partial measurements with deep delay autoencoders

    Joseph Bakarji, Kathleen Champion, J Nathan Kutz, and Steven L Brunton. Discov- ering governing equations from partial measurements with deep delay autoencoders. Proceedings of the Royal Society A, 479(2276):20230422, 2023

    2023

  5. [5]

    Invariant measures in time-delay coordinates for unique dynamical system identification.Physical Review Letters, 135(16):167202, 2025

    Jonah Botvinick-Greenhouse, Robert Martin, and Yunan Yang. Invariant measures in time-delay coordinates for unique dynamical system identification.Physical Review Letters, 135(16):167202, 2025

    2025

  6. [6]

    Measure- theoretic time-delay embedding.Journal of Statistical Physics, 192(12):171, 2025

    Jonah Botvinick-Greenhouse, Maria Oprea, Romit Maulik, and Yunan Yang. Measure- theoretic time-delay embedding.Journal of Statistical Physics, 192(12):171, 2025

    2025

  7. [7]

    Does observability affect proso- ciality? Proceedings of the Royal Society B: Biological Sciences, 285(1875):20180116, 2018

    Alex Bradley, Claire Lawrence, and Eamonn Ferguson. Does observability affect proso- ciality? Proceedings of the Royal Society B: Biological Sciences, 285(1875):20180116, 2018

    2018

  8. [8]

    Extracting qualitative dynamics from experimental data

    David S Broomhead and Gregory P King. Extracting qualitative dynamics from experimental data. Physica D: Nonlinear Phenomena, 20(2-3):217–236, 1986

    1986

Show all 51 references
  1. [9]

    Chaos as an intermittently forced linear system.Nature communications, 8(1):19, 2017

    Steven L Brunton, Bingni W Brunton, Joshua L Proctor, Eurika Kaiser, and J Nathan Kutz. Chaos as an intermittently forced linear system.Nature communications, 8(1):19, 2017

    2017

  2. [10]

    Modern koopman theory for dynamical systems.arXiv preprint arXiv:2102.12086, 2021

    Steven L Brunton, Marko Budišić, Eurika Kaiser, and J Nathan Kutz. Modern koopman theory for dynamical systems.arXiv preprint arXiv:2102.12086, 2021

    2021 arXiv

  3. [11]

    Discovering governing equa- tions from data by sparse identification of nonlinear dynamical systems.Proceedings of the national academy of sciences, 113(15):3932–3937, 2016

    Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Discovering governing equa- tions from data by sparse identification of nonlinear dynamical systems.Proceedings of the national academy of sciences, 113(15):3932–3937, 2016

    2016

  4. [12]

    Data- driven discovery of coordinates and governing equations.Proceedings of the National Academy of Sciences, 116(45):22445–22451, 2019

    Kathleen Champion, Bethany Lusch, J Nathan Kutz, and Steven L Brunton. Data- driven discovery of coordinates and governing equations.Proceedings of the National Academy of Sciences, 116(45):22445–22451, 2019

    2019

  5. [13]

    Differential embedding of the lorenz attractor.Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 81(6):066220, 2010

    Daniel J Cross and R Gilmore. Differential embedding of the lorenz attractor.Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 81(6):066220, 2010

    2010

  6. [14]

    Delay-coordinate maps and the spectra of koopman operators.Journal of Statistical Physics, 175(6):1107–1145, 2019

    Suddhasattwa Das and Dimitrios Giannakis. Delay-coordinate maps and the spectra of koopman operators.Journal of Statistical Physics, 175(6):1107–1145, 2019. 24

    2019

  7. [15]

    Causal discovery in symmetric dynamic systems with convergent cross mapping.arXiv preprint arXiv:2505.04815, 2025

    Yiting Duan, Yi Guo, Jack Yang, and Ming Yin. Causal discovery in symmetric dynamic systems with convergent cross mapping.arXiv preprint arXiv:2505.04815, 2025

    2025 arXiv

  8. [16]

    Ergodic theory of chaos and strange attractors

    J-P Eckmann and David Ruelle. Ergodic theory of chaos and strange attractors. Reviews of modern physics, 57(3):617, 1985

    1985

  9. [17]

    Causal inference from cross-sectional earth system data with geographical convergent cross mapping.nature communications, 14(1):5875, 2023

    Bingbo Gao, Jianyu Yang, Ziyue Chen, George Sugihara, Manchun Li, Alfred Stein, Mei-Po Kwan, and Jinfeng Wang. Causal inference from cross-sectional earth system data with geographical convergent cross mapping.nature communications, 14(1):5875, 2023

    2023

  10. [18]

    Delay-coordinate maps, coherence, and approximate spectra of evolution operators

    Dimitrios Giannakis. Delay-coordinate maps, coherence, and approximate spectra of evolution operators. Research in the Mathematical Sciences, 8(1):8, 2021

    2021

  11. [19]

    Assessing observability of chaotic systems using delay differential analysis

    Christopher E Gonzalez, Claudia Lainscsek, Terrence J Sejnowski, and Christophe Letellier. Assessing observability of chaotic systems using delay differential analysis. Chaos: An Interdisciplinary Journal of Nonlinear Science, 30(10), 2020

    2020

  12. [20]

    Princeton university press, 2020

    James D Hamilton.Time series analysis. Princeton university press, 2020

    2020

  13. [21]

    Nonlinear controllability and observability

    Robert Hermann and Arthur Krener. Nonlinear controllability and observability. IEEE Transactions on automatic control, 22(5):728–740, 2003

    2003

  14. [22]

    Structured time-delay models for dynamical systems with connections to frenet–serret frame

    Seth M Hirsh, Sara M Ichinaga, Steven L Brunton, J Nathan Kutz, and Bingni W Brunton. Structured time-delay models for dynamical systems with connections to frenet–serret frame. Proceedings of the Royal Society A, 477(2254):20210097, 2021

    2021

  15. [23]

    Learning discrepancy models from experimental data

    Kadierdan Kaheman, Eurika Kaiser, Benjamin Strom, J Nathan Kutz, and Steven L Brunton. Learning discrepancy models from experimental data. arXiv preprint arXiv:1909.08574, 2019

    1909 arXiv

  16. [24]

    Time-delay observables for koopman: Theory and applications

    Mason Kamb, Eurika Kaiser, Steven L Brunton, and J Nathan Kutz. Time-delay observables for koopman: Theory and applications. SIAM Journal on Applied Dynamical Systems, 19(2):886–917, 2020

    2020

  17. [25]

    Determining embedding dimension for phase-space reconstruction using a geometrical construction.Physical review A, 45(6):3403, 1992

    Matthew B Kennel, Reggie Brown, and Henry DI Abarbanel. Determining embedding dimension for phase-space reconstruction using a geometrical construction.Physical review A, 45(6):3403, 1992

    1992

  18. [26]

    Data-driven approximation of the koopman generator: Model reduction, system identification, and control.Physica D: Nonlinear Phenomena, 406:132416, 2020

    Stefan Klus, Frank Nüsken, Péter Koltai, Hao Wu, Ioannis G Kevrekidis, and Christof Schütte. Data-driven approximation of the koopman generator: Model reduction, system identification, and control.Physica D: Nonlinear Phenomena, 406:132416, 2020

    2020

  19. [27]

    Linear predictors for nonlinear dynamical systems: Global stability and control.Automatica, 93:149–160, 2018

    Milan Korda and Igor Mezić. Linear predictors for nonlinear dynamical systems: Global stability and control.Automatica, 93:149–160, 2018

    2018

  20. [28]

    Parsimony as the ultimate regularizer for physics-informed machine learning.Nonlinear Dynamics, 107(3):1801–1817, 2022

    J Nathan Kutz and Steven L Brunton. Parsimony as the ultimate regularizer for physics-informed machine learning.Nonlinear Dynamics, 107(3):1801–1817, 2022

    2022

  21. [29]

    Global modeling of the rössler system from the z-variable.Physics Letters A, 314(5-6):409–427, 2003

    Claudia Lainscsek, Christophe Letellier, and Irina Gorodnitsky. Global modeling of the rössler system from the z-variable.Physics Letters A, 314(5-6):409–427, 2003. 25

    2003

  22. [30]

    Christophe Letellier and Luis A Aguirre. Investigating nonlinear dynamics from time series: The influence of symmetries and the choice of observables.Chaos: An Interdisciplinary Journal of Nonlinear Science, 12(3):549–558, 2002

    2002

  23. [31]

    Relation between observability and differential embeddings for nonlinear dynamics.Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 71(6):066213, 2005

    Christophe Letellier, Luis A Aguirre, and Jean Maquet. Relation between observability and differential embeddings for nonlinear dynamics.Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 71(6):066213, 2005

    2005

  24. [32]

    A symbolic network-based nonlinear theory for dynamical systems observ- ability

    Christophe Letellier, Irene Sendiña-Nadal, Ezequiel Bianco-Martinez, and Murilo S Baptista. A symbolic network-based nonlinear theory for dynamical systems observ- ability. Scientific reports, 8(1):3785, 2018

    2018

  25. [33]

    Control principles of complex systems

    Yang-Yu Liu and Albert-László Barabási. Control principles of complex systems. Reviews of Modern Physics, 88(3):035006, 2016

    2016

  26. [34]

    Controllability of complex networks.nature, 473(7346):167–173, 2011

    Yang-Yu Liu, Jean-Jacques Slotine, and Albert-László Barabási. Controllability of complex networks.nature, 473(7346):167–173, 2011

    2011

  27. [35]

    Observability of complex systems.Proceedings of the National Academy of Sciences, 110(7):2460–2465, 2013

    Yang-Yu Liu, Jean-Jacques Slotine, and Albert-László Barabási. Observability of complex systems.Proceedings of the National Academy of Sciences, 110(7):2460–2465, 2013

    2013

  28. [36]

    Recurrent outbreaks of measles, chickenpox and mumps: I

    Wayne P London and James A Yorke. Recurrent outbreaks of measles, chickenpox and mumps: I. seasonal variation in contact rates.American journal of epidemiology, 98(6):453–468, 1973

    1973

  29. [37]

    Deep learning for universal linear embeddings of nonlinear dynamics.Nature Communications, 9(1):4950, 2018

    Bethany Lusch, J Nathan Kutz, and Steven L Brunton. Deep learning for universal linear embeddings of nonlinear dynamics.Nature Communications, 9(1):4950, 2018

    2018

  30. [38]

    Geometry from a time series.Physical review letters, 45(9):712, 1980

    Norman H Packard, James P Crutchfield, J Doyne Farmer, and Robert S Shaw. Geometry from a time series.Physical review letters, 45(9):712, 1980

    1980

  31. [39]

    Observation of a strange attractor

    J-C Roux, Reuben H Simoyi, and Harry L Swinney. Observation of a strange attractor. Physica D: Nonlinear Phenomena, 8(1-2):257–266, 1983

    1983

  32. [40]

    Embedology.Journal of statistical Physics, 65(3):579–616, 1991

    Tim Sauer, James A Yorke, and Martin Casdagli. Embedology.Journal of statistical Physics, 65(3):579–616, 1991

    1991

  33. [41]

    Do strange attractors govern ecological systems? BioScience, 35(6):342–350, 1985

    William M Schaffer and Mark Kot. Do strange attractors govern ecological systems? BioScience, 35(6):342–350, 1985

    1985

  34. [42]

    Delay embeddings for forced systems

    Jaroslav Stark. Delay embeddings for forced systems. I. deterministic forcing.Journal of Nonlinear Science, 9(3):255–332, 1999

    1999

  35. [43]

    Broomhead, Mark E

    Jaroslav Stark, David S. Broomhead, Mark E. Davies, and John Huke. Delay embeddings for forced systems. II. stochastic forcing.Journal of Nonlinear Science, 13(6):519–577, 2003

    2003

  36. [44]

    Consistent nonparametric regression.The annals of statistics, pages 595–620, 1977

    Charles J Stone. Consistent nonparametric regression.The annals of statistics, pages 595–620, 1977

    1977

  37. [45]

    Detecting causality in complex ecosystems.science, 338(6106):496–500, 2012

    George Sugihara, Robert May, Hao Ye, Chih-hao Hsieh, Ethan Deyle, Michael Fogarty, and Stephan Munch. Detecting causality in complex ecosystems.science, 338(6106):496–500, 2012. 26

    2012

  38. [46]

    Nonlinear forecasting as a way of distinguishing chaos from measurement error in time series.Nature, 344(6268):734–741, 1990

    George Sugihara and Robert M May. Nonlinear forecasting as a way of distinguishing chaos from measurement error in time series.Nature, 344(6268):734–741, 1990

    1990

  39. [47]

    Detecting strange attractors in turbulence

    Floris Takens. Detecting strange attractors in turbulence. InDynamical Systems and Turbulence, Warwick 1980: proceedings of a symposium held at the University of Warwick 1979/80, pages 366–381. Springer, 2006

    1980

  40. [48]

    Spurious dimension from correlation algorithms applied to limited time-series data

    James Theiler. Spurious dimension from correlation algorithms applied to limited time-series data. Physical review A, 34(3):2427, 1986

    1986

  41. [49]

    A data–driven approximation of the koopman operator: Extending dynamic mode decomposition

    Matthew O Williams, Ioannis G Kevrekidis, and Clarence W Rowley. A data–driven approximation of the koopman operator: Extending dynamic mode decomposition. Journal of Nonlinear Science, 25(6):1307–1346, 2015

    2015

  42. [50]

    Optimal transport for parameter identification of chaotic dynamics via invariant measures

    Yunan Yang, Levon Nurbekyan, Elisa Negrini, Robert Martin, and Mirjeta Pasha. Optimal transport for parameter identification of chaotic dynamics via invariant measures. SIAM Journal on Applied Dynamical Systems, 22(1):269–310, 2023

    2023

  43. [51]

    Distinguishing time-delayed causal interactions using convergent cross mapping.Scientific reports, 5(1):14750, 2015

    Hao Ye, Ethan R Deyle, Luis J Gilarranz, and George Sugihara. Distinguishing time-delayed causal interactions using convergent cross mapping.Scientific reports, 5(1):14750, 2015. 27

    2015

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.