pith. sign in

arxiv: 2604.06980 · v1 · submitted 2026-04-08 · 📡 eess.SY · cs.SY

Stochastic Adaptive Control for Systems with Nonlinear Parameterization: Almost Sure Stability and Tracking

Lantian Zhang , Bo Wahlberg , Silun Zhang This is my paper

Pith reviewed 2026-05-10 18:30 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords adaptive controlnonlinear parameterizationstochastic systemsweighted least squaresalmost sure stabilitytrackingstrong consistencynonlinear systems
0
0 comments X

The pith

An online nonlinear weighted least-squares estimator with attenuating excitation achieves almost sure global stability and average tracking for nonlinearly parameterized stochastic systems.

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

This paper develops an adaptive control method for stochastic systems whose dynamics depend nonlinearly on unknown parameters. It introduces a nonlinear weighted least-squares algorithm that estimates the parameters consistently from data without requiring persistent excitation. The control law then incorporates an attenuating excitation signal to ensure the estimates improve while maintaining stability. If the claims hold, such systems can be controlled reliably in applications like neural networks and social dynamics where excitation is costly or impractical. The result is global stability and long-run average tracking of reference trajectories.

Core claim

The resulting closed-loop system is globally stable and the system trajectory can track, in a long-run average sense, the reference trajectory generated with the true system parameters. This is achieved via an online nonlinear weighted least-squares parameter estimation algorithm that attains global strong consistency without persistent excitation conditions, combined with an adaptive control algorithm using an attenuating excitation signal.

What carries the argument

The nonlinear weighted least-squares (WLS) estimator for parameter consistency and the attenuating excitation signal in the feedback control law.

Load-bearing premise

The nonlinear weighted least-squares estimator achieves global strong consistency without persistent excitation conditions on the trajectory data.

What would settle it

A simulation or example where the parameter estimates fail to converge strongly to the true values despite the attenuating excitation, or where the closed-loop system diverges, would falsify the stability and tracking claims.

Figures

Figures reproduced from arXiv: 2604.06980 by Bo Wahlberg, Lantian Zhang, Silun Zhang.

Figure 1
Figure 1. Figure 1: Topology of the four agents. Yellow and green rep [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Trajectories of the parameter estimation error for dif [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
read the original abstract

This paper concerns the adaptive control problem for a class of nonlinear stochastic systems in which the state update is given by a nonlinear function of linear dynamics plus additive stochastic noise. Such systems arise in a wide range of applications, including recurrent neural networks, social dynamics, and signal processing. Despite their importance, adaptive control for these systems remains relatively unexplored in the literature. This gap is primarily due to the inherently nonconvex dependence of the system dynamics on unknown parameters, which significantly complicates both controller design and analysis. To address these challenges, we propose an online nonlinear weighted least-squares (WLS)-based parameter estimation algorithm and establish the global strong consistency of the resulting parameter estimates. In contrast to most existing results, our consistency analysis does not rely on restrictive assumptions such as persistent excitation conditions of the trajectory data, making it applicable to stochastic adaptive control settings. Building on the proposed estimator, we further develop an adaptive control algorithm with an attenuating excitation signal that can effectively combine adaptive estimation and feedback control. Finally, we are able to show that the resulting closed-loop system is globally stable and that the system trajectory can track, in a long-run average sense, the reference trajectory generated with the true system parameters. The proposed methods and theoretical results are finally validated through simulations in two nonlinear interaction network applications.

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 paper addresses adaptive control for stochastic nonlinearly parameterized systems of the form x_{t+1} = f(A(θ)x_t + B u_t) + w_t. It proposes an online nonlinear weighted least-squares (WLS) estimator, proves its global strong consistency without persistent excitation (PE) conditions on the data, and combines it with a certainty-equivalence controller that injects an attenuating excitation signal. The main results are almost-sure global stability of the closed loop and long-run average tracking of the reference trajectory generated by the true parameters. The claims are supported by theoretical analysis and simulations on nonlinear interaction networks.

Significance. If the global strong consistency of the nonlinear WLS estimator holds without PE and without hidden dependence on the control signal, the result would be significant for adaptive control of nonconvex parameterizations arising in recurrent neural networks, social dynamics, and signal processing. The almost-sure stability and average tracking claims are strong; the avoidance of PE is a notable technical contribution relative to standard stochastic adaptive control literature.

major comments (2)
  1. [§3] §3 (Estimator Consistency): The global strong consistency of the nonlinear WLS estimator (Theorem 1) is the load-bearing step for all subsequent stability and tracking claims. The argument appears to rely on the weighting matrix W and the specific attenuation schedule, but it is not clear from the derivation whether injectivity or strong convexity conditions on f are explicitly assumed or derived, nor how the non-convex loss is shown to have a unique global minimizer almost surely without PE on the regressors. A concrete counter-example or growth condition on f that could violate consistency should be ruled out.
  2. [§4] §4 (Closed-Loop Stability): The proof that the certainty-equivalence controller with attenuating excitation yields global stability and average tracking (Theorem 2) uses the estimator consistency result. However, the attenuating excitation is introduced for control purposes; it is unclear whether the consistency analysis in §3 is fully independent of this signal or whether the boundedness of states (needed for the regressor richness) is established without circularity. The long-run average tracking claim should be accompanied by an explicit rate or bound that does not degrade with the attenuation schedule.
minor comments (2)
  1. [§2] Notation for the nonlinear function f and the weighting matrix W should be introduced with explicit assumptions on their properties (e.g., Lipschitz constants or boundedness) at the first appearance.
  2. [§5] The simulation section would benefit from reporting the specific attenuation rate used and sensitivity of the tracking error to that choice.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful and constructive review of our manuscript. The comments raise important points about the clarity of the consistency proof and the structure of the closed-loop analysis. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [§3] §3 (Estimator Consistency): The global strong consistency of the nonlinear WLS estimator (Theorem 1) is the load-bearing step for all subsequent stability and tracking claims. The argument appears to rely on the weighting matrix W and the specific attenuation schedule, but it is not clear from the derivation whether injectivity or strong convexity conditions on f are explicitly assumed or derived, nor how the non-convex loss is shown to have a unique global minimizer almost surely without PE on the regressors. A concrete counter-example or growth condition on f that could violate consistency should be ruled out.

    Authors: We appreciate this observation. The proof of Theorem 1 establishes global strong consistency by showing that the weighted nonlinear least-squares criterion converges almost surely to a function whose unique minimizer is the true parameter θ*. This relies on the specific structure of the system (the composition of the linear-in-parameter map with the known nonlinearity f) together with the positive-definiteness of the weighting matrix W and the chosen attenuation schedule. Under the standing assumptions (continuous differentiability of f and the linear growth condition stated in Assumption 1), the map θ ↦ f(A(θ)x + Bu) is injective for each fixed (x,u) in the relevant domain; this injectivity is derived rather than postulated separately and is used to rule out other candidate minimizers. The non-convexity of the loss is handled by a direct argument that exploits the strong law of large numbers applied to the martingale difference noise terms; persistent excitation of the regressors is not required because the additive noise supplies the necessary richness in probability. To improve clarity we will (i) explicitly restate the derived injectivity property in the statement of Theorem 1 and (ii) add a short remark that excludes potential counter-examples by invoking the linear growth bound on f. No additional strong-convexity assumption is needed. revision: partial

  2. Referee: [§4] §4 (Closed-Loop Stability): The proof that the certainty-equivalence controller with attenuating excitation yields global stability and average tracking (Theorem 2) uses the estimator consistency result. However, the attenuating excitation is introduced for control purposes; it is unclear whether the consistency analysis in §3 is fully independent of this signal or whether the boundedness of states (needed for the regressor richness) is established without circularity. The long-run average tracking claim should be accompanied by an explicit rate or bound that does not degrade with the attenuation schedule.

    Authors: The consistency analysis of §3 is formulated for any input sequence that keeps the state process in a set of finite moments; it does not depend on the particular form of the control law. In the closed-loop proof we first establish, via a separate Lyapunov argument that accounts for the vanishing excitation, that the state remains bounded almost surely. Only after this boundedness step do we invoke the consistency result. We acknowledge that the current ordering of arguments may give the impression of circularity and will therefore restructure the proof of Theorem 2 to present the boundedness step first, followed by consistency and then the tracking conclusion. The attenuating excitation is constructed so that its contribution to the regressor vanishes at a rate compatible with the weighting schedule, preserving the almost-sure convergence. For the long-run average tracking we prove that the Cesàro mean of the tracking error converges to zero almost surely. An explicit non-asymptotic rate is not derived because any such rate would necessarily depend on the particular attenuation schedule chosen to guarantee stability; we will add a remark clarifying this dependence and noting that the average convergence itself remains independent of the schedule. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained

full rationale

The paper defines a nonlinear WLS estimator, claims global strong consistency without PE (via direct analysis of the loss and noise model), then augments the certainty-equivalence controller with an independent attenuating excitation term. The closed-loop stability and average tracking results are derived from the estimator's convergence property plus standard stochastic Lyapunov arguments. No equation or step reduces the consistency claim to the control law, nor renames a fitted quantity as a prediction, nor relies on a self-citation chain for the uniqueness or injectivity conditions. The weighting and attenuation schedules are explicit design choices whose effect on the proof is stated separately from the estimator definition itself.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claims rest on the system being expressible as a nonlinear map of linear dynamics plus additive noise, on standard stochastic convergence tools, and on the existence of a suitable attenuating excitation schedule; no new entities are postulated.

free parameters (2)
  • weighting matrix in WLS
    Chosen to handle the nonlinear parameterization; its specific form is a design parameter whose selection affects consistency.
  • attenuation rate of excitation signal
    Decaying schedule is introduced to balance estimation and control; its rate is a free design choice.
axioms (2)
  • domain assumption The nonlinear function satisfies conditions that allow the WLS estimator to be strongly consistent without persistent excitation.
    Invoked to drop the usual PE requirement; location is the consistency analysis paragraph.
  • standard math Additive stochastic noise is such that standard martingale or ergodic theorems apply to the closed-loop trajectory.
    Used for almost-sure stability and average tracking.

pith-pipeline@v0.9.0 · 5532 in / 1471 out tokens · 26059 ms · 2026-05-10T18:30:22.412768+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

49 extracted references · 49 canonical work pages

  1. [1]

    Bibitem after note You are using a bibitem after a note in a subbibitems environment; note should the last item in a subbibitems environment \@itemnote @bb@error [] @noitemnote \@tempa \@noitemnote \@noitemnote \@itemnote @bibitem\@bibitem @lbibitem\@lbibitem \@bibitem#1 \@itemnote @bibitem #1 \@tempa @noitemnote \@lbibitem[#1]#2 \@itemnote @lbibitem[#1] ...

    work page

  2. [2]

    @stdbsttrue @ctr \@lbibitem [ @ctr] \@bibitem##1 @ctr \@lbibitem[ @ctr] ##1 @bb@error\@mkbberr @filesw @natbibloaded \@auxout \@itemslabel @bibnum a-- @ctr \@auxout \@itemslabel :s @bibnum \@auxout \@itemslabel @bibnum a-- @ctr \@auxout \@itemslabel :s @bibnum @ctr @bibnum @citex\@citex \@tempcntc @citex[#1]#2 @filesw \@auxout #2 \@tempcnta @\@tempcntb @n...

    work page 1991

  3. [3]

    Elman, J. L. (1990). Finding structure in time. Cognitive Science , 14(2), 179--211

    work page 1990

  4. [4]

    D., Swart, T

    Mienye, I. D., Swart, T. G., & Obaido, G. (2024). A comprehensive review of deep learning: Architectures, recent advances, and applications. Information , 15(12), 755

    work page 2024

  5. [5]

    M., & Zimmermann, H

    Sch\"afer, A. M., & Zimmermann, H. G. (2006). Recurrent neural networks are universal approximators, In International Conference on Artificial Neural Networks , Berlin, Heidelberg: Springer Berlin Heidelberg, 632--640

    work page 2006

  6. [6]

    D., Pichler, E

    Keeler, J. D., Pichler, E. E., & Ross, J. (1989). Noise in neural networks: Thresholds, hysteresis, and neuromodulation of signal-to-noise. Proceedings of the National Academy of Sciences , 86(5), 1712--1716

    work page 1989

  7. [7]

    Wu, J. (2001). Introduction to Neural Dynamics and Signal Transmission Delay . Walter de Gruyter

    work page 2001

  8. [8]

    N., & Ioannides, Y

    Durlauf, S. N., & Ioannides, Y. M. (2010). Social interactions. Annual Review of Economics , 2(1), 451--478

    work page 2010

  9. [9]

    Xing, Y., He, X., Fang, H., & Johansson, K. H. (2022). Recursive network estimation for a model with binary-valued states. IEEE Transactions on Automatic Control , 68(7), 3872--3887

    work page 2022

  10. [10]

    M., Mazo, J

    Ebert, P. M., Mazo, J. E., & Taylor, M. G. (1969). Overflow oscillations in digital filters. Bell System Technical Journal , 48(9), 2999--3020

    work page 1969

  11. [11]

    Ooba, T. (2003). Stability of linear discrete dynamics employing state saturation arithmetic. IEEE Transactions on Automatic Control , 48(4), 626--630

    work page 2003

  12. [12]

    Bizyaeva, A., Franci, A., & Leonard, N. E. (2022). Nonlinear opinion dynamics with tunable sensitivity. IEEE Transactions on Automatic Control , 68(3), 1415--1430

    work page 2022

  13. [13]

    Su, W., Chen, X., Yu, Y., & Chen, G. (2021). Noise-based control of opinion dynamics. IEEE Transactions on Automatic Control , 67(6), 3134--3140

    work page 2021

  14. [14]

    D., & Iudice, F

    Ancona, C., Lellis, P. D., & Iudice, F. L. (2023). Influencing opinions in a nonlinear pinning control model. IEEE Control Systems Letters , 7, 1945--1950

    work page 2023

  15. [15]

    Schoukens, J., & Ljung, L. (2019). Nonlinear system identification: A user-oriented road map. IEEE Control Systems Magazine , 39(6), 28--99

    work page 2019

  16. [16]

    Chen, H., & Guo, L. (1991). Identification and Stochastic Adaptive Control . Boston, MA: Birkhäuser

    work page 1991

  17. [17]

    Wahlberg, B., Welsh, J., & Ljung, L. (2015). Identification of stochastic Wiener systems using indirect inference. IFAC-PapersOnLine , 48(28), 620--625

    work page 2015

  18. [18]

    Guo, L. (1996). Self-convergence of weighted least squares with applications to stochastic adaptive control. IEEE Transactions on Automatic Control , 41(1), 79--89

    work page 1996

  19. [19]

    Foster, D., Sarkar, T., & Rakhlin, A. (2020). Learning nonlinear dynamical systems from a single trajectory. In Learning for Dynamics and Control , 851--861

    work page 2020

  20. [20]

    Oymak, S. (2019). Stochastic gradient descent learns state equations with nonlinear activations. In Conference on Learning Theory , 2551--2579

    work page 2019

  21. [21]

    Kowshik, S., Nagaraj, D., Jain, P., & Netrapalli, P. (2021). Near-optimal offline and streaming algorithms for learning non-linear dynamical systems. Advances in Neural Information Processing Systems , 34, 8518--8531

    work page 2021

  22. [22]

    Sattar, Y., & Oymak, S. (2022). Non-asymptotic and accurate learning of nonlinear dynamical systems. Journal of Machine Learning Research , 23(140), 1--49

    work page 2022

  23. [23]

    J., & Wittenmark, B

    Åström, K. J., & Wittenmark, B. (1973). On self tuning regulators. Automatica , 9(2), 185--199

    work page 1973

  24. [24]

    L., & Wei, C

    Lai, T. L., & Wei, C. Z. (1982). Least squares estimates in stochastic regression models with applications to identification and control of dynamic systems. The Annals of Statistics , 10(1), 154--166

    work page 1982

  25. [25]

    C., Ramadge, P

    Goodwin, G. C., Ramadge, P. J., & Caines, P. E. (1981). Discrete time stochastic adaptive control. SIAM Journal on Control and Optimization , 19(6), 829--853

    work page 1981

  26. [26]

    Kumar, P. R. (1990). Convergence of adaptive control schemes using least squares parameter estimates. IEEE Transactions on Automatic Control , 35(4), 416--424

    work page 1990

  27. [27]

    Guo, L. (1995). Convergence and logarithm laws of self-tuning regulators. Automatica , 31(3), 435--450

    work page 1995

  28. [28]

    Liu, N., Zhao, C., Tan, S., & Lü, J. (2025). Optimal adaptive control of linear stochastic systems with quadratic cost function. IEEE Transactions on Automatic Control . 70(10), 7024 - 7031

    work page 2025

  29. [29]

    Dörfler, F., Tesi, P., & De Persis, C. (2023). On the certainty-equivalence approach to direct data-driven LQR design. IEEE Transactions on Automatic Control , 68(12), 7989--7996

    work page 2023

  30. [30]

    Chiuso, A., Fabris, M., Breschi, V., & Formentin, S. (2025). Harnessing uncertainty for a separation principle in direct data-driven predictive control. Automatica , 173, 112070

    work page 2025

  31. [31]

    Bian, T., Jiang, Y., & Jiang, Z. P. (2016). Adaptive dynamic programming for stochastic systems with state and control dependent noise. IEEE Transactions on Automatic Control , 61(12), 4170--4175

    work page 2016

  32. [32]

    Guo, L. (1997). On critical stability of discrete-time adaptive nonlinear control. IEEE Transactions on Automatic Control , 42(11), 1488--1499

    work page 1997

  33. [33]

    Bercu, B., & Portier, B. (2002). Adaptive control of parametric nonlinear autoregressive models via a new martingale approach. IEEE Transactions on Automatic Control, 47(9), 1524--1528

    work page 2002

  34. [34]

    Li, C., & Lam, J. (2013). Stabilization of discrete-time nonlinear uncertain systems by feedback based on LS algorithm. SIAM Journal on Control and Optimization, 51(2), 1128--1151

    work page 2013

  35. [35]

    Li, W., & Krstic, M. (2020). Stochastic adaptive nonlinear control with filterless least squares. IEEE Transactions on Automatic Control , 66(9), 3893--3905

    work page 2020

  36. [36]

    Liu, Z., & Li, C. (2025). Optimal adaptive output regulation of discrete-time nonlinear stochastic systems. SIAM Journal on Control and Optimization , 63(4), 2369--2396

    work page 2025

  37. [37]

    Ma, H. (2007). Finite-model adaptive control using WLS-like algorithm. Automatica , 43(4), 677--684

    work page 2007

  38. [38]

    Li, C., & Chen, M. Z. Q. (2015). Simultaneous identification and stabilization of nonlinearly parameterized discrete-time systems by nonlinear least squares algorithm. IEEE Transactions on Automatic Control , 61(7), 1810--1823

    work page 2015

  39. [39]

    Zhang, L., Zhao, Y., & Guo, L. (2022). Identification and adaptation with binary-valued observations under non-persistent excitation. Automatica , 138, 110158

    work page 2022

  40. [40]

    Zhang, L., & Guo, L. (2025). Adaptive tracking control with binary-valued output observations. IEEE Transactions on Automatic Control . DOI: 10.1109/TAC.2025.3631410

    work page doi:10.1109/tac.2025.3631410 2025

  41. [41]

    Zhang, L., & Zhang, S. (2025). Online learning for nonlinear dynamical systems without the IID condition. arXiv preprint arXiv:2504.02995

    work page arXiv 2025

  42. [42]

    Zhao, Y., Guo, J., & Zhang, J. (2013). Adaptive tracking control of linear systems with binary-valued observations and periodic target. IEEE Transactions on Automatic Control, 58(5), 1293--1298

    work page 2013

  43. [43]

    Zhao, W., Weyer, E., Yin, G., Dong, D., Zhang, Y., & Shen, T. (2023). Adaptive regulation of block-oriented nonlinear systems using binary sensors with applications to automotive engine control. IEEE Transactions on Automatic Control, 68(3), 1369--1382

    work page 2023

  44. [44]

    E., Guo, L., & Pasik-Duncan, B

    Duncan, T. E., Guo, L., & Pasik-Duncan, B. (1999). Adaptive continuous-time linear quadratic Gaussian control. IEEE Transactions on Automatic Control, 44(9), 1653--1662

    work page 1999

  45. [45]

    Hager, W. W. (1989). Updating the inverse of a matrix. SIAM Review, 31(2), 221--239

    work page 1989

  46. [46]

    Van der Schaft, A. (2000). L_2 -gain and Passivity Techniques in Nonlinear Control . Berlin, Heidelberg: Springer Berlin Heidelberg

    work page 2000

  47. [47]

    Bloem, M., Alpcan, T., & Başar, T. (2009). Optimal and robust epidemic response for multiple networks. Control Engineering Practice , 17, 525--533

    work page 2009

  48. [48]

    (2025) Collective decision-making dynamics in hypernetworks

    Fontan, A., & Zhang, S. (2025) Collective decision-making dynamics in hypernetworks. IEEE 64th Conference on Decision and Control (CDC) , 6222--6227

    work page 2025

  49. [49]

    Hildebrandt, T. H. (1942). Remarks on the Abel-Dini theorem. The American Mathematical Monthly , 49(7), 441--445

    work page 1942