pith. machine review for the scientific record. sign in

arxiv: 2604.15238 · v2 · submitted 2026-04-16 · 📡 eess.SY · cs.LG· cs.SY· math.OC

Recognition: unknown

A Nonlinear Separation Principle via Contraction Theory: Applications to Neural Networks, Control, and Learning

Anand Gokhale , Anton V. Proskurnikov , Yu Kawano , Francesco Bullo
Authors on Pith no claims yet

Pith reviewed 2026-05-10 10:27 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SYmath.OC
keywords nonlinear separation principlecontraction theoryglobal exponential stabilityrecurrent neural networksLMI conditionsoutput feedback controlimplicit neural networks
0
0 comments X

The pith

If both a state-feedback controller and an observer are contracting, their interconnection is globally exponentially stable.

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

The paper establishes a nonlinear separation principle using contraction theory. This principle shows that global exponential stability holds for the closed-loop system when the controller and observer each satisfy contraction conditions independently. It provides LMI-based certificates for the contractivity of firing-rate and Hopfield neural network models, with structural results favoring monotone activations. These tools are then applied to design controllers and observers for plants modeled by recurrent neural networks, including low-gain integral action for reference tracking. An algebraic parameterization of the LMIs also yields a class of implicit neural networks that achieve competitive accuracy on image classification tasks.

Core claim

The central claim is that the interconnection of a contracting state-feedback controller and a contracting observer yields global exponential stability of the closed-loop system. This separation principle is derived via contraction theory and extended to parametric uncertainties and equilibrium tracking. Sharp LMI conditions are derived to certify contractivity for firing-rate and Hopfield RNNs, with structural relationships showing that monotone activations maximize the admissible weight space. These are used to solve output reference tracking for RNN plants with LMI synthesis and low-gain integral control, plus an unconstrained parameterization for implicit NNs.

What carries the argument

The contraction property of a dynamical system, which requires a metric in which the symmetric part of the Jacobian is uniformly negative definite, ensuring exponential convergence of trajectories independent of initial conditions.

If this is right

  • Global exponential stability of the full closed-loop system follows directly from separate contraction of the controller and observer.
  • LMI conditions certify contractivity for firing-rate and Hopfield RNNs and extend to graph RNNs and interconnected systems.
  • LMI synthesis methods exist for designing feedback controllers and observers for plants modeled by recurrent neural networks.
  • A low-gain integral controller can be added to eliminate steady-state error while preserving the contraction property.
  • An exact algebraic parameterization of the contraction LMIs enables the design of expressive implicit neural networks.

Where Pith is reading between the lines

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

  • The separation principle could simplify verification of stability in learning-based control by checking contraction on the controller and observer separately rather than on the full interconnection.
  • The algebraic parameterization of LMIs might support training of implicit networks that inherit contraction-based stability guarantees without additional constraints.
  • The approach suggests a route to modular design in domains such as robotics where sensor and actuator subsystems can be made contracting independently.

Load-bearing premise

Both the controller and the observer are assumed to satisfy the contraction condition on their Jacobians independently of each other and of the plant.

What would settle it

A counterexample consisting of a specific nonlinear plant, controller, and observer where both the controller and observer are contracting yet the closed-loop system fails to be globally exponentially stable would disprove the separation principle.

Figures

Figures reproduced from arXiv: 2604.15238 by Anand Gokhale, Anton V. Proskurnikov, Francesco Bullo, Yu Kawano.

Figure 1
Figure 1. Figure 1: A summary of relationships for the contractivity conditions from Table I. The sets W(·, ·, ·) are described in Theorem 14. The discrete time CONE condition restricts the weight matrices the most, whereas the continuous time MONE condition enables maximum expressivity. Theorem 14 (Reductions and duality of the certificates): Let W(M, T , N ) denote the set of synaptic matrices W satisfying the contraction c… view at source ↗
Figure 2
Figure 2. Figure 2: For a two tank system modeled by an FRNN, we utilize our design mechanism for the full state feedback controller, the contracting observer and the integral gain to design a closed loop system capable of tracking references, validating the proposed theoretical results. trace, and a positive determinant. Both these conditions are met under assumption (A4). By continuity, M(ε) + ηI3 remains Hurwitz for small … view at source ↗
read the original abstract

This paper establishes a nonlinear separation principle based on contraction theory and derives sharp stability conditions for recurrent neural networks (RNNs). First, we introduce a nonlinear separation principle that guarantees global exponential stability for the interconnection of a contracting state-feedback controller and a contracting observer, alongside parametric extensions for robustness and equilibrium tracking. Second, we derive sharp linear matrix inequality (LMI) conditions that guarantee the contractivity of both firing rate and Hopfield neural network architectures. We establish structural relationships among these certificates-demonstrating that continuous-time models with monotone non-decreasing activations maximize the admissible weight space-and extend these stability guarantees to interconnected systems and Graph RNNs. Third, we combine our separation principle and LMI framework to solve the output reference tracking problem for RNN-modeled plants. We provide LMI synthesis methods for feedback controllers and observers, and rigorously design a low-gain integral controller to eliminate steady-state error. Finally, we derive an exact, unconstrained algebraic parameterization of our contraction LMIs to design highly expressive implicit neural networks, achieving competitive accuracy and parameter efficiency on standard image classification benchmarks.

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

0 major / 3 minor

Summary. The manuscript establishes a nonlinear separation principle based on contraction theory, proving that the interconnection of a contracting state-feedback controller and a contracting observer yields global exponential stability of the closed-loop error dynamics. It derives LMI conditions certifying contractivity for firing-rate and Hopfield RNN architectures, demonstrates structural relationships (including maximization of admissible weight space for monotone activations), extends the framework to interconnected systems and Graph RNNs, and applies the results to output reference tracking via LMI-synthesized controllers plus a low-gain integral term. Finally, it supplies an exact algebraic parameterization of the contraction LMIs to construct implicit neural networks and reports competitive accuracy and parameter efficiency on standard image-classification benchmarks.

Significance. If the central derivations hold, the work supplies a rigorous, metric-based route to stability guarantees for nonlinear plants that avoids linearization and supplies constructive LMI certificates together with an unconstrained algebraic parameterization. These elements are directly usable for controller/observer synthesis in RNN-modeled systems and for designing stable implicit networks. The explicit differential-dynamics proof of the separation implication and the provision of reproducible LMI conditions constitute clear strengths.

minor comments (3)
  1. The construction of the combined contraction metric when the controller and observer employ distinct metrics should be written out explicitly (with the resulting differential inequality) rather than left implicit in the interconnection argument.
  2. In the LMI section, the claim that the certificates are 'sharp' would be strengthened by a brief remark on whether the LMIs are also necessary or only sufficient; a simple scalar example illustrating the boundary of the admissible weight set would clarify the structural relationship.
  3. The benchmark tables for the implicit-network parameterization should report the number of parameters and test accuracy for at least one standard baseline (e.g., a comparable explicit RNN or feed-forward network) to make the 'parameter efficiency' claim directly comparable.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive and accurate summary of our manuscript on the nonlinear separation principle via contraction theory, including its LMI certificates for RNN contractivity, extensions to interconnected and graph-based systems, output tracking applications, and implicit neural network parameterization. We appreciate the recognition of the work's significance for stability guarantees without linearization and for constructive synthesis methods. The recommendation for minor revision is noted. However, the report contains no specific major comments to address point by point.

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The central result is a conditional theorem: if a state-feedback controller and observer are each contracting (w.r.t. possibly different metrics), then their interconnection is globally exponentially stable. The manuscript supplies an explicit proof via the differential dynamics of the combined system. LMI certificates for contractivity of firing-rate and Hopfield RNNs are derived directly from the Jacobian contraction condition, not fitted to target data. The algebraic parameterization of the LMIs for implicit networks is an exact re-expression of those same certificates. No load-bearing step reduces by construction to its own inputs, no self-citation chain is invoked to justify the implication, and the contraction assumption is stated as given rather than derived from the target result. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the existence of contraction metrics for the controller and observer, the validity of the LMI conditions as sufficient certificates, and the assumption that the plant can be modeled by an RNN. No free parameters are explicitly fitted in the abstract, but the choice of contraction metric is a potential free parameter.

axioms (2)
  • domain assumption The system dynamics admit a contraction metric under which the Jacobian satisfies a uniform negative definiteness condition.
    Invoked to guarantee global exponential stability of the closed-loop interconnection.
  • domain assumption The activation functions are monotone non-decreasing.
    Used to maximize the admissible weight space in the LMI conditions for RNN contractivity.

pith-pipeline@v0.9.0 · 5510 in / 1485 out tokens · 39091 ms · 2026-05-10T10:27:28.202877+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

42 extracted references · 33 canonical work pages · 3 internal anchors

  1. [1]

    IFAC-PapersOnLine48(28), 1082–1087 (2015) https://doi.org/10.1016/j.ifacol

    V . Andrieu and S. Tarbouriech. LMI conditions for contraction and synchronization. InIFAC Symposium on Nonlinear Control Sys- tems, volume 52, pages 616–621, 2019.doi:10.1016/j.ifacol. 2019.12.030

    work page doi:10.1016/j.ifacol 2019

  2. [2]

    A. N. Atassi and H. K. Khalil. A separation principle for the stabilization of a class of nonlinear systems.IEEE Transactions on Automatic Control, 44(9):1672–1687, 2002.doi:10.1109/9.788534

    work page doi:10.1109/9.788534 2002

  3. [3]

    S. Bai, J. Z. Kolter, and V . Koltun. Deep equilibrium models. In Advances in Neural Information Processing Systems, 2019. URL: https://arxiv.org/abs/1909.01377

    work page arXiv 2019

  4. [4]

    Baker, Q

    J. Baker, Q. Wang, C. D. Hauck, and B. Wang. Implicit graph neural networks: A monotone operator viewpoint. InInt. Conf. on Machine Learning, volume 202 ofProceedings of Machine Learning Research, pages 1521–1548, 2023. URL: https://proceedings.mlr.press/ v202/baker23a.html

    2023

  5. [5]

    D. S. Bernstein.Matrix Mathematics. Princeton University Press, 2 edition, 2009, ISBN 0691140391

    2009

  6. [6]

    S. Boyd, L. El Ghaoui, E. Feron, and V . Balakrishnan.Linear Matrix Inequalities in System and Control Theory. SIAM, 1994, ISBN 089871334X

    1994

  7. [7]

    Bullo.Contraction Theory for Dynamical Systems

    F. Bullo.Contraction Theory for Dynamical Systems. Kindle Direct Publishing, 1.3 edition, 2026, ISBN 979-8836646806. URL: https:// fbullo.github.io/ctds

    2026

  8. [8]

    Rme-gan: A learning framework for radio map estimation based on conditional generative adversarial network,

    V . Centorrino, A. Gokhale, A. Davydov, G. Russo, and F. Bullo. Euclidean contractivity of neural networks with symmetric weights. IEEE Control Systems Letters, 7:1724–1729, 2023.doi:10.1109/ LCSYS.2023.3278250

    work page arXiv 2023

  9. [9]

    Colombino, E

    M. Colombino, E. Dall’Anese, and A. Bernstein. Online optimization as a feedback controller: Stability and tracking.IEEE Transactions on Control of Network Systems, 7(1):422–432, 2020.doi:10.1109/ TCNS.2019.2906916

    work page arXiv 2020

  10. [10]

    D’Alto and M

    L. D’Alto and M. Corless. Incremental quadratic stability.Numerical Algebra, Control and Optimization, 3:175–201, 2013.doi:10.3934/ naco.2013.3.175

    2013

  11. [11]

    D’Amico, A

    W. D’Amico, A. La Bella, and M. Farina. An incremental input-to- state stability condition for a class of recurrent neural networks.IEEE Transactions on Automatic Control, 69(4):2221–2236, 2024.doi:10. 1109/tac.2023.3327937

    work page arXiv 2024

  12. [13]

    Davydov, V

    A. Davydov, V . Centorrino, A. Gokhale, G. Russo, and F. Bullo. Time- varying convex optimization: A contraction and equilibrium tracking approach.IEEE Transactions on Automatic Control, 70(11):7446–7460, 2025.doi:10.1109/TAC.2025.3576043

    work page doi:10.1109/tac.2025.3576043 2025

  13. [14]

    Drgona, A

    J. Drgona, A. Tuor, J. Koch, M. Shapiro, B. Jacob, and D. Vrabie. NeuroMANCER: Neural Modules with Adaptive Nonlinear Constraints and Efficient Regularizations. 2023. URL: https://github.com/pnnl/ neuromancer

    2023

  14. [15]

    El Ghaoui, F

    L. El Ghaoui, F. Gu, B. Travacca, A. Askari, and A. Tsai. Implicit deep learning.SIAM Journal on Mathematics of Data Science, 3(3):930–958, 2021.doi:10.1137/20M1358517

    work page doi:10.1137/20m1358517 2021

  15. [16]

    Esfandiari and H

    F. Esfandiari and H. K. Khalil. Output feedback stabilization of fully linearizable systems.International Journal of Control, 56(5):1007– 1037, 1992.doi:10.1080/00207179208934355

    work page doi:10.1080/00207179208934355 1992

  16. [17]

    Fazlyab, A

    M. Fazlyab, A. Robey, H. Hassani, M. Morari, and G. J. Pappas. Efficient and accurate estimation of Lipschitz constants for deep neural networks. InAdvances in Neural Information Processing Systems, 2019. URL: https://arxiv.org/abs/1906.04893

    work page arXiv 2019

  17. [18]

    Small-gain analysis of exponential incremental input/output-to-state stability for large-scale distributed systems

    C. Gatke, J. D. Schiller, and M. A. M ¨uller. Small-gain analysis of exponential incremental input/output-to-state stability for large-scale distributed systems.arXiv preprint arXiv:2604.07081, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  18. [19]

    Giaccagli, V

    M. Giaccagli, V . Andrieu, S. Tarbouriech, and D. Astolfi. LMI conditions for contraction, integral action, and output feedback stabilization for a class of nonlinear systems.Automatica, 154:111106, 2023.doi: 10.1016/j.automatica.2023.111106

    work page doi:10.1016/j.automatica.2023.111106 2023

  19. [20]

    T., & Le, X.-B

    A. Gokhale, A. V . Proskurnikov, Y . Kawano, and F. Bullo. Contracting neural networks: Sharp LMI conditions with applications to integral control and deep learning. InIEEE Conf. on Decision and Control, Honolulu, Hawaii, 2026. Submitted.doi:10.48550/arXiv.2604. 00119

    work page doi:10.48550/arxiv.2604 2026

  20. [21]

    F. Gu, H. Chang, W. Zhu, S. Sojoudi, and L. El Ghaoui. Implicit graph neural networks. InAdvances in Neural Information Processing Systems, 2020. URL: https://arxiv.org/abs/2009.06211

    work page arXiv 2020

  21. [22]

    J. P. Hespanha.Linear Systems Theory. Princeton University Press, 2009, ISBN 0691140219

    2009

  22. [23]

    Jafarpour, A

    S. Jafarpour, A. Davydov, A. V . Proskurnikov, and F. Bullo. Robust implicit networks via non-Euclidean contractions. InAdvances in Neural Information Processing Systems, December 2021.doi:10. 48550/arXiv.2106.03194

    work page arXiv 2021

  23. [24]

    T. N. Kipf and M. Welling. Semi-supervised classification with graph convolutional networks.arXiv preprint arXiv:1609.02907, 2016

    work page internal anchor Pith review arXiv 2016

  24. [25]

    Kozachkov, M

    L. Kozachkov, M. Ennis, and J.-J. E. Slotine. RNNs of RNNs: Recursive construction of stable assemblies of recurrent neural networks. In Advances in Neural Information Processing Systems, December 2022. doi:10.48550/arXiv.2106.08928

    work page doi:10.48550/arxiv.2106.08928 2022

  25. [26]

    J. Liu, L. Ding, S. Osher, and W. Yin. Implicit models: Expressive power scales with test-time compute.arXiv preprint, 2025.doi: 10.48550/arXiv.2510.03638

    work page doi:10.48550/arxiv.2510.03638 2025

  26. [27]

    I. R. Manchester and J.-J. E. Slotine. Transverse contraction criteria for existence, stability, and robustness of a limit cycle.Systems & Control Letters, 63:32–38, 2014.doi:10.1016/j.sysconle.2013.10. 005

    work page doi:10.1016/j.sysconle.2013.10 2014

  27. [28]

    Nicolás, The bar derived category of a curved dg algebra, Journal of Pure and Applied Algebra 212 (2008) 2633–2659

    A. Nikolakopoulou, M. Hong, and R. D. Braatz. Dynamic state feedback controller and observer design for dynamic artificial neural network models.Automatica, 146:110622, 2022.doi:10.1016/j. automatica.2022.110622

    work page doi:10.1016/j 2022

  28. [29]

    M. Poli, S. Massaroli, J. Park, A. Yamashita, H. Asama, and J. Park. Graph neural ordinary differential equations.arXiv preprint arXiv:1911.07532, 2019

    work page arXiv 1911

  29. [30]

    Revay, R

    M. Revay, R. Wang, and I. R. Manchester. Lipschitz bounded equi- librium networks.arXiv preprint arXiv:2010.01732, 2020.doi: 10.48550/arXiv.2010.01732

    work page doi:10.48550/arxiv.2010.01732 2010

  30. [31]

    Revay, R

    M. Revay, R. Wang, and I. R. Manchester. A convex parameterization of robust recurrent neural networks.IEEE Control Systems Letters, 5(4):1363–1368, 2021.doi:10.1109/LCSYS.2020.3038221

    work page doi:10.1109/lcsys.2020.3038221 2021

  31. [32]

    D. J. Rose. Matrix identities of the fast Fourier transform.Linear Algebra and its Applications, 29:423–443, 1980.doi:10.1016/ 0024-3795(80)90253-0

    1980

  32. [33]

    C. J. Rozell, D. H. Johnson, R. G. Baraniuk, and B. A. Olshausen. Sparse coding via thresholding and local competition in neural circuits.Neu- ral Computation, 20(10):2526–2563, 2008.doi:10.1162/neco. 2008.03-07-486

    work page doi:10.1162/neco 2008

  33. [34]

    Schoukens and J

    M. Schoukens and J. P. No ¨el. Three benchmarks addressing open challenges in nonlinear system identification.IFAC World Congress, 50(1):446–451, 2017.doi:10.1016/j.ifacol.2017.08.071

    work page doi:10.1016/j.ifacol.2017.08.071 2017

  34. [35]

    Shiriaev, R

    A. Shiriaev, R. Johansson, A. Robertsson, and L. Freidovich. Separation principle for a class of nonlinear feedback systems augmented with observers.IFAC Proceedings Volumes, 41(2):6196–6201, 2008.doi: 10.3182/20080706-5-KR-1001.01046

    work page doi:10.3182/20080706-5-kr-1001.01046 2008

  35. [36]

    J. W. Simpson-Porco. Analysis and synthesis of low-gain integral con- trollers for nonlinear systems.IEEE Transactions on Automatic Control, 66(9):4148–4159, 2021.doi:10.1109/tac.2020.3035569

    work page doi:10.1109/tac.2020.3035569 2021

  36. [37]

    E. D. Sontag and Y . Wang. Output-to-state stability and detectability of nonlinear systems.Systems & Control Letters, 29(5):279–290, 1997. doi:10.1016/S0167-6911(97)90013-X

    work page doi:10.1016/s0167-6911(97)90013-x 1997

  37. [38]

    D. W. Tank and J. J. Hopfield. Simple ”neural” optimization networks: An A/D converter, signal decision circuit, and a linear programming circuit.IEEE Transactions on Circuits and Systems, 33(5):533–541, 1986.doi:10.1109/TCS.1986.1085953

    work page doi:10.1109/tcs.1986.1085953 1986

  38. [39]

    Teel and L

    A. Teel and L. Praly. Tools for semiglobal stabilization by partial state and output feedback.SIAM Journal on Control and Optimization, 33(5):1443–1488, 1995.doi:10.1137/S0363012992241430

    work page doi:10.1137/s0363012992241430 1995

  39. [40]

    Vidyasagar

    M. Vidyasagar. On the stabilization of nonlinear systems using state detection.IEEE Transactions on Automatic Control, 25(3):504–509, 1980.doi:10.1109/TAC.1980.1102376

    work page doi:10.1109/tac.1980.1102376 1980

  40. [41]

    Winston and J

    E. Winston and J. Z. Kolter. Monotone operator equilibrium networks. InAdvances in Neural Information Processing Systems, 2020. URL: https://arxiv.org/abs/2006.08591

    work page arXiv 2020

  41. [42]

    Xhonneux, M

    L.-P. Xhonneux, M. Qu, and J. Tang. Continuous graph neural networks. InInternational conference on machine learning, pages 10432–10441. PMLR, 2020

    2020

  42. [43]

    Controller Design for Structured State-space Models via Contraction Theory

    M. Zakwan, V . Gupta, A. Karimi, E. C. Balta, and G. Ferrari-Trecate. Controller design for structured state-space models via contraction theory.arXiv preprint arXiv:2604.07069, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026