Pith. sign in

REVIEW 2 major objections 6 minor 42 references

For positive linear systems drawn from a known finite set, an explicit certainty-equivalence law stabilizes the plant and bounds the ℓ1-gain without any initially stabilizing controller.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-30 20:14 UTC pith:ZEGYI7AU

load-bearing objection Clean, usable extension of finite-set minimax adaptive control to positive systems with explicit ℓ1 certificates; sufficient conditions, not optimal dual control. the 2 major comments →

arxiv 2607.26816 v1 pith:ZEGYI7AU submitted 2026-07-29 math.OC

Minimax adaptive control for finite sets of positive linear systems

Fethi Bencherki , Anders Rantzer This is my paper
classification math.OC MSC 93C4093B3690C3993D21
keywords minimax adaptive controlpositive systemsgame-theoretic controldynamic programmingBellman inequalitydata-driven controlℓ1-gaincertainty equivalence
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper treats discrete-time positive linear systems whose matrices lie in a known finite list, with disturbances that may be negative so long as the next state stays nonnegative. It casts control as a game against an adversary that picks both the true plant and the disturbances, then rewrites the game in terms of next-state variables and one cumulative history score per candidate plant. An explicit piecewise-linear value function built from pairwise cost vectors solves the Bellman inequality, so the certainty-equivalence policy that always applies the gain of the currently best-fitting model is stabilizing, positivity-preserving, and certifies a finite cost and a finite ℓ1-gain from disturbances to stage cost. After enough data have singled out the true model, the same law collapses to the ordinary H∞-type optimal controller for that plant. The construction needs no a-priori stabilizing feedback and is illustrated on a two-reservoir network and a four-mode job-queueing network.

Core claim

When the unknown positive plant belongs to a finite set and inputs are constrained by a common bound |u|≤Ex, there exist nonnegative pairwise vectors p_ij and gains K_i such that the max of the linear functions p_ij^⊤x−½(z^{(i)}+z^{(j)}) solves the Bellman inequality of the reformulated game; the induced certainty-equivalence policy therefore stabilizes every plant in the set, preserves positivity, and guarantees J_μ(x_0)≤max p_ij^⊤x_0 together with a bounded ℓ1-gain.

What carries the argument

Model-specific history scores z^{(i)} that accumulate the γ-weighted ℓ1 prediction error of each candidate plant, together with the pairwise value functions V_ij=p_ij^⊤x−½(z^{(i)}+z^{(j)}) whose pointwise maximum solves the Bellman inequality and yields the certainty-equivalence law u_t=K_{k_t}x_t with k_t=arg min_i z^{(i)}_t.

Load-bearing premise

There must exist one common input-bound matrix and a family of nonnegative cost vectors, one for every pair of models, that simultaneously satisfy a list of linear inequalities; those inequalities are only sufficient, not necessary.

What would settle it

For a concrete finite set that is known to be simultaneously stabilizable by some positivity-preserving law, check whether Algorithm 1 (or any solver) can produce p_ij and γ satisfying all of the inequalities (5.7); if no such vectors exist, or if the resulting certainty-equivalence trajectory diverges on a positivity-preserving disturbance, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Stabilizing adaptive controllers for positive networks can be designed without first finding a single gain that works for every plant in the uncertainty set.
  • Once the history scores identify the true model, the closed loop automatically reverts to the model-based H∞-type optimum for that plant.
  • The same certificates supply an explicit upper bound on the induced ℓ1-gain from adversarial disturbances to the weighted stage cost.
  • The framework applies directly to reservoir networks and multiclass fluid queues whose mode or transfer direction is unknown but finite.

Where Pith is reading between the lines

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

  • Replacing the finite list by a continuum of plants will require history variables that no longer grow with the number of models, analogous to a single covariance matrix.
  • The same pairwise construction may extend, with only notational change, to the nonnegative-input class used for stochastic shortest-path problems on positive networks.
  • Tighter numerical procedures for the p_ij inequalities, borrowed from the quadratic minimax literature, would shrink the certified gain and make larger model sets practical.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper develops a minimax adaptive control framework for discrete-time positive linear systems whose unknown dynamics belong to a finite model set M. The controller and an adversary play a zero-sum game in which the adversary chooses both positivity-preserving disturbances and the active plant. By reparameterizing via next-state variables and model-specific cumulative history scalars z^{(A,B)}, the authors obtain an equivalent standard minimax dynamic game (Problem 2 / Theorem 1). They then give explicit max-of-linear solutions to the associated Bellman inequality: first for unknown input sign (Theorem 2, Corollary 1), then for a general finite set of plants (Theorem 3), yielding certainty-equivalence policies u_t = K_{k_t} x_t that preserve positivity, require no initially stabilizing controller, and certify a cost/ℓ1-gain bound. The model-based special case is recovered under a relaxed disturbance class (Appendix B). Two numerical examples (two-reservoir network; four-mode queueing network) and public code illustrate the policies.

Significance. The contribution is a clean, structured extension of minimax adaptive control to positive systems with linear stage costs and ℓ1-type robustness. Explicit pairwise certificates p_ij, the history-based reformulation tailored to the ℓ1 setting, and the relaxed positivity-preserving disturbance class (beyond nonnegative w) are genuine technical additions relative to the quadratic finite-set literature and to model-based positive H∞/minimax work. Strengths that should be credited include: (i) a stated DP equivalence (Theorem 1) with proofs in the appendix; (ii) closed-form sufficient Bellman solutions and a scalar CE gain formula; (iii) reproducible code linked in the examples. The results are certificates rather than optimal dual controllers, which the paper largely acknowledges and lists as future work; within that scope the paper is a solid and usable contribution for scalable positive-network control.

major comments (2)
  1. [Section 5.2, Example 3 / Figure 5] Example 3 (and Figure 5) introduces an abrupt true-mode switch at t=30 and presents the adaptive law as identifying the new mode. The problem formulation and Theorems 1–3 assume a single fixed unknown pair (A*,B*) ∈ M (Assumption 1); mode switching is outside the stated game. Either restrict the example to a fixed unknown mode, or add a precise caveat that switching is only empirical illustration beyond the theorems, and do not present it as validation of the certified bound.
  2. [Theorem 3, Algorithm 1, Remark 11; abstract / §1.1 (C4)] Theorem 3 and Algorithm 1 give only sufficient conditions for the Bellman inequality via (5.7). Remark 11 notes non-tightness, but the abstract and contribution (C4) read closer to a constructive solution of the adaptive problem. Please state prominently (abstract/intro and at Theorem 3) that feasibility of {p_ij, K_i, γ} is an assumption, that Algorithm 1 need not be optimal, and—if possible—give a short sufficient feasibility discussion (e.g., when each plant admits a model-based p_ii and γ can be enlarged to cover mixed terms even if some mismatched closed loops are unstable).
minor comments (6)
  1. [§1.3, §3] Notation Z_t for the collection of histories is overloaded with the common use of Z for integers; a different symbol (e.g., H_t or Z_t with a calligraphic face) would reduce friction.
  2. [§5.1 and passim] In §5.1, “Without a loss of optimality, Such inputs” has a capitalization/typo; several similar small grammar issues appear (e.g., “nonnegtaive”, “qunatity”). A careful copy-edit pass is needed.
  3. [Appendix B, Theorem 4] Appendix B omits the full model-based proof as “readily adapted from [23]”. A short self-contained verification of the key step under w ∈ W(x,u) (especially the max over signed positivity-preserving w) would make the claimed extension easier to check.
  4. [Figures 3 and 5] Figure 3/5 legends are dense; stating the true plant/mode and the numerical γ, p_ij used in each panel caption would help reproducibility alongside the GitHub link.
  5. [§3, paragraph after (3.2)] Contribution (C2) and Remark 2 stress that allowing sign-indefinite positivity-preserving disturbances is essential for the adaptive reformulation. Cross-reference this earlier when the next-state constraint v ≥ 0 is introduced in §3, so the motivation is visible before the history construction.
  6. [Corollary 1 / Remark 8] Corollary 1’s piecewise γ_CE(a) is nice; a one-line comparison to the known-sign model-based gain (which is 1 when r=0 and the system is reset by u = −(a/b)x) would clarify the adaptation price.

Circularity Check

1 steps flagged

No significant circularity: standard minimax DP with postulated linear certificates verified by direct inequality checks; self-citations supply background tools, not load-bearing uniqueness.

specific steps
  1. self citation load bearing [Section 3, Remark 5; Section 1.1 (C3); citations to [32]]
    "The reformulation technique used here follows a principle similar to that in [32], with one important difference. In [32], the history variable is defined in terms of a global covariance historical-data matrix that is independent of the model set M. By contrast, in our formulation, the history variables are defined locally for each possible pair (A,B) in M."

    Minor: the history-variable reformulation idea is imported from the same research line ([32], Rantzer). It is not load-bearing uniqueness or a forbidden alternative; the paper re-derives equivalence for the ell_1 / positive setting in Theorem 1 with full proof in Appendix C.1, and the local z^(A,B) construction is different from the covariance matrix in [32]. Score contribution is at most 1.

full rationale

The paper's central chain is: reformulate the original game via next-state and model-specific history variables (Problem 2), establish value equivalence and that any solution of the Bellman inequality yields a cost upper bound (Theorem 1), then postulate a max-of-pairwise linear value function V = max Vij with Vij = p_ij^T x - (1/2)(z^i + z^j) and verify FV <= V under the linear inequalities (5.7) (Theorems 2-3). The p_ij and K_i are design variables chosen to satisfy those inequalities (Algorithm 1 / Remark 7); they are not fitted to data that already encode the claimed gain, nor defined in terms of the target J*. History variables z^(A,B) are cumulative prediction residuals by construction, used as a sufficient statistic, not as a fitted prediction of the cost bound. Self-citations to Rantzer [32] (history reformulation idea), Gurpegui et al. [23] (model-based positive LP), and related group work supply independent background statements that are re-derived or extended in the appendices (Appendix B for the positivity-preserving disturbance extension). The paper itself flags that (5.7) and Algorithm 1 are only sufficient and need not be tight (Remark 11), and lists exact Bellman / continuum M as future work. No step reduces the claimed certificate to its inputs by definition.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 2 invented entities

Load-bearing content is standard discrete-time minimax DP plus domain structure of positive linear systems with a finite known model class and a common positivity-preserving input constraint. Design weights (s,r,γ,E) and candidate gains are chosen by the designer to satisfy linear inequalities; they are not data-fitted physical constants. No new physical entities are postulated—history scores and pairwise value vectors are mathematical devices.

free parameters (3)
  • Disturbance penalty vector γ = Example-dependent (e.g. [42/5, 25], [15,15,15])
    Chosen large enough so p_ij≤γ and the mixed inequalities hold; Algorithm 1 and Remark 7 treat γ as a design knob, not an estimated physical constant.
  • Stage-cost weights s,r and input bound E
    Design parameters fixed by Assumption 3 (A≥|B|E, s≥E^⊤|r|) to enforce positivity and nonnegative stage cost; selected by the user for each application.
  • Pairwise cost vectors p_ij and gains K_i
    Solved from linear fixed-point/max inequalities (5.8)–(5.10); the recipe is sufficient and may be tightened, so certified bounds depend on this design choice.
axioms (5)
  • domain assumption True plant (A★,B★) lies in a known compact/finite set M (Assumption 1; finite M in §5).
    Without a known finite list the model-local histories and CE index arg min_i z^{(i)} are undefined as stated.
  • domain assumption Common E with A≥|B|E for all (A,B)∈M and s≥E^⊤|r| (Assumption 3), so |u|≤Ex preserves positivity and nonnegative stage cost.
    Enters all closed-loop positivity and cost arguments from §5 onward and in the model-based Appendix B.
  • domain assumption Disturbances need only satisfy w≥−(Ax+Bu) (positivity-preserving), not w≥0.
    Used to make next-state constraint v≥0 independent of unknown (A,B), enabling the reformulation in §3 (Remark 2).
  • standard math Standard minimax dynamic programming: monotone value iteration, Bellman inequality implies cost upper bound (Lemma 3 / Theorem 1).
    Classical zero-sum DP background cited via [8,24]; proofs in Appendix C rely on it.
  • ad hoc to paper Existence of p_ij, K_i satisfying the linear inequalities (5.7) with 0≤p_ij≤γ (sufficient, not necessary).
    This is the paper-specific certificate; if no such vectors exist the explicit CE guarantee of Theorem 3 does not apply even if some other adaptive law works.
invented entities (2)
  • Model-specific cumulative history scalars z^{(A,B)}_t no independent evidence
    purpose: Compress past prediction errors per candidate plant so the controller can run CE without full trajectory dependence.
    Defined in (3.3)–(3.5); mathematical bookkeeping, not a physical object. Independent evidence is not applicable beyond the closed-loop simulations.
  • Pairwise max-linear value function V=max_{i,j}(p_ij^⊤x−½(z^{(i)}+z^{(j)})) no independent evidence
    purpose: Witness for the Bellman inequality under finite model uncertainty.
    Ansatz in Theorem 3; standard Lyapunov/value construction rather than a new ontological entity.

pith-pipeline@v1.2.0-daily-grok45 · 33919 in / 3549 out tokens · 67513 ms · 2026-07-30T20:14:18.840393+00:00 · methodology

0 comments
read the original abstract

We present a minimax adaptive control framework for discrete-time positive linear systems with parametric uncertainty and adversarial disturbances. The uncertainty in the system dynamics is assumed to lie in a finite set of possible plants. We formulate the problem as a dynamic game between the controller, which minimizes the cost, and an adversary, which selects both the disturbances and the plant dynamics to maximize the cost. An equivalent reformulation of the original game transforms the problem into a standard minimax two-player zero-sum dynamic game. This enables the problem to be addressed via minimax dynamic programming. We provide an explicit solution to the Bellman inequality, yielding stabilizing, positivity preserving policies without requiring an initially stabilizing controller. The resulting controller enjoys robustness guarantees in the form of bounded $\ell_1$-gain from disturbances to errors. Once the uncertain parameters have been sufficiently estimated, the controller behaves like a standard $\mathcal H_\infty$-type optimal controller for positive linear systems. The theoretical findings are supported by numerical experiments illustrating the resulting adaptive controller in action.

Figures

Figures reproduced from arXiv: 2607.26816 by Anders Rantzer, Fethi Bencherki.

Figure 1
Figure 1. Figure 1: Feedback interconnection of a discrete-time positive linear system with uncertain parameters, external disturbances, unmodelled dynamics, and an adaptive controller. The goal is to design the controller so as to minimize the worst case induced ℓ1-gain from disturbances to errors, while preserving positivity and accounting for parametric uncertainty. based problem still admits a solution via linear programm… view at source ↗
Figure 2
Figure 2. Figure 2: Two-reservoir water network with uncertain transfer direction. The states x1 and x2 denote stored water volumes, ut is the pump command, and wt represents exogenous inflows and withdrawals. The matrix A captures passive storage and hydraulic coupling, while the uncertainty ±B means that a positive control input may transfer water in either direction. Next, we derive an upper bound on the minimum achievable… view at source ↗
Figure 3
Figure 3. Figure 3: Closed loop simulation for Example 1, comparing the adaptive controller for the unknown input direction with the optimal controller that knows the true sign. The panels show, from top left to bottom right, x1, x2, the control input ut , the positivity-preserving disturbance wt , the history variables z (+) and z (−) , and the true and estimated signs. The true sign is fixed but unknown to the adaptive cont… view at source ↗
Figure 4
Figure 4. Figure 4: Three queue multiclass job queueing network. The scalar admission command ut regulates the external workload entering queues Q1 and Q2. Queues Q1, Q2, and Q3 are processed by servers S1, S2, and S3, respectively. A portion of the workload served from Q1 may be routed to Q2 or directly to the shared downstream queue Q3, while another portion may leave the modeled network. Similarly, a portion of the workloa… view at source ↗
Figure 5
Figure 5. Figure 5: Closed loop trajectories under the adaptive certainty equivalence controller and the optimal controller with known mode. The dashed vertical line indicates the mode switching time. The figures show the workloads in queues Q1–Q3, the admission command, the history variables, and the true and estimated modes. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗

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 · 1 canonical work pages

  1. [1]

    Explicit solutions to Bellman equation for positive systems with linear cost

    A. Rantzer. “Explicit solutions to Bellman equation for positive systems with linear cost”. In:61st IEEE Conf. Decis. Control. 2022, pp. 6154–6155

    2022

  2. [2]

    A. A. Al Makdah and F. Pasqualetti.Model-based and Data-based Dynamic Output Feedback for Externally Positive Systems. arXiv preprint arXiv:2305.02472. 2023

    Pith/arXiv arXiv 2023

  3. [3]

    Adaptive control and intersections with reinforcement learning

    A. M. Annaswamy. “Adaptive control and intersections with reinforcement learning”. In: Annu. Rev. Control Robot. Auton. Syst.6.1 (2023), pp. 65–93. 32

    2023

  4. [4]

    A historical perspective of adaptive control and learning

    A. M. Annaswamy and A. L. Fradkov. “A historical perspective of adaptive control and learning”. In:Annu. Rev. Control52 (2021), pp. 18–41

    2021

  5. [5]

    A Linear Programming Approach to Routing Control in Networks of Constrained Nonlinear Positive Systems with Concave Flow Rates

    H. Arneson, N. Dousse, and C. Langbort. “A Linear Programming Approach to Routing Control in Networks of Constrained Nonlinear Positive Systems with Concave Flow Rates”. In:Automatica68 (2016), pp. 357–368

    2016

  6. [6]

    K. J. Astrom and B. Wittenmark.Adaptive control. Addison-Wesley Longman Publishing Co., Inc., 1994

    1994

  7. [7]

    On self tuning regulators

    K. J. Åström and B. Wittenmark. “On self tuning regulators”. In:Automatica9.2 (1973), pp. 185–199

    1973

  8. [8]

    Başar and P

    T. Başar and P. Bernhard.H-infinity optimal control and related minimax design problems: a dynamic game approach. Springer Sci. & Bus. Media, 2008

    2008

  9. [9]

    Adaptive Control of Positive Systems with Application to Learning SSP

    F. Bencherki and A. Rantzer. “Adaptive Control of Positive Systems with Application to Learning SSP”. In:Proc. 7th Annu. Learn. Dyn. Control Conf.PMLR. 2025, pp. 660–672

    2025

  10. [10]

    Data-Driven Adaptive Dispatching Policies for Processing Networks

    F. Bencherki and A. Rantzer. “Data-Driven Adaptive Dispatching Policies for Processing Networks”. In: vol. 8. IEEE, 2024, pp. 2841–2846

    2024

  11. [11]

    Robust simultaneous stabilization via minimax adaptive control

    F. Bencherki and A. Rantzer. “Robust simultaneous stabilization via minimax adaptive control”. In:Proc. 2023 IEEE 62nd Conf. Decis. Control (CDC). IEEE. 2023, pp. 2503– 2508

    2023

  12. [12]

    Berman and R

    A. Berman and R. J. Plemmons.Nonnegative matrices in the mathematical sciences. SIAM, 1994

    1994

  13. [13]

    Optimal control of multiclass fluid queueing networks: A machine learning approach

    D. Bertsimas and C. W. Kim. “Optimal control of multiclass fluid queueing networks: A machine learning approach”. In:arXiv preprint arXiv:2307.12405(2023)

    Pith/arXiv arXiv 2023

  14. [14]

    Optimal control of compartmental models: The exact solution

    F. Blanchini, P. Bolzern, P. Colaneri, G. De Nicolao, and G. Giordano. “Optimal control of compartmental models: The exact solution”. In:Automatica147 (2023), p. 110680

    2023

  15. [15]

    Synthesis of minimax adaptive controller for a finite set of linear systems

    D. Cederberg, A. Hansson, and A. Rantzer. “Synthesis of minimax adaptive controller for a finite set of linear systems”. In:Proc. 2022 IEEE 61st Conf. Decis. Control (CDC). IEEE. 2022, pp. 1380–1384

    2022

  16. [16]

    A Compartmental Model for Traffic Networks and Its Dynamical Behavior

    S. Coogan and M. Arcak. “A Compartmental Model for Traffic Networks and Its Dynamical Behavior”. In:IEEE Trans. Autom. Control60.10 (2015), pp. 2698–2703.doi:10.1109/ TAC.2015.2411916

    arXiv 2015

  17. [17]

    Minimax adaptive control of uncertain plants

    G. Didinsky and T. Basar. “Minimax adaptive control of uncertain plants”. In:Proc. 1994 33rd IEEE Conf. Decis. Control. Vol. 3. IEEE. 1994, pp. 2839–2844

    1994

  18. [18]

    Farina and S

    L. Farina and S. Rinaldi.Positive linear systems: theory and applications. John Wiley & Sons, 2011

    2011

  19. [19]

    Discrete time stochastic adaptive control

    G. C. Goodwin, P. J. Ramadge, and P. E. Caines. “Discrete time stochastic adaptive control”. In:SIAM J. Control Optim.19.6 (1981), pp. 829–853

    1981

  20. [20]

    G. C. Goodwin and K. S. Sin.Adaptive filtering prediction and control. Courier Corporation, 2014

    2014

  21. [21]

    Convergence and logarithm laws of self-tuning regulators

    L. Guo. “Convergence and logarithm laws of self-tuning regulators”. In:Automatica31.3 (1995), pp. 435–450

    1995

  22. [22]

    Minimax linear regulator problems for positive systems

    A. Gurpegui, M. Jeeninga, E. Tegling, and A. Rantzer. “Minimax linear regulator problems for positive systems”. In:IEEE Trans. Autom. Control(2026)

    2026

  23. [23]

    A minimax optimal controller for positive systems

    A. Gurpegui, E. Tegling, and A. Rantzer. “A minimax optimal controller for positive systems”. In:arXiv preprint arXiv:2502.01180(2025)

    Pith/arXiv arXiv 2025

  24. [24]

    Minimax linear optimal control of positive systems

    A. Gurpegui, E. Tegling, and A. Rantzer. “Minimax linear optimal control of positive systems”. In:IEEE Control Syst. Lett.7 (2023), pp. 3920–3925

    2023

  25. [25]

    Data Informativity for Distributed Positive Stabilization

    T. Iwata, S.-i. Azuma, R. Ariizumi, and T. Asai. “Data Informativity for Distributed Positive Stabilization”. In:IEEE Control Syst. Lett.8 (2024), pp. 826–831

    2024

  26. [26]

    Exact dynamic programming for positive systems with linear optimal cost

    Y. Li and A. Rantzer. “Exact dynamic programming for positive systems with linear optimal cost”. In:IEEE Trans. Autom. Control69.12 (2024), pp. 8738–8750. 33

    2024

  27. [27]

    From self-tuning regulators to reinforcement learning and back again

    N. Matni, A. Proutiere, A. Rantzer, and S. Tu. “From self-tuning regulators to reinforcement learning and back again”. In:Proc. 2019 IEEE 58th Conf. Decis. Control (CDC). IEEE. 2019, pp. 3724–3740

    2019

  28. [28]

    Data-Driven Control of Positive Linear Systems using Linear Programming

    J. Miller, T. Dai, M. Sznaier, and B. Shafai. “Data-Driven Control of Positive Linear Systems using Linear Programming”. In:Proc. 2023 62nd IEEE Conf. Decis. Control (CDC). IEEE. 2023, pp. 1588–1594

    2023

  29. [29]

    Adaptive control using multiple models

    K. S. Narendra and J. Balakrishnan. “Adaptive control using multiple models”. In:IEEE Trans. Autom. Control42.2 (1997), pp. 171–187

    1997

  30. [30]

    Heuristic search for linear positive systems

    D. Ohlin, A. Rantzer, and E. Tegling. “Heuristic search for linear positive systems”. In: Automatica192 (2026), p. 113153

    2026

  31. [31]

    Optimal control of linear cost networks

    D. Ohlin, E. Tegling, and A. Rantzer. “Optimal control of linear cost networks”. In:Eur. J. Control80 (2024), p. 101068

    2024

  32. [32]

    Minimax adaptive control for a finite set of linear systems

    A. Rantzer. “Minimax adaptive control for a finite set of linear systems”. In:Proc. Mach. Learn. Res. (L4DC). PMLR. 2021, pp. 893–904

    2021

  33. [33]

    Scalable Control of Positive Systems

    A. Rantzer. “Scalable Control of Positive Systems”. In:Eur. J. Control24 (2015), pp. 72–80

    2015

  34. [34]

    Scalable control of positive systems

    A. Rantzer and M. E. Valcher. “Scalable control of positive systems”. In:Annu. Rev. Control Robot. Auton. Syst.4.1 (2021), pp. 319–341

    2021

  35. [35]

    An online learning analysis of minimax adaptive control

    V. Renganathan, A. Iannelli, and A. Rantzer. “An online learning analysis of minimax adaptive control”. In:Proc. 2023 IEEE 62nd Conf. Decis. Control (CDC). IEEE. 2023, pp. 1034–1039

    2023

  36. [36]

    Formal Synthesis of Control Strategies for Positive Monotone Systems

    S. Sadraddini and C. Belta. “Formal Synthesis of Control Strategies for Positive Monotone Systems”. In:IEEE Trans. Autom. Control64.2 (2019), pp. 480–495.doi:10.1109/TAC. 2018.2814631

    arXiv 2019

  37. [37]

    Data-Driven Positive Stabilization of Linear Systems

    B. Shafai, A. Moradmand, and M. Siami. “Data-Driven Positive Stabilization of Linear Systems”. In:Proc. 2022 8th Int. Conf. Control, Decis. Inf. Technol. (CoDIT). IEEE. 2022, pp. 1031–1036

    2022

  38. [38]

    Throughput Properties of a Queueing Network with Distributed Dynamic Routing and Flow Control

    L. Tassiulas and A. Ephremides. “Throughput Properties of a Queueing Network with Distributed Dynamic Routing and Flow Control”. In:Adv. Appl. Probab.28.1 (1996), pp. 285–307.doi:10.2307/1427922

    work page doi:10.2307/1427922 1996

  39. [39]

    Statistical learning theory for control: A finite-sample perspective

    A. Tsiamis, I. Ziemann, N. Matni, and G. J. Pappas. “Statistical learning theory for control: A finite-sample perspective”. In:IEEE Control Syst. Mag.43.6 (2023), pp. 67–97

    2023

  40. [40]

    Examples and counterexamples in finiteℓ2-gain adaptive control

    G. Vinnicombe. “Examples and counterexamples in finiteℓ2-gain adaptive control”. In: Leuven: Sixteenth International Symposium on Mathematical Theory of Networks and Systems (MTNS2004). 2004

    2004

  41. [41]

    Data Driven Positive Subspace System Identification

    Y. Wang and B. Shafai. “Data Driven Positive Subspace System Identification”. In:IFAC- PapersOnLine58.15 (2024), pp. 372–377

    2024

  42. [42]

    Data-Driven Identification and Control of Positive Systems

    Y. Wang and B. Shafai. “Data-Driven Identification and Control of Positive Systems”. In: Integrated Systems: Data Driven Engineering. Springer, 2024, pp. 289–307. 34

    2024