pith. sign in

arxiv: 2606.17797 · v1 · pith:G3SO3KTYnew · submitted 2026-06-16 · 🧮 math.OC

Koopman Lifting with Certified Error Bounds for Joint Inference in Nonlinear Networks

Chuansen Peng , Xiaojing Shen , Yunmin Zhu This is my paper

Pith reviewed 2026-06-26 23:25 UTC · model grok-4.3

classification 🧮 math.OC
keywords Koopman operatornetwork topology inferencegroup sparse regularizationnonlinear dynamical systemsjoint state estimationerror boundsADMM algorithm
0
0 comments X

The pith

Koopman operator embedding with separable node-wise dictionaries converts joint state and topology inference in nonlinear networks into a linear filtering task equipped with explicit three-term error bounds.

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

The paper introduces Koopman-GKFA, a method that embeds nonlinear network dynamics into an approximately linear system through a Koopman operator constructed from separable node-wise dictionaries. This embedding makes the unknown graph topology appear as block-sparse structure in the lifted linear operator, which is then recovered by group-sparse convex optimization while node states are estimated by Kalman filtering. A structural homomorphism lemma shows that the block sparsity pattern exactly matches the original graph topology under the separability condition. The framework also supplies a certified mean-squared error bound that decomposes total error into Koopman truncation error, measurement noise, and topology residual, and proves that this bound decreases monotonically as the dictionary dimension increases. Experiments on oscillator models, gene networks, and traffic data show improved recovery compared with extended and unscented Kalman filters.

Core claim

Under a separable-dictionary condition, block sparsity of the lifted coupling operator is isomorphic to the graph topology, enabling group-sparse regularization; the framework supplies a three-term certified mean-squared error bound that decomposes error into Koopman truncation, observation noise, and topology residual, with monotone consistency as dictionary dimension grows.

What carries the argument

The structural homomorphism lemma, which proves that block sparsity in the lifted coupling operator corresponds isomorphically to the original graph topology when the dictionary is separable across nodes.

If this is right

  • Standard linear tools such as Kalman filters become applicable to the lifted system for state estimation.
  • Topology inference reduces to a convex block-structured group-sparse problem solvable by ADMM with linear convergence guarantees.
  • The three-term error bound certifies accuracy of both state estimates and recovered topology.
  • An exponential forgetting factor extends the method to time-varying topologies while preserving convergence.

Where Pith is reading between the lines

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

  • The error decomposition could be used to adaptively enlarge the dictionary only in directions that reduce the dominant error term.
  • If separability holds for common physical dictionaries, the same lifting might apply to inverse problems on other networked systems such as power grids or neural circuits.
  • The polynomial scaling reported for high-dimensional cases suggests the approach could be tested on networks with thousands of nodes where particle filters become intractable.

Load-bearing premise

The dictionary used for the Koopman lifting must be separable node by node so that block sparsity in the lifted operator directly reflects the graph topology.

What would settle it

A concrete network example in which the separable-dictionary condition holds yet the group-sparse solution recovers a coupling operator whose block-sparsity pattern differs from the true graph edges, or in which increasing dictionary dimension fails to reduce the certified mean-squared error.

Figures

Figures reproduced from arXiv: 2606.17797 by Chuansen Peng, Xiaojing Shen, Yunmin Zhu.

Figure 1
Figure 1. Figure 1: State estimation RMSE trajectories of all com [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Topology recovery F1 scores over time, showing that Koopman-GKFA achieves the highest steady-state edge-detection accuracy among all evaluated methods. 0 50 100 150 200 250 Time step k 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 T o p olo g y e r r o r ‖ ̂ A − A‖F/‖A‖F KF-ADMM EKF-ADMM UKF-ADMM SG-KF Koopman-GKFA [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Normalized Frobenius topology error over [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Steady-state state estimation RMSE as a func [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: State estimation RMSE versus measurement [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 9
Figure 9. Figure 9: Ablation on regularization strategy: F1 score comparison among the proposed group-sparse block reg￾ularizer, entrywise ℓ1 penalty, and ridge regression, vali￾dating the critical role of block sparsity. 48 16 32 64 128 Koopman dictionary dimension N 1.00 1.25 1.50 1.75 2.00 2.25 2.50 State RMSE Koopman-GKFA PCRLB 48 16 32 64 128 Koopman dictionary dimension N 0.2 0.3 0.4 0.5 0.6 T o p olo g y Error A A F/ A… view at source ↗
Figure 10
Figure 10. Figure 10: State estimation RMSE and topology error [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Effect of the forgetting factor γ on topology tracking accuracy under stationary and time-varying set￾tings, revealing a U-shaped optimum near γ ∗≈0.97 in the non-stationary regime. roborating the Structural Homomorphism Lemma: un￾der the separable-dictionary condition, the block sparsity pattern of the Koopman operator is isomorphic to the graph topology, and exploiting this structure is critical for rel… view at source ↗
Figure 12
Figure 12. Figure 12: Estimated adjacency matrices of all methods [PITH_FULL_IMAGE:figures/full_fig_p014_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Network graph with edges colored by detection [PITH_FULL_IMAGE:figures/full_fig_p014_13.png] view at source ↗
Figure 16
Figure 16. Figure 16: State estimation RMSE in the joint es￾timation setting comparing Koopman-GKFA with PF￾ADMM (M∈{200, 1000}), UKF-ADMM, and EKF￾ADMM, showing that Koopman-GKFA achieves the low￾est steady-state error despite the superior oracle accuracy of the particle filter. 0 50 100 150 200 250 Time step k 0.0 0.2 0.4 0.6 0.8 1.0 T o p olo g y error ‖ ̂ Ak − A‖F Koopman-GKFA PF-ADMM (M = 1000) UKF-ADMM EKF-ADMM [PITH_FU… view at source ↗
Figure 17
Figure 17. Figure 17: Steady-state topology Frobenius error in the [PITH_FULL_IMAGE:figures/full_fig_p015_17.png] view at source ↗
Figure 20
Figure 20. Figure 20: Topology estimation error versus Hill coeffi [PITH_FULL_IMAGE:figures/full_fig_p016_20.png] view at source ↗
Figure 18
Figure 18. Figure 18: State estimation RMSE versus network size [PITH_FULL_IMAGE:figures/full_fig_p016_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Topology Frobenius error versus network size [PITH_FULL_IMAGE:figures/full_fig_p016_19.png] view at source ↗
Figure 22
Figure 22. Figure 22: Frobenius-norm topology tracking error over [PITH_FULL_IMAGE:figures/full_fig_p017_22.png] view at source ↗
Figure 21
Figure 21. Figure 21: Per-step velocity estimation RMSE on the [PITH_FULL_IMAGE:figures/full_fig_p017_21.png] view at source ↗
Figure 24
Figure 24. Figure 24: ROC curves for directed edge detection on the [PITH_FULL_IMAGE:figures/full_fig_p018_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: F1-score, precision, and recall (mean ± std over five replicates) for all methods on DREAM4, con￾firming the consistent superiority of Koopman-GKFA in topology inference. Ground Truth Koopman-GKFA SG-KF EKF-ADMM UKF-ADMM KF-ADMM 0.0 0.2 0.4 0.6 0.8 1.0 Edge weight [PITH_FULL_IMAGE:figures/full_fig_p018_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Estimated adjacency matrices on the DREAM4 benchmark alongside the ground truth, show￾ing that Koopman-GKFA recovers the sparse regulatory structure with the fewest false positives and false nega￾tives. Topology inference results [PITH_FULL_IMAGE:figures/full_fig_p018_26.png] view at source ↗
read the original abstract

Jointly inferring latent node states and unknown network topology in nonlinear graphical dynamical systems is a fundamental yet largely unsolved problem, where the mutual entanglement of continuous states and discrete structure renders accurate recovery of either quantity critically dependent on the other. We propose \textbf{Koopman-GKFA} (Koopman Group-sparse Kalman Filter--ADMM), a unified framework that lifts nonlinear network dynamics into an approximately linear system via Koopman operator embedding with a separable node-wise dictionary, enabling optimal linear filtering for state estimation and provably convergent convex optimization for topology inference. Three theoretical contributions underpin the framework: (i)~a \emph{structural homomorphism lemma} proving that, under a separable-dictionary condition, block sparsity of the lifted coupling operator is isomorphic to the graph topology, providing the rigorous foundation for group-sparse regularization; (ii)~a block-structured group-sparse ADMM topology subproblem with certified linear convergence, extended by an exponential forgetting factor to track time-varying topologies; and (iii)~a \emph{three-term certified mean-squared error bound} that decomposes total estimation error into Koopman truncation, observation noise, and topology residual components, with monotone consistency established as the dictionary dimension grows. Extensive experiments on synthetic benchmarks (Kuramoto oscillators, Hill-kinetics gene-regulatory networks) and real-world datasets (NGSIM US-101, DREAM4) demonstrate that Koopman-GKFA consistently outperforms EKF-, UKF-, and particle-filter-based joint estimators in both state estimation and topology recovery, while exhibiting polynomial computational scaling and strong robustness in high-dimensional nonlinear settings.

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 proposes Koopman-GKFA, a framework for joint state and topology inference in nonlinear network dynamical systems. It lifts the dynamics via Koopman embedding with a separable node-wise dictionary to obtain an approximately linear system, then applies linear filtering for states and group-sparse ADMM for topology recovery. The central claims are (i) a structural homomorphism lemma showing that, under a separable-dictionary condition, block sparsity of the lifted coupling operator is isomorphic to the underlying graph topology; (ii) a block-structured group-sparse ADMM subproblem with certified linear convergence (extended by exponential forgetting for time-varying topologies); and (iii) a three-term certified MSE bound decomposing error into Koopman truncation, observation noise, and topology residual, with monotone consistency as dictionary dimension increases. Experiments on Kuramoto oscillators, Hill-kinetics networks, NGSIM, and DREAM4 report superior performance over EKF/UKF/particle-filter baselines.

Significance. If the structural homomorphism lemma and the three-term bound can be rigorously established with the separable-dictionary condition explicitly characterized and verified on the benchmark dictionaries, the work would supply a principled, certifiably consistent approach to joint inference that directly ties topology recovery to group-sparse regularization. The monotone consistency result with growing dictionary dimension and the polynomial scaling claim would be notable strengths for high-dimensional nonlinear network problems.

major comments (2)
  1. [Abstract (contribution (i))] Abstract, contribution (i): The structural homomorphism lemma asserts that block sparsity of the lifted coupling operator is isomorphic to graph topology under a 'separable-dictionary condition,' yet the manuscript supplies no explicit characterization of the function class for which separability holds. It is also not shown that the dictionaries employed in the Kuramoto and Hill-kinetics experiments satisfy the condition. This premise is load-bearing for the justification of group-sparse regularization and for the subsequent ADMM and error-bound guarantees.
  2. [Abstract (contribution (iii))] Abstract, contribution (iii): The three-term certified MSE bound is stated to decompose total error into Koopman truncation, observation noise, and topology residual components with monotone consistency as dictionary dimension grows, but the derivation of the decomposition and any numerical verification of bound tightness are not referenced. Without these, the 'certified' claim cannot be assessed as independent of the final performance numbers.
minor comments (2)
  1. [Abstract] The abstract lists four datasets but does not report the specific dictionary dimensions or forgetting-factor values used, which would help readers assess the practical scaling.
  2. [Abstract] Notation for the lifted coupling operator and the separable dictionary should be introduced with a short equation or definition even in the abstract to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting the need for greater explicitness in the theoretical claims. We address each major comment below and will make the indicated revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: Abstract, contribution (i): The structural homomorphism lemma asserts that block sparsity of the lifted coupling operator is isomorphic to graph topology under a 'separable-dictionary condition,' yet the manuscript supplies no explicit characterization of the function class for which separability holds. It is also not shown that the dictionaries employed in the Kuramoto and Hill-kinetics experiments satisfy the condition. This premise is load-bearing for the justification of group-sparse regularization and for the subsequent ADMM and error-bound guarantees.

    Authors: We agree that an explicit characterization of the separable-dictionary condition is required. The condition is used in the proof of the homomorphism lemma (Lemma 1) but is not stated as a standalone definition. In the revision we will insert a formal definition (Definition 2) characterizing separable dictionaries as those that are direct sums of node-local functions with no cross-node interactions. We will also add Proposition 1 proving that common node-wise bases (polynomials, Fourier, RBFs) satisfy the condition by construction, and we will explicitly verify and state that the dictionaries used in the Kuramoto (node-wise Fourier) and Hill-kinetics (node-wise sigmoidal) experiments meet this definition. These additions will be placed immediately before the lemma statement. revision: yes

  2. Referee: Abstract, contribution (iii): The three-term certified MSE bound is stated to decompose total error into Koopman truncation, observation noise, and topology residual components with monotone consistency as dictionary dimension grows, but the derivation of the decomposition and any numerical verification of bound tightness are not referenced. Without these, the 'certified' claim cannot be assessed as independent of the final performance numbers.

    Authors: The decomposition and monotone consistency are derived in Theorem 3 (Section 4) via the triangle inequality applied to the lifted-state error, with each term bounded separately and the dictionary-dimension consistency following from the universal approximation property of the chosen function class. The proof is in Appendix B. We acknowledge, however, that the abstract and main-text statements lack explicit forward references to the theorem and that no numerical check of bound tightness is provided. In the revision we will add cross-references in the abstract and Section 4, include a new figure (Figure 8) comparing the certified bound against empirical MSE on the Kuramoto and Hill-kinetics examples, and expand the appendix proof with an additional intermediate lemma for readability. revision: yes

Circularity Check

0 steps flagged

No circularity: derivations rest on external assumptions and independent proofs

full rationale

The paper's core claims—the structural homomorphism lemma under an explicitly stated separable-dictionary condition, the block-structured ADMM with linear convergence, and the three-term MSE bound with monotone consistency—are presented as derived from the Koopman lifting and convex optimization setup without reducing any prediction or result to a fitted parameter or self-referential definition. No equations equate outputs to inputs by construction, no fitted quantities are relabeled as predictions, and no load-bearing steps rely on self-citations. The separable-dictionary condition functions as a premise rather than a derived or tautological element, leaving the framework self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

Based solely on abstract; the separable-dictionary condition is the central modeling assumption enabling the isomorphism, while dictionary dimension and forgetting factor are mentioned as tunable elements that affect consistency and tracking.

free parameters (2)
  • dictionary dimension
    Grows to achieve monotone consistency of the MSE bound
  • exponential forgetting factor
    Introduced to track time-varying topologies in the ADMM subproblem
axioms (1)
  • domain assumption separable-dictionary condition
    Invoked to establish that block sparsity of the lifted coupling operator is isomorphic to graph topology

pith-pipeline@v0.9.1-grok · 5822 in / 1266 out tokens · 37517 ms · 2026-06-26T23:25:33.097812+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

45 extracted references

  1. [1]

    Power system dynamic state es- timation: Motivations, definitions, methodologies, and future work

    Junbo Zhao, Antonio G´ omez-Exp´ osito, Marcos Netto, Lamine Mili, Ali Abur, Vladimir Terzija, In- nocent Kamwa, Bikash Pal, Abhinav Kumar Singh, Junjian Qi, et al. Power system dynamic state es- timation: Motivations, definitions, methodologies, and future work. IEEE Transactions on Power Sys- tems, 34(4):3188–3198, 2019

    2019

  2. [2]

    How to infer gene networks from expression profiles

    Mukesh Bansal, Vincenzo Belcastro, Alberto Ambesi-Impiombato, and Diego Di Bernardo. How to infer gene networks from expression profiles. Molecular Systems Biology, 3:78, 2007

    2007

  3. [3]

    W. Helly. Simulation of bottlenecks in single-lane traffic flow. In Proceedings of Symposium on Theory Traffic Flow, pages 207–238, 1959

    1959

  4. [4]

    Nonlinear state estimation for hu- manoid robot walking

    Stylianos Piperakis, Maria Koskinopoulou, and Panos Trahanias. Nonlinear state estimation for hu- manoid robot walking. IEEE Robotics and Automa- tion Letters, 3(4):3347–3354, 2018

    2018

  5. [5]

    Learning graphs from data: A signal representation perspective

    Xiaowen Dong, Dorina Thanou, Michael Rabbat, and Pascal Frossard. Learning graphs from data: A signal representation perspective. IEEE Signal Pro- cessing Magazine, 36(3):44–63, 2019

    2019

  6. [6]

    Network topology in- ference from spectral templates

    Santiago Segarra, Antonio G Marques, Gonzalo Ma- teos, and Alejandro Ribeiro. Network topology in- ference from spectral templates. IEEE Transactions on Signal and Information Processing over Networks, 3(3):467–483, 2017

    2017

  7. [7]

    Connecting the dots: Identifying network structure via graph signal pro- cessing

    Gonzalo Mateos, Santiago Segarra, Antonio G Mar- ques, and Alejandro Ribeiro. Connecting the dots: Identifying network structure via graph signal pro- cessing. IEEE Signal Processing Magazine, 36(3):16– 43, 2019

    2019

  8. [8]

    Topology inference for network sys- tems: Causality perspective and nonasymptotic per- formance

    Yushan Li, Jianping He, Cailian Chen, and Xin- ping Guan. Topology inference for network sys- tems: Causality perspective and nonasymptotic per- formance. IEEE Transactions on Automatic Control, 69(6):3483–3498, 2023

    2023

  9. [9]

    Infer- ring topology of network systems by few excitations: Probability analysis and fusion

    Qing Jiao, Yushan Li, and Jianping He. Infer- ring topology of network systems by few excitations: Probability analysis and fusion. IEEE Transactions on Automatic Control, 2025

    2025

  10. [10]

    A new approach to linear filtering and prediction problems

    Rudolph Emil Kalman. A new approach to linear filtering and prediction problems. Journal of Basic Engineering, 82(1):35–45, 1960

    1960

  11. [11]

    Adaptation, learning, and optimiza- tion over networks

    Ali H Sayed. Adaptation, learning, and optimiza- tion over networks. Foundations and Trends ® in Machine Learning, 7(4-5):311–801, 2014

    2014

  12. [12]

    Joint graph learning and sig- nal recovery via Kalman filter for multivariate auto-regressive processes

    Mahmoud Ramezani-Mayiami and Baltasar Beferull-Lozano. Joint graph learning and sig- nal recovery via Kalman filter for multivariate auto-regressive processes. In Proceedings of Eu- ropean Signal Processing Conference (EUSIPCO) , pages 907–911. IEEE, 2018

    2018

  13. [13]

    Joint topology learning and graph signal recovery via Kalman filter in causal data processes

    Mahmoud Ramezani-Mayiami and Baltasar Beferull-Lozano. Joint topology learning and graph signal recovery via Kalman filter in causal data processes. In Proceedings of International Workshop on Machine Learning for Signal Processing (MLSP), pages 1–6. IEEE, 2018

    2018

  14. [14]

    Graphical in- ference in linear-Gaussian state-space models

    V´ ıctor Elvira and Emilie Chouzenoux. Graphical in- ference in linear-Gaussian state-space models. IEEE Transactions on Signal Processing , 70:4757–4771, 2022

    2022

  15. [15]

    Joint state estimation and topology in- ference for graphical dynamical systems

    Pengfei Fang, Wenling Li, Jia Song, Xiaoming Li, and Li Ma. Joint state estimation and topology in- ference for graphical dynamical systems. Signal Pro- cessing, 237:110070, 2025

    2025

  16. [16]

    Distributed optimization and statisti- cal learning via the alternating direction method of multipliers

    Parikh Neal, Chu Eric, Peleato Borja, and Eckstein Jonathan. Distributed optimization and statisti- cal learning via the alternating direction method of multipliers. Foundations and Trends ® in Machine learning, 3(1):1–122, 2011

    2011

  17. [17]

    On the linear convergence of the alternating direction method of multipliers

    Mingyi Hong and Zhi-Quan Luo. On the linear convergence of the alternating direction method of multipliers. Mathematical Programming, 162(1):165– 199, 2017

    2017

  18. [18]

    Application of state-space meth- ods to navigation problems

    Stanley F Schmidt. Application of state-space meth- ods to navigation problems. In Advances in Control Systems, volume 3, pages 293–340. Elsevier, 1966

    1966

  19. [19]

    Unscented filtering and nonlinear estimation

    Simon J Julier and Jeffrey K Uhlmann. Unscented filtering and nonlinear estimation. Proceedings of the IEEE, 92(3):401–422, 2004

    2004

  20. [20]

    A tutorial on particle filters for online nonlinear/non-Gaussian Bayesian tracking

    M Sanjeev Arulampalam, Simon Maskell, Neil Gor- don, and Tim Clapp. A tutorial on particle filters for online nonlinear/non-Gaussian Bayesian tracking. IEEE Transactions on Signal Processing, 50(2):174– 188, 2002

    2002

  21. [21]

    Extended Kalman filter for graph signals in nonlin- ear dynamic systems

    Guy Sagi, Nir Shlezinger, and Tirza Routtenberg. Extended Kalman filter for graph signals in nonlin- ear dynamic systems. In Proceedings of IEEE Inter- national Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE, 2023. 25

    2023

  22. [22]

    Unscented Kalman filter of graph signals

    Wenling Li, Xiaoyan Fu, Bin Zhang, and Yang Liu. Unscented Kalman filter of graph signals. Automat- ica, 148:110796, 2023

    2023

  23. [23]

    State-space network topology identifi- cation from partial observations

    Mario Coutino, Elvin Isufi, Takanori Maehara, and Geert Leus. State-space network topology identifi- cation from partial observations. IEEE Transactions on Signal and Information Processing over Networks, 6:211–225, 2020

    2020

  24. [24]

    In- ductive representation learning on large graphs

    Will Hamilton, Zhitao Ying, and Jure Leskovec. In- ductive representation learning on large graphs. In Proceeding of Advances in Neural Information Pro- cessing Systems (NeurIPS) , volume 30, 2017

    2017

  25. [25]

    GSP- KalmanNet: Tracking graph signals via neural-aided Kalman filtering

    Itay Buchnik, Guy Sagi, Nimrod Leinwand, Yuval Loya, Nir Shlezinger, and Tirza Routtenberg. GSP- KalmanNet: Tracking graph signals via neural-aided Kalman filtering. IEEE Transactions on Signal Pro- cessing, 72:3700–3716, 2024

    2024

  26. [26]

    Hamiltonian systems and transformation in Hilbert space

    Bernard O Koopman. Hamiltonian systems and transformation in Hilbert space. Proceedings of the National Academy of Sciences , 17(5):315–318, 1931

    1931

  27. [27]

    Spectral properties of dynamical sys- tems, model reduction and decompositions

    Igor Mezi´ c. Spectral properties of dynamical sys- tems, model reduction and decompositions. Nonlin- ear Dynamics, 41(1):309–325, 2005

    2005

  28. [28]

    Dynamic mode decomposition of numerical and experimental data

    Peter J Schmid. Dynamic mode decomposition of numerical and experimental data. Journal of Fluid Mechanics, 656:5–28, 2010

    2010

  29. [29]

    A data–driven approxima- tion of the Koopman operator: Extending dynamic mode decomposition

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

    2015

  30. [30]

    On convergence of extended dynamic mode decomposition to the Koopman operator

    Milan Korda and Igor Mezi´ c. On convergence of extended dynamic mode decomposition to the Koopman operator. Journal of Nonlinear Science , 28(2):687–710, 2018

    2018

  31. [31]

    Koopman invariant subspaces and finite linear representations of non- linear dynamical systems for control

    Steven L Brunton, Bingni W Brunton, Joshua L Proctor, and J Nathan Kutz. Koopman invariant subspaces and finite linear representations of non- linear dynamical systems for control. PloS one , 11(2):e0150171, 2016

    2016

  32. [32]

    Generalizing Koopman theory to allow for in- puts and control

    Joshua L Proctor, Steven L Brunton, and J Nathan Kutz. Generalizing Koopman theory to allow for in- puts and control. SIAM Journal on Applied Dynam- ical Systems, 17(1):909–930, 2018

    2018

  33. [33]

    Koopman- based lifting techniques for nonlinear systems identi- fication

    Alexandre Mauroy and Jorge Goncalves. Koopman- based lifting techniques for nonlinear systems identi- fication. IEEE Transactions on Automatic Control , 65(6):2550–2565, 2019

    2019

  34. [34]

    Bilineariza- tion, reachability, and optimal control of control- affine nonlinear systems: A Koopman spectral ap- proach

    Debdipta Goswami and Derek A Paley. Bilineariza- tion, reachability, and optimal control of control- affine nonlinear systems: A Koopman spectral ap- proach. IEEE Transactions on Automatic Control , 67(6):2715–2728, 2021

    2021

  35. [35]

    Optimal construc- tion of Koopman eigenfunctions for prediction and control

    Milan Korda and Igor Mezi´ c. Optimal construc- tion of Koopman eigenfunctions for prediction and control. IEEE Transactions on Automatic Control , 65(12):5114–5129, 2020

    2020

  36. [36]

    Deep learning for universal linear embeddings of nonlinear dynamics

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

    2018

  37. [37]

    Model selection and estima- tion in regression with grouped variables

    Ming Yuan and Yi Lin. Model selection and estima- tion in regression with grouped variables. Journal of the Royal Statistical Society Series B: Statistical Methodology, 68(1):49–67, 2006

    2006

  38. [38]

    Tracking switched dynamic network topologies from informa- tion cascades

    Brian Baingana and Georgios B Giannakis. Tracking switched dynamic network topologies from informa- tion cascades. IEEE Transactions on Signal Process- ing, 65(4):985–997, 2016

    2016

  39. [39]

    Proximal splitting methods in signal processing

    Patrick L Combettes and Jean-Christophe Pesquet. Proximal splitting methods in signal processing. In Fixed-Point Algorithms for Inverse Problems in Science and Engineering , pages 185–212. Springer, 2011

    2011

  40. [40]

    On model selection con- sistency of LASSO

    Peng Zhao and Bin Yu. On model selection con- sistency of LASSO. Journal of Machine Learning Research, 7:2541–2563, 2006

    2006

  41. [41]

    Consistency of the group LASSO and multiple kernel learning

    Francis R Bach. Consistency of the group LASSO and multiple kernel learning. Journal of Machine Learning Research, 9(6), 2008

    2008

  42. [42]

    Chemical turbulence

    Yoshiki Kuramoto. Chemical turbulence. In Chemi- cal Oscillations, Waves, and Turbulence , pages 111–

  43. [43]

    On the assessment of vehicle trajectory data accuracy and application to the Next Gener- ation SIMulation (NGSIM) program data

    Vincenzo Punzo, Maria Teresa Borzacchiello, and Bi- agio Ciuffo. On the assessment of vehicle trajectory data accuracy and application to the Next Gener- ation SIMulation (NGSIM) program data. Trans- portation Research Part C: Emerging Technologies , 19(6):1243–1262, 2011

    2011

  44. [44]

    Generating realistic in silico gene networks for performance assessment of reverse engineering methods

    Daniel Marbach, Thomas Schaffter, Claudio Mat- tiussi, and Dario Floreano. Generating realistic in silico gene networks for performance assessment of reverse engineering methods. Journal of Computa- tional Biology, 16(2):229–239, 2009

    2009

  45. [45]

    Matrix Anal- ysis

    Roger A Horn and Charles R Johnson. Matrix Anal- ysis. Cambridge University Press, 2 edition, 2012. 26

    2012