Pith. sign in

REVIEW 31 references

Feedback Stabilization of Polynomial Systems: From Model-based to Data-driven Methods

T0 review · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read New sum-of-squares conditions allow globally stabilizing controllers for structured polynomial systems, including a data-driven version that is robust to bounded noise and prior parameter knowledge.

desk verdict A competent, incremental paper: the model-based relaxation is clean, the data-driven extension is real, and the reader's two technical objections don't hold up on close reading—the actual caveat is the a priori noise bound. read the letter →

arxiv 2505.14457 v1 pith:UZKJTOLY submitted 2025-05-20 math.OC

classification math.OC
keywords approachdata-drivenfunctiongloballyapunovmodel-basedpolynomialstabilization
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

Many physical systems, like a mass on a spring, have states that are not all directly pushed by the input. This paper studies polynomial versions of such systems. A classic tool is the Lyapunov function, a sort of energy-like measure. If you can find one whose value always decreases along the system's motion, you can prove the system settles at zero. Earlier methods for polynomial systems only gave global guarantees when this function grew without bound at infinity. The authors replace that requirement with a milder one: outside a large ball, the Lyapunov function must decrease at a rate bounded away from zero. That opens the door to a wider class of Lyapunov functions.

The authors then move to the harder case where the equations are unknown. They assume the unknown parts enter linearly, and the polynomial building blocks are known. From a handful of noisy samples of state, input, and derivative, they build a set of systems that are all compatible with the data. Their controller is designed to stabilize every system in that set, not just the true one. The price is a condition called an S-lemma inequality, which they turn into a sum-of-squares optimization problem. Knowledge that certain parameters are known, such as rows of the state matrix, can be plugged in easily and makes the conditions less conservative.

Three numerical examples, including a two-mass hardening spring, show controllers and Lyapunov functions that pass the tests. The paper does not provide code, and one of the SOS conditions is bilinear in the decision variables, so the authors do not explain exactly how their solver found the reported solutions.

Extended reading notes

Core claim

Theorem 1: if conditions (a)-(d) hold, the controller K(x)=L(x)P^{-1}(x1)Z(x) renders the origin of system (1) globally asymptotically stable, without requiring V to be radially unbounded. Theorem 2 extends the same guarantee to every system in the data-compatible set Sigma under bounded noise, so the data are informative for stabilization.

Load-bearing premise

The data-driven theorem assumes the noise energy bound Phi_11 in (19)/(20) is known a priori and that the true system satisfies it. If actual disturbances exceed this bound, the true system is not in the compatible set Sigma and the synthesized controller has no formal guarantee. This boundary enters at Section III-A, Eq. (19).

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.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The model-based theorem assumes polynomial representation F=HZ, positive-definite P, and a certificate Z with a unique zero. The data-driven theorem adds a known energy bound on noise and a full-rank data matrix. The only nonstandard external input is the specialized S-lemma from the authors' companion paper [23], which is not proved here.

free parameters (2)
  • epsilon_1, epsilon_2, epsilon_3, c, r = Examples: epsilon_1=0.1, epsilon_2=0.01, epsilon_3=1+x1^2 or 2+2x1^2
    Hand-chosen tuning functions and constants in each example to satisfy conditions (a)-(b) and make the SOS problem feasible; no systematic selection rule is given.
  • Maximum degrees of P and L = 4 and 6 in Examples 2-4
    Chosen per example; the feasibility and size of the SOS program depend on these degrees.
assumptions (6)
  • domain assumption Z(x)=0 if and only if x=0 (Assumption 1)
    Needed to make V positive definite and to preserve the equilibrium; stated in Section II-A and used in Theorems 1-2.
  • domain assumption There exists a polynomial matrix H with F(x)=H(x)Z(x)
    The entire V and M construction in Eq. (6) depends on this factorization; H is chosen after Z in the examples.
  • domain assumption The matrix [F^T U^T G^T] has full column rank (Assumption 2)
    Guarantees D has full row rank and N22<0, which is required for the specialized S-lemma; stated in Section III-B.
  • domain assumption Noise is bounded by the energy inequality (19)/(20) with known Phi_11
    All data-compatibility and robustness statements depend on this model of unknown noise; entered in Section III-A.
  • ad hoc to paper Specialized S-lemma (Lemma 2) from the authors' companion paper [23]
    A specialized variant of the S-lemma cited to [23, Lem. 7]; proof not included in this paper and it is key to converting universal quantification over compatible systems into an SOS condition.
  • standard math Khalil Theorem 4.1 local asymptotic stability criterion
    Used in the proof of Lemma 1 to obtain a local basin of attraction from positive definite V and negative definite derivative.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Feedback Stabilization of Polynomial Systems: From Model-based to Data-driven Methods." pith.science (2026). https://pith.science/paper/UZKJTOLY

@misc{pith2026250514457,
  author       = {Pith},
  title        = {Pith review of: Feedback Stabilization of Polynomial Systems: From Model-based to Data-driven Methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UZKJTOLY}},
  note         = {Machine review of arXiv:2505.14457}
}
read the original abstract

In this study, we propose new global stabilization approaches for a class of polynomial systems in both model-based and data-driven settings. The existing model-based approach guarantees global asymptotic stability of the closed-loop system only when the Lyapunov function is radially unbounded, which limits its applicability. To overcome this limitation, we develop a new global stabilization approach that allows a broader class of Lyapunov function candidates. Furthermore, we extend this approach to the data-driven setting, considering Lyapunov function candidates with the same functional structure. Using data corrupted by bounded noise, we derive conditions for constructing globally stabilizing controllers for unknown polynomial systems. Beyond handling noise, the proposed data-driven approach can be readily adapted to incorporate further prior knowledge of system parameters to reduce conservatism. In both approaches, sum-of-squares relaxation is used to ensure computational tractability of the involved conditions.

Figures

Figures reproduced from arXiv: 2505.14457 by the authors.

Figure 1
Figure 1. Phase portrait of the closed-loop system and the level sets of V (x) in Example 1. The gray arrows depict the closed-loop vector field and the black lines are trajectories starting from the edges and converging to the origin. The level sets are indicated by red dashed lines. A. Problem formulation Given F and G, we denote the system of the form (1) by (A1, A2, B2). In this section, we consider a true system (As1, As… view at source ↗
Figure 2
Figure 2. Phase portrait of the closed-loop system and the level sets of V (x) in Example 2 (see (a)) and Example 3 (see (b)). The gray arrows depict the closed-loop vector field and the black lines are trajectories starting from the edges and converging to the origin. The level sets are indicated by red dashed lines. where J ∈ R 3 [x] is arbitrary. It can be verified that for the true system (34), there exist L ∈ R 1×2 [x] a… view at source ↗
Figure 3
Figure 3. The mass-spring system with a hardening spring. After taking F(ζ) = ζ2 ζ3 ζ4 ζ 3 2 ⊤ and G(ζ) = 1, we have As1 =   0 1 0 0 0 1 −1 0 −0.5 −0.05 0 −0.5   , As2 = [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Trajectories of the true closed-loop system for four different initial states. The zero state in each subplot is depicted by a red line. where P24(ζ2) = P42(ζ2) = 0.4522 + 0.1062ζ 2 2 , P34(ζ2) = P43(ζ2) = 0.4036 − 0.0540ζ 2 2 , P44(ζ2) = 1.3096 + 0.2286ζ 4 2 , L4(ζ) =…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 30 canonical work pages

  1. [6]

    Nonlinear c ontrol synthesis by sum of squares optimization: A Lyapunov-based approach,

    S. Prajna, A. Papachristodoulou, and F. Wu, “Nonlinear c ontrol synthesis by sum of squares optimization: A Lyapunov-based approach, ” in Proc. 5th Asian Control Conf. , 2004, pp. 157–165

  2. [23]

    Data- driven stabilization of polynomial systems using density f unctions,

    H. Huang, M. K. Camlibel, R. Carloni, and H. J. van Waarde , “Data- driven stabilization of polynomial systems using density f unctions,” 2025, arXiv:2503.07092

  3. [1]

    Dissipative dynamical systems Part I: Ge neral theory,

    J. C. Willems, “Dissipative dynamical systems Part I: Ge neral theory,” Archive Rational Mech. Anal. , vol. 45, no. 5, pp. 321–351, 1972

  4. [2]

    A ‘universal’ construction of Artstein’s theorem on nonlinear stabilization,

    E. D. Sontag, “A ‘universal’ construction of Artstein’s theorem on nonlinear stabilization,” Syst. Control Lett. , vol. 13, no. 2, pp. 117–123, 1989

  5. [3]

    On characterizations of the inp ut-to-state stability property,

    E. D. Sontag and Y . Wang, “On characterizations of the inp ut-to-state stability property,” Syst. Control Lett. , vol. 24, no. 5, pp. 351–359, 1995

  6. [4]

    H. K. Khalil, Nonlinear systems, 3rd ed. Upper Saddle River, NJ, USA: Prentice Hall, 2002

  7. [5]

    Virtual ref erence feedback tuning: a direct method for the design of feedback c ontrollers,

    M. C. Campi, A. Lecchini, and S. M. Savaresi, “Virtual ref erence feedback tuning: a direct method for the design of feedback c ontrollers,” Automatica, vol. 38, no. 8, pp. 1337–1346, 2002

  8. [7]

    Data-driven model-free adaptive control for a class of MIMO nonlinear discrete-time systems,

    Z. S. Hou and S. T. Jin, “Data-driven model-free adaptive control for a class of MIMO nonlinear discrete-time systems,” IEEE Trans. Neural Netw., vol. 22, no. 12, pp. 2173–2188, 2011

Show all 31 references
  1. [8]

    Port-Hamiltonian s ystems theory: An introductory overview,

    A. J. van der Schaft and D. Jeltsema, “Port-Hamiltonian s ystems theory: An introductory overview,” F ound. Trends Syst. and Control, vol. 1, no. 2-3, pp. 173–378, 2014

  2. [9]

    Dat a-driven control of nonlinear systems: An on-line direct approach,

    M. Tanaskovic, L. Fagiano, C. Novara, and M. Morari, “Dat a-driven control of nonlinear systems: An on-line direct approach,” Automatica, vol. 75, pp. 1–10, 2017

  3. [10]

    Output regulation for nonl inear systems: An overview,

    C. I. Byrnes and A. Isidori, “Output regulation for nonl inear systems: An overview,” Int. J. Robust Nonlinear Control , vol. 10, no. 5, pp. 323–337, 2000

  4. [11]

    Global adaptive dynamic progr amming for continuous-time nonlinear systems,

    Y . Jiang and Z. P . Jiang, “Global adaptive dynamic progr amming for continuous-time nonlinear systems,” IEEE Trans. Autom. Control , vol. 60, no. 11, pp. 2917–2929, 2015

  5. [12]

    Isidori, Nonlinear control systems: an introduction

    A. Isidori, Nonlinear control systems: an introduction . Springer, 1985

  6. [13]

    On the passivity- based impedance control of flexible joint robots,

    C. Ott, A. Albu-Schaffer, A. Kugi, and G. Hirzinger, “On the passivity- based impedance control of flexible joint robots,” IEEE Trans. Robot. , vol. 24, no. 2, pp. 416–429, 2008

  7. [14]

    Nonlinear con trol synthesis by convex optimization,

    S. Prajna, P . A. Parrilo, and A. Rantzer, “Nonlinear con trol synthesis by convex optimization,” IEEE Trans. Autom. Control , vol. 49, no. 2, pp. 310–314, 2004

  8. [15]

    A dual to Lyapunov’s stability theorem,

    A. Rantzer, “A dual to Lyapunov’s stability theorem,” Syst. Control Lett., vol. 42, no. 3, pp. 161–168, 2001

  9. [16]

    Semidefinite programming relaxations f or semialgebraic problems,

    P . A. Parrilo, “Semidefinite programming relaxations f or semialgebraic problems,” Mathematical programming, vol. 96, pp. 293–320, 2003

  10. [17]

    Data informativity: a new perspective on data-driven analysis a nd control,

    H. J. van Waarde, J. Eising, H. L. Trentelman, and M. K. Ca mlibel, “Data informativity: a new perspective on data-driven analysis a nd control,” IEEE Trans. Automa. Control , vol. 65, no. 11, pp. 4753–4768, 2020

  11. [18]

    Model-free control,

    M. Fliess and C. Join, “Model-free control,” Int. J. Control , vol. 86, no. 12, pp. 2228–2252, 2013

  12. [19]

    Data-driven stabiliz ation of nonlinear polynomial systems with noisy data,

    M. Guo, C. De Persis, and P . Tesi, “Data-driven stabiliz ation of nonlinear polynomial systems with noisy data,” IEEE Trans. Autom. Control , vol. 67, no. 8, pp. 4210–4217, 2021

  13. [20]

    A semi-algebraic optimization a pproach to data- driven control of continuous-time nonlinear systems,

    T. Dai and M. Sznaier, “A semi-algebraic optimization a pproach to data- driven control of continuous-time nonlinear systems,” IEEE Control Syst. Lett., vol. 5, no. 2, pp. 487–492, 2020

  14. [21]

    A note on persistency of excitation,

    J. C. Willems, P . Rapisarda, I. Markovsky, and B. L. M. De Moor, “A note on persistency of excitation,” Syst. Control Lett. , vol. 54, no. 4, pp. 325–329, 2005

  15. [22]

    Formulas for data-driven cont rol: Stabilization, optimality, and robustness,

    C. De Persis and P . Tesi, “Formulas for data-driven cont rol: Stabilization, optimality, and robustness,” IEEE Trans. Autom. Control , vol. 65, no. 3, pp. 909–924, 2020

  16. [24]

    Z. W. Jarvis-Wloszek, Lyapunov based analysis and controller synthesis for polynomial systems using sum-of-squares optimization . University of California, Berkeley, 2003

  17. [25]

    Control desi gn along trajectories with sums of squares programming,

    A. Majumdar, A. A. Ahmadi, and R. Tedrake, “Control desi gn along trajectories with sums of squares programming,” in Proc. IEEE Int. Conf. Robot. Autom. , 2013, pp. 4054–4061

  18. [26]

    The informativity approach: To data-driven analysis and c ontrol,

    H. J. V an Waarde, J. Eising, M. K. Camlibel, and H. L. Tren telman, “The informativity approach: To data-driven analysis and c ontrol,” IEEE Control Systems Magazine , vol. 43, no. 6, pp. 32–66, 2023

  19. [27]

    Quadratic matrix inequalities with applications to data- based control,

    H. J. van Waarde, M. K. Camlibel, J. Eising, and H. L. Tren telman, “Quadratic matrix inequalities with applications to data- based control,” SIAM J. Control Optim. , vol. 61, no. 4, pp. 2251–2281, 2023

  20. [28]

    A survey of the S-lemma,

    I. P´ olik and T. Terlaky, “A survey of the S-lemma,” SIAM review, vol. 49, no. 3, pp. 371–418, 2007

  21. [29]

    Y almip: A toolbox for modeling and optimi zation in matlab,

    J. L ¨ ofberg, “Y almip: A toolbox for modeling and optimi zation in matlab,” in Proc. IEEE Int. Conf. Robot. Autom. , 2004, pp. 284–289

  22. [30]

    Pre- and post-processing sum-of-squares program s in practice,

    ——, “Pre- and post-processing sum-of-squares program s in practice,” IEEE Trans. Autom. Control , vol. 54, no. 5, pp. 1007–1011, 2009

  23. [31]

    V ersion 10.0.30., 2022

    MOSEK ApS, The MOSEK optimization toolbox for MATLAB manual. V ersion 10.0.30., 2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.