Pith. sign in

REVIEW 1 major objections 6 minor 1 cited by

Data-Driven Decentralized Control Design for Discrete-Time Large-Scale Systems

T0 review · 1 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Decentralized stabilizing controllers can be computed directly from measured state, control, and interconnection data, without identifying the system's model.

desk verdict A useful but rough extension of Persis–Tesi to interconnected systems; Theorem 4 needs an explicit symmetry/PD constraint on the Lyapunov matrix before it is correct as stated. read the letter →

arxiv 2411.10243 v1 pith:JEGNPBEO submitted 2024-11-15 eess.SY cs.SY

Jiaping Liao , Shuaizheng Lu , Tao Wang , Weiming Xiang This is my paper · ORCID
classification eess.SYcs.SY MSC 93A1493C5593D20
keywords data-drivencontroldecentralizedlarge-scalesystemsstabilizationlinearmatrixinequalitiesdiscrete-timepersistencyofexcitationinterconnected
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

This paper establishes that for a class of discrete-time large-scale systems composed of nonlinear interconnected subsystems, decentralized state-feedback controllers can be designed directly from measured data without first identifying the subsystem matrices. The key is an exact data-driven representation of the closed-loop system, valid when the recorded control and interconnection input sequences are persistently exciting. Under a sector-bound condition on the interconnection nonlinearities, the stabilizing gains are obtained by solving a semi-definite program (an LMI) per subsystem, with an auxiliary constraint that zeros out the interconnection-related data. The result is demonstrated on a chain of five spring-mass systems whose velocities converge to a target.

What carries the argument

The load-bearing mechanism is Theorem 3's exact data-driven representation of the closed loop, which converts the unknown matrices $\{A_i, B_i, G_i\}$ into the measured data block $X_{i,[1,T]}$ and two parameter matrices $H_{i,1}$, $H_{i,2}$. The matrix $H_{i,1}$ carries the controller gain through $K_i = U_{i,[0,T-1]} H_{i,1}$, while the constraint $\Phi_{i,[0,T-1]} Q_i = 0$ ensures that the row block of data corresponding to interconnection inputs is annihilated, so the effective feedback only uses local state. Stability is established through the Lyapunov matrix $S_i^{-1}$ and the LMI (23), which is feasible if there exist $Q_i$ making the data-driven closed-loop matrix contractive with respect to the sector bounds $W_{ij}$.

What would settle it

Take a two-subsystem instance of the spring-mass chain, deliberately set $W_{ij}$ to a value smaller than the true Lipschitz constant of $g_{ij}$, and record data while satisfying the rank condition. If the resulting LMI (23) is feasible but the closed-loop system under the computed $K_i$ fails to converge, that would show the stability certificate is invalid when the sector-bound knowledge is wrong. Alternatively, corrupt the measured interconnection inputs with small noise (e.g., $\|\eta(k)\| \approx 0.01\|\phi_i(k)\|$) and re-run the design; instability under the resulting gains would demonstrate the method's sensitivity to exact measurability.

Watch

Extended reading notes

Core claim

The central discovery is that the closed-loop behavior of each controlled subsystem can be written entirely in terms of data matrices: $x_i(k+1) = X_{i,[1,T]}(H_{i,1} x_i(k) + H_{i,2} \phi_i(k))$, where $H_{i,1}$ and $H_{i,2}$ are solutions of linear equations built from collected state, input, and interconnection input matrices. Because these solutions exist whenever the input data is persistently exciting of order $n_i+1$, the feedback gain becomes a parameter of the data representation rather than a function of an unknown model. Stability is enforced by a Lyapunov function $V = \sum_i x_i^\top S_i^{-1} x_i$ with $S_i = X_{i,[0,T-1]} Q_i$ and an LMI equivalent to $\sum_i (\Delta V_i + \hat{\Theta}_i) < 0$, where the $\hat{\Theta}_i$ terms account for the interconnection bounds. Theorem 4 therefore reduces decentralized stabilization to a data-only semidefinite program: find $Q_i$ satisfying (23) and $\Phi_{i,[0,T-1]} Q_i = 0$, and set $K_i = U_{i,[0,T-1]} Q_i S_i^{-1}$, which asymptotically stabilizes the large-scale system.

Load-bearing premise

The load-bearing assumption is that every interconnection input $g_{ij}(x_j)$ can be measured exactly and that the sector-bound matrices $W_{ij}$ are known; if the coupling signals between subsystems cannot be measured or bounded, the data matrices and the LMI cannot be formed, and the whole design collapses.

Editorial extensions

If this is right

  • If the LMI (23) is feasible and the data satisfy the rank condition, the gains $K_i = U_{i,[0,T-1]} Q_i S_i^{-1}$ guarantee asymptotic stability of the interconnected closed-loop system, without any model identification step.
  • The design is fully decentralized: each subsystem's gain is computed from its own local data and the interconnection input data, and the resulting controller uses only local state feedback.
  • The approach applies to any interconnection nonlinearity satisfying the sector inequality $\|g_{ij}(r)-g_{ij}(s)\| \le \|W_{ij}(r-s)\|$, with the $W_{ij}$ matrices entering the LMI; no other model information is needed.
  • The validation on a five-mass spring chain shows the method in action, with the velocities of all masses converging to the target in about four seconds.
  • Because the design is a semidefinite program per subsystem, the computational cost scales with subsystem size rather than the full large-scale system.

Reading between the lines

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

  • The paper requires exact measurement of interconnection inputs and known sector-bound matrices $W_{ij}$; if those are only estimated with error, the stability certificate degrades, and one could add robustness margins to the LMI to account for noise, though that extension is not made here.
  • The same data representation could be reused for other objectives, such as LQR-like performance or disturbance rejection, by replacing the Lyapunov inequality in (23) with the corresponding performance inequality while keeping the identification-free structure.
  • The persistence-of-excitation condition on interconnection inputs means that data collection experiments must actively excite the coupling channels; in applications where interconnections cannot be injected externally, one might rely on naturally occurring excitation or use closed-loop data, which would be a natural extension of the approach.
  • Since the LMI is solved per subsystem, the method is naturally parallelizable and could scale to very large networks, but the conservativeness incurred by bounding all interconnections with known $W_{ij}$ suggests that exploiting the actual sparse interconnection topology could reduce conservatism.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 6 minor

Summary. The paper proposes a data-driven decentralized state-feedback stabilization method for discrete-time large-scale systems with unknown subsystem dynamics and sector-bounded interconnections. Under a rank/persistency-of-excitation condition, Lemmas 1–2 and Theorem 3 show that the open-loop and closed-loop subsystem dynamics can be represented exactly using collected data matrices U_{i,[0,T-1]}, Φ_{i,[0,T-1]}, X_{i,[0,T-1]}, and X_{i,[1,T]}. The main result, Theorem 4, formulates an LMI in variables Q_i, with S_i = X_{i,[0,T-1]}Q_i and the constraint Φ_{i,[0,T-1]}Q_i = 0, and gives the decentralized gains K_i = U_{i,[0,T-1]}Q_i S_i^{-1}. Stability is proved via a sum of local Lyapunov functions and the sector bounds on interconnections. A simulation on a five-mass spring-chain illustrates the design.

Significance. If the central theorem is repaired, the paper makes a useful incremental contribution by extending the Persis–Tesi data-driven framework to decentralized control of interconnected large-scale systems. The data-driven representation in Lemma 2 and Theorem 3 is sound under the stated rank condition, and the SDP formulation is direct: the controller gains are parameterized by data matrices without an explicit model identification step. The authors openly acknowledge the main limitations (no measurement noise, no delays or uncertainties). The numerical example is simple, but the theory is the primary contribution. The paper is not circular: Q_i are decision variables of an LMI, not parameters fitted to reproduce a desired stability outcome.

major comments (1)
  1. [Theorem 4, Eqs. (23)–(26)] The variable S_i = X_{i,[0,T-1]} Q_i is not constrained to be symmetric or positive definite. For the block matrix in (23) to be a valid negative-definite LMI, the diagonal blocks -S_i must be symmetric, and negative definiteness would then imply S_i ≻ 0 and hence invertibility. However, the theorem only says 'there exist matrices Q_i'; a free Q_i will generally produce a nonsymmetric S_i, making the expression in (23) undefined as an LMI. This also affects the proof: the Lyapunov function V_i = x_i^T S_i^{-1} x_i in (26) and the congruence transformation with diag{S_i^{-1}, I, I, I} before (29) are justified only if S_i is symmetric positive definite. This is a load-bearing gap, because (23) is the only synthesis condition and (25) is undefined if S_i is not invertible. The authors should add the explicit constraints S_i = S_i^T ≻ 0 (e.g., declare S_i symmetric and impose the linear equality S_i = X_{i,[0,T-1]} Q_i), or introduce a symmetric slack variable in place of S_i.
minor comments (6)
  1. [Section 2.1, Eq. (3)] The definition of G_i repeats G_{i1}; it should read G_i = [G_{i1} G_{i2} ··· G_{iM}].
  2. [Appendix A, Eq. (A.2)] The term (I - Y†Y)w is dimensionally inconsistent; the general least-squares solution should be Ξ* = X Y† + Z(I - Y†Y) for a matrix Z of size n_i × T. The conclusion is unaffected because the projector vanishes under (11).
  3. [Theorem 4] The dimensions of the decision variables should be stated, e.g., Q_i ∈ R^{T × n_i}, and the matrix W_i should be defined in the theorem statement rather than only in the proof.
  4. [Remark 1] The estimator \bar w_{ij} = max ||g_{ij}(x_j)||/||x_j|| is undefined at x_j = 0; it should be stated as a supremum over x_j ≠ 0.
  5. [Section 4, Step (3)] The instructions say to form U_{i,[0,T-1]} again after it was formed in Step (2); rephrase to avoid duplication.
  6. [Assumption 1] Exact measurability of all interconnection inputs g_{ij}(x_j) is a strong requirement for large-scale systems; the paper would benefit from a discussion of how these signals are obtained in practice, particularly in the spring-mass example.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the data-driven synthesis uses standard external persistence-of-excitation results and an LMI sufficient condition; the S_i symmetry gap is a rigor issue, not a circular step.

full rationale

Every load-bearing derivation step is either self-contained or rests on external, cited results (Willems et al. 2005; Persis and Tesi 2020; Penrose 1956). The rank-condition data representation in Lemma 2 and Theorem 3 is the standard least-squares/persistence-of-excitation identity, not a re-labeling of the paper's own conclusion. Theorem 4's LMI (23) is a sufficient Lyapunov condition: the decision variables Q_i and controller gains K_i = U_i Q_i S_i^{-1} are chosen to satisfy (23)-(24), and the closed-loop terms in the Lyapunov difference (27)-(30) are obtained by Schur complement and congruence from (23). No parameter is fitted to reproduce a target stability outcome, and no result by the same authors is load-bearing; the self-citations in the introduction (Wang et al. 2020; Xiang et al. 2014) are contextual rather than used to justify the main derivation. The symmetry/positive-definiteness gap in S_i identified by the skeptic is a proof-rigor issue (the star notation requires S_i = S_i^T, and -S_i ≺ 0 would imply S_i ≻ 0 if that symmetry were stated), not a circular reduction. Therefore no circularity step is present.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central design depends on the data rank condition, the sector-bounded interconnection assumption, and the implicit symmetry of S_i. No new physical entities are introduced.

free parameters (1)
  • W_ij sector-bound matrices or scalars = e.g., \bar w_ij I with \bar w_ij = max ||g_ij(x_j)||/||x_j|| (Remark 1)
    The LMI (23) depends on W_i; in practice these are estimated from the same offline data, so they are data-fitted constants, not known system parameters.
assumptions (4)
  • domain assumption The interconnection inputs g_ij(x_j) are measurable and satisfy the sector bound ||g_ij(r)-g_ij(s)|| <= ||W_ij(r-s)|| with known W_ij.
    Assumption 1 and Remark 1; this is needed to form Φ_i data and to apply the S-procedure in the stability proof.
  • domain assumption The pairs (A_i, [B_i, G_i]) are controllable and the data matrix satisfies rank condition (11).
    Section 2.1 states controllability; the rank condition is required for the exact data-driven representation in Lemma 2 and Theorem 3.
  • standard math The fundamental lemma of Willems et al. (2005), i.e., persistency of excitation of order n_i+1 implies the data matrix spans the system behavior.
    Used in Lemma 1 and the proofs of Lemma 2 and Theorem 3; the paper cites Willems et al. and Persis and Tesi.
  • ad hoc to paper S_i = X_i Q_i is symmetric positive definite and invertible, which is implicit in the LMI but not stated.
    The diagonal block -S_i and the proof's use of S_i^{-1} require S_i symmetric; this is an unflagged assumption in Theorem 4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-Driven Decentralized Control Design for Discrete-Time Large-Scale Systems." pith.science (2026). https://pith.science/paper/JEGNPBEO

@misc{pith2026241110243,
  author       = {Pith},
  title        = {Pith review of: Data-Driven Decentralized Control Design for Discrete-Time Large-Scale Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JEGNPBEO}},
  note         = {Machine review of arXiv:2411.10243}
}
read the original abstract

In this paper, a data-driven approach is developed for controller design for a class of discrete-time large-scale systems, where a large-scale system can be expressed in an equivalent data-driven form and the decentralized controllers can be parameterized by the data collected from its subsystems, i.e., system state, control input, and interconnection input. Based on the developed data-driven method and the Lyapunov approach, a data-driven semi-definite programming problem is constructed to obtain decentralized stabilizing controllers. The proposed approach has been validated on a mass-spring chain model, with the significant advantage of avoiding extensive modeling processes.

Figures

Figures reproduced from arXiv: 2411.10243 by the authors.

Figure 1
Figure 1. State responses of five masses under decentral [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Dissipativity-Based Data-Driven Decentralized Control of Interconnected Systems

    eess.SY 2025-09 conditional novelty 6.0 of 10

    Data-driven decentralized control of interconnected discrete-time LTI systems is achieved by synthesizing local dissipative controllers and certifying global stability with LMIs from local data and noise bounds.

Reference graph

Works this paper leans on

21 extracted references · 17 canonical work pages · cited by 1 Pith paper

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sent...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sentence := ...

  4. [4]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize ":" * " " *...

  5. [5]

    Baggio, G., Katewa, V., and Pasqualetti, F. (2019). Data-driven minimum-energy controls for linear systems. IEEE Control Systems Letters, 3(3), 589--594

    work page 2019

  6. [6]

    and Tesi, P

    De Persis, C. and Tesi, P. (2021). Low-complexity learning of linear quadratic regulators from noisy data. Automatica, 128, 109548

    work page 2021

  7. [7]

    Guo, M., De Persis, C., and Tesi, P. (2021). Data-driven stabilization of nonlinear polynomial systems with noisy data. IEEE Transactions on Automatic Control, 67(8), 4210--4217

    work page 2021

  8. [8]

    and Liu, T

    Hu, K. and Liu, T. (2023). Data-driven output-feedback control for unknown switched linear systems. IEEE Control Systems Letters, 7, 2299--2304

    work page 2023

Show all 21 references
  1. [9]

    and Liu, T

    Jiang, Z.P. and Liu, T. (2018). Small-gain theory for stability and control of dynamical networks: A survey. Annual Reviews in Control, 46, 58--79

    2018

  2. [10]

    Kordestani, M., Safavi, A.A., and Saif, M. (2021). Recent survey of large-scale systems: Architectures, controller strategies, and industrial applications. IEEE Systems Journal, 15(4), 5440--5453

    2021

  3. [11]

    Penrose, R. (1956). On best approximate solutions of linear matrix equations. In Mathematical Proceedings of the Cambridge Philosophical Society, volume 52, 17--19. Cambridge University Press

    1956

  4. [12]

    and Tesi, P

    Persis, C.D. and Tesi, P. (2020). Formulas for data-driven control: Stabilization, optimality, and robustness. IEEE Transactions on Automatic Control, 65(3), 909--924

    2020

  5. [13]

    Qin, S.J. (2006). An overview of subspace identification. Computers & Chemical Engineering, 30(10), 1502--1513

    2006

  6. [14]

    Rajan, R., Fernandez, F.M., and Yang, Y. (2021). Primary frequency control techniques for large-scale pv-integrated power systems: A review. Renewable and Sustainable Energy Reviews, 144, 110998

    2021

  7. [15]

    Rotulo, M., De Persis, C., and Tesi, P. (2022). Online learning of data-driven controllers for unknown switched linear systems. Automatica, 145, 110519

    2022

  8. [16]

    Silva, G.R.G.d., Bazanella, A.S., Lorenzini, C., and Campestrini, L. (2019). Data-driven LQR control design. IEEE Control Systems Letters, 3(1), 180--185

    2019

  9. [17]

    Ugrinovskii, V.A., Petersen, I.R., Savkin, A.V., and Ugrinovskaya, E.Y. (2000). Decentralized state-feedback stabilization and robust control of uncertain large-scale systems with integrally constrained interconnections. Systems & Control Letters, 40(2), 107--119

    2000

  10. [18]

    van der Veen, G., van Wingerden, J.W., Bergamasco, M., Lovera, M., and Verhaegen, M. (2013). Closed-loop subspace identification methods: an overview. IET Control Theory & Applications, 7(10), 1339--1358

    2013

  11. [19]

    Wang, T., Wang, X., and Xiang, W. (2020). Reachable set estimation and decentralized control synthesis for a class of large-scale switched systems. ISA Transactions, 103, 75--85

    2020

  12. [20]

    Willems, J.C., Rapisarda, P., Markovsky, I., and De Moor, B.L.M. (2005). A note on persistency of excitation. Systems & Control Letters, 54(4), 325--329

    2005

  13. [21]

    Xiang, W., Xiao, J., and Han, L. (2014). Decentralized weighted control for a class of large-scale systems with multi-modes. International Journal of Robust and Nonlinear Control, 24(18), 3387--3408

    2014

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.