pith. sign in

arxiv: 2512.18356 · v5 · submitted 2025-12-20 · 📡 eess.SY · cs.SY· math.OC

Robust H2/H-infinity control under stochastic requirements: minimizing conditional value-at-risk instead of worst-case performance

Ervan Kassarian , Francesco Sanfedino , Daniel Alazard , Andrea Marrazza This is my paper

Pith reviewed 2026-05-16 20:50 UTC · model grok-4.3

classification 📡 eess.SY cs.SYmath.OC
keywords robust controlconditional value-at-riskH-infinitystochastic optimizationMonte Carlo samplingparametric uncertaintycontroller design
0
0 comments X

The pith

Optimizing robust H2/H-infinity controllers by minimizing conditional value-at-risk instead of worst-case performance reduces design conservatism.

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

The paper introduces an alternative to traditional robust control that minimizes the conditional value-at-risk of performance metrics rather than their worst-case value. This leverages Monte Carlo sampling of parametric uncertainties to produce less conservative controller designs. The approach accepts some performance degradation in rare extreme scenarios in exchange for better performance in more probable cases. It is demonstrated on a mechanical system where overall performance improves significantly.

Core claim

By optimizing the controller with respect to the conditional value at risk of the performance under sampled uncertainties, rather than the worst-case performance, the design avoids excessive conservatism while still addressing stochastic parametric variations through sampling.

What carries the argument

Conditional value-at-risk (CVaR) minimization applied to the closed-loop performance cost under Monte Carlo sampled parametric uncertainties, serving as the objective in place of the standard worst-case H-infinity or H2 norm.

Load-bearing premise

That de-emphasizing rare worst-case parametric configurations through CVaR optimization will still produce controllers whose performance is acceptable when those rare cases occur.

What would settle it

Observing a controller from this method that exhibits unacceptably poor performance in one of the low-probability uncertainty samples that was not heavily weighted in the CVaR calculation.

Figures

Figures reproduced from arXiv: 2512.18356 by Andrea Marrazza, Daniel Alazard, Ervan Kassarian, Francesco Sanfedino.

Figure 2
Figure 2. Figure 2: Probability density p zw k (α) of the loss function ||Tzw(s, k, δ)|| with β-VAR, β-CVAR and deterministic worst case (only defined under Assumption 1). Assumption 3 is adopted here following the work of [19] to properly introduce Definition 1. Definition 1 (β-VAR, β-CVAR) Given the loss function Lzw(k, δ) satisfying Assump￾tions 1, 2 and 3, the β-Value at risk (β-VAR), noted VARzw β (k), is: VARzw β (k) = … view at source ↗
Figure 1
Figure 1. Figure 1: LFR representation of Tzw(s, k, δ) Assumption 1 The random variable δ has bounded sup￾port Dδ, and its probability distribution admits a density p(δ). Assuming a bounded support is common in control engineering, as it is necessary in the traditional de￾terministic framework to define a worst case (as in Eq. (1)). In this paper, it is necessary to realistically make Assumption 2 below. Without loss of gener… view at source ↗
Figure 3
Figure 3. Figure 3: Benchmark: closed-loop system The uncertain parameters represent mechanical prop￾erties. Once normalized, they all follow a Gaussian distribution of mean 0 and standard deviation 1/3, truncated at 3 σ, except for one parameter (representing an angle) which follows a uniform distribution between -1 and 1. Additionally, the vectors δ that do not satisfy a nonlinear constraint c(δ) ≤ 0, related to some mechan… view at source ↗
Figure 4
Figure 4. Figure 4: Empirical probability densities with solution [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Empirical probability densities with solution [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
read the original abstract

Conventional robust H2/H-infinity control minimizes the worst-case performance, often leading to a conservative design driven by very rare parametric configurations. To reduce this conservatism while taking advantage of the stochastic properties of Monte Carlo sampling and its compatibility with parallel computing, we introduce an alternative paradigm that optimizes the controller with respect to a stochastic criterion, namely the conditional value at risk. We present the problem formulation and discuss several open challenges toward a general synthesis framework. The potential of this approach is illustrated on a mechanical system, where it significantly improves overall performance by tolerating some degradation in very rare worst-case scenarios.

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 / 1 minor

Summary. The paper proposes replacing worst-case minimization in robust H2/H∞ control with minimization of conditional value-at-risk (CVaR) over a Monte Carlo-sampled distribution of parametric uncertainties. This aims to reduce conservatism by tolerating degradation in rare tail events while leveraging stochastic sampling properties. The manuscript presents the problem formulation, discusses open synthesis challenges, and illustrates the idea on a mechanical system example claiming improved overall performance.

Significance. If the CVaR-based paradigm can be equipped with synthesis algorithms and closed-loop guarantees, it would offer a less conservative alternative to classical robust control by incorporating probabilistic information, potentially improving average-case performance in sampled uncertainty settings without fully sacrificing tail robustness.

major comments (2)
  1. Problem formulation section: the transition from standard H2/H∞ worst-case optimization to CVaR minimization is stated at a conceptual level but supplies no explicit mathematical derivation, optimization problem statement, or closed-loop stability/performance guarantees, which are load-bearing for establishing the approach as a viable paradigm.
  2. Mechanical system illustration: the claim of 'significantly improves overall performance' is unsupported by any quantitative metrics, comparison tables, or data against standard H2/H∞ designs, leaving the practical benefit unevaluated.
minor comments (1)
  1. Abstract: the mention of 'several open challenges' is not enumerated, which reduces clarity on the scope of the contribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address the major points below and indicate the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: Problem formulation section: the transition from standard H2/H∞ worst-case optimization to CVaR minimization is stated at a conceptual level but supplies no explicit mathematical derivation, optimization problem statement, or closed-loop stability/performance guarantees, which are load-bearing for establishing the approach as a viable paradigm.

    Authors: The manuscript is structured as an introduction of the CVaR paradigm together with an explicit discussion of open synthesis challenges, rather than a complete synthesis theory. In the revision we will add an explicit optimization problem statement that replaces the worst-case operator with CVaR over the Monte-Carlo-sampled uncertainty set, together with the corresponding derivation from the classical formulation. Closed-loop stability and performance guarantees are among the open challenges already flagged in the paper; we will keep this limitation clearly stated and will not claim guarantees that are not yet available. revision: partial

  2. Referee: Mechanical system illustration: the claim of 'significantly improves overall performance' is unsupported by any quantitative metrics, comparison tables, or data against standard H2/H∞ designs, leaving the practical benefit unevaluated.

    Authors: We agree that the current example is only qualitative. In the revised manuscript we will augment the mechanical-system section with quantitative comparisons, including tables that report average closed-loop cost, CVaR values, and worst-case values for both the CVaR-designed controller and the classical H2/H∞ controller under the same uncertainty samples. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper proposes a methodological shift from worst-case H2/H-infinity minimization to CVaR minimization over Monte Carlo-sampled parametric uncertainties. No load-bearing equations or derivations are supplied that reduce the claimed controller improvement to a fitted parameter, self-definition, or self-citation chain; the CVaR criterion is imported as an external stochastic risk measure, the formulation explicitly flags open synthesis challenges, and the mechanical-system illustration serves only as a demonstration rather than a closed derivation. The central claim is therefore self-contained as an alternative paradigm without internal reduction to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The proposal rests on the domain assumption that Monte Carlo sampling faithfully captures the relevant stochastic properties of parametric uncertainty and that CVaR provides a meaningful and tunable trade-off between average and tail performance.

axioms (1)
  • domain assumption Monte Carlo sampling of parametric uncertainty accurately represents the stochastic properties needed for CVaR evaluation
    Invoked to justify compatibility with parallel computing and the stochastic criterion

pith-pipeline@v0.9.0 · 5412 in / 1179 out tokens · 17177 ms · 2026-05-16T20:50:44.831345+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages · 2 internal anchors

  1. [1]

    The H infinity Control Problem is Solved,

    P. Apkarian and D. Noll, “The H infinity Control Problem is Solved,” Aerospace Lab, no. 13, pp. 1–27, 2017

    work page 2017

  2. [2]

    Worst-case search in constrained uncertainty space for robust H-infinity synthesis,

    E. Kassarian, F. Sanfedino, D. Alazard, and A. Marrazza, “Worst-case search in constrained uncertainty space for robust H-infinity synthesis,” Nov. 2025, arXiv:2511.15480 [eess]. [Online]. A vailable: http://arxiv.org/abs/2511.15480

    work page arXiv 2025

  3. [3]

    Parametric Robust Structured Control Design,

    P. Apkarian, M. N. Dao, and D. Noll, “Parametric Robust Structured Control Design,” IEEE Transactions on Automatic Control, vol. 60, no. 7, pp. 1857–1869, 2015, arXiv: 1405.4202

    work page arXiv 2015

  4. [4]

    Clarke, Optimization and nonsmooth analysis

    F. Clarke, Optimization and nonsmooth analysis. Society for industrial and Applied Mathematics, 1990

    work page 1990

  5. [5]

    Boyd and C

    S. Boyd and C. Barratt, Linear Controller Design: Limits of Performance. Englewood Cliffs, NJ: Prentice Hall., 1991

    work page 1991

  6. [6]

    Nonsmooth H infinity synthesis,

    P. Apkarian and D. Noll, “Nonsmooth H infinity synthesis,” IEEE Transactions on Automatic Control, vol. 51, no. 1, pp. 71–86, 2006

    work page 2006

  7. [7]

    Probabilistic robustness analysis: explicit bounds for the minimum number of samples,

    R. Tempo, E. Bai, and F. Dabbene, “Probabilistic robustness analysis: explicit bounds for the minimum number of samples,” in Proceedings of 35th IEEE Conference on Decision and Control, vol. 3. Kobe, Japan: IEEE, 1996, pp. 3424–3428. [Online]. A vailable: http://ieeexplore.ieee.org/ document/573690/

    work page 1996

  8. [8]

    Analysis of feedback systems with structured uncertainties,

    J. Doyle, “Analysis of feedback systems with structured uncertainties,” IEE Proceedings D Control Theory and Applications, vol. 129, no. 6, p. 242, 1982. [Online]. A vailable: https://digital-library.theiet.org/content/ journals/10.1049/ip-d.1982.0053

    work page doi:10.1049/ip-d.1982.0053 1982

  9. [9]

    Systems modeling, analysis and control (SMAC) toolbox: An insight into the robustness analysis library,

    C. Roos, “Systems modeling, analysis and control (SMAC) toolbox: An insight into the robustness analysis library,” in 2013 IEEE Conference on Computer Aided Control System Design (CACSD). Hyderabad, India: IEEE, Aug. 2013, pp. 176–181. [Online]. A vailable: http://ieeexplore.ieee.org/ document/6663479/

    work page arXiv 2013

  10. [10]

    Advanced probabilistic mu -analysis techniques for AOCS validation,

    J.-M. Biannic, C. Roos, S. Bennani, F. Boquet, V. Preda, and B. Girouart, “Advanced probabilistic mu -analysis techniques for AOCS validation,” European Journal of Control, vol. 62, pp. 120–129, Nov. 2021. [Online]. A vailable: https: //linkinghub.elsevier.com/retrieve/pii/S0947358021000789

    work page 2021

  11. [11]

    On the complexity of purely complex ￿ computation and related problems in multidimensional systems,

    O. Toker and H. Ozbay, “On the complexity of purely complex ￿ computation and related problems in multidimensional systems,” IEEE Transactions on Automatic Control, vol. 43, no. 3, pp. 409–414, Mar. 1998. [Online]. A vailable: http: //ieeexplore.ieee.org/document/661609/

    work page 1998

  12. [12]

    Shapiro, D

    A. Shapiro, D. Dentcheva, and A. Ruszczyński, Lectures on Stochastic Programming: Modeling and Theory. Society for Industrial and Applied Mathematics, Jan. 2009. [Online]. A vailable: http://epubs.siam.org/doi/book/10.1137/ 1.9780898718751

    work page 2009

  13. [13]

    Wasserstein Distributionally Robust Motion Control for Collision A voidance Using Conditional Value-at-Risk,

    A. Hakobyan and I. Yang, “Wasserstein Distributionally Robust Motion Control for Collision A voidance Using Conditional Value-at-Risk,” IEEE Transactions on Robotics, vol. 38, no. 2, pp. 939–957, Apr. 2022. [Online]. A vailable: https://ieeexplore.ieee.org/document/9547384/

    work page arXiv 2022

  14. [14]

    A Risk-Aware Control: Integrating Worst- Case CVaR with Control Barrier Function,

    M. Kishida, “A Risk-Aware Control: Integrating Worst- Case CVaR with Control Barrier Function,” Aug. 2023, arXiv:2308.14265 [math]. [Online]. A vailable: http://arxiv. org/abs/2308.14265

    work page arXiv 2023

  15. [15]

    Distributionally Robust Control of Constrained Stochastic Systems,

    B. Van Parys, D. Kuhn, P. Goulart, and M. Morari, “Distributionally Robust Control of Constrained Stochastic Systems,” IEEE Transactions on Automatic Control, pp. 1–1, 2015. [Online]. A vailable: http://ieeexplore.ieee.org/ document/7122259/

    work page arXiv 2015

  16. [16]

    Computational Design of Optimal Output Feedback Controllers,

    T. Rautert and E. W. Sachs, “Computational Design of Optimal Output Feedback Controllers,” SIAM Journal on Optimization, vol. 7, no. 3, pp. 837–852, Aug

    work page

  17. [17]

    A vailable: http://epubs.siam.org/doi/10.1137/ S1052623495290441

    [Online]. A vailable: http://epubs.siam.org/doi/10.1137/ S1052623495290441

    work page

  18. [18]

    Mixed H2/Hinfinity control via nonsmooth optimization,

    P. Apkarian, D. Noll, and A. Rondepierre, “Mixed H2/Hinfinity control via nonsmooth optimization,” in Proceedings of the 48h IEEE Conference on Decision and Control (CDC) held jointly with 2009 28th Chinese Control Conference. Shanghai, China: IEEE, Dec. 2009, pp. 6460–6465. [Online]. A vailable: http://ieeexplore.ieee.org/ document/5400165/

    work page arXiv 2009

  19. [19]

    H2 for HIFOO

    D. Arzelier, D. Georgia, S. Gumussoy, and D. Henrion, “H2 for HIFOO,” Oct. 2010, arXiv:1010.1442 [math]. [Online]. A vailable: http://arxiv.org/abs/1010.1442

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  20. [20]

    Optimization of conditional value-at-risk,

    R. T. Rockafellar and S. Uryasev, “Optimization of conditional value-at-risk,” The Journal of Risk, vol. 2, no. 3, pp. 21–41, 2000. [Online]. A vail- able: http://www.risk.net/journal-of-risk/technical-paper/ 2161159/optimization-conditional-value-risk

    work page 2000

  21. [21]

    Monte Carlo sampling-based methods for stochastic optimization,

    T. Homem-de Mello and G. Bayraksan, “Monte Carlo sampling-based methods for stochastic optimization,” Surveys in Operations Research and Management Science, vol. 19, no. 1, pp. 56–85, Jan. 2014. [Online]. A vailable: https: //linkinghub.elsevier.com/retrieve/pii/S1876735414000038

    work page 2014

  22. [22]

    Monte Carlo sampling approach to stochastic programming,

    A. Shapiro, “Monte Carlo sampling approach to stochastic programming,” ESAIM: Proceedings, vol. 13, pp. 65–73, Dec. 2003. [Online]. A vailable: http://www.esaim-proc.org/ 10.1051/proc:2003003

    work page doi:10.1051/proc:2003003 2003

  23. [23]

    Computing VaR and CVaR using stochastic approximation and adaptive unconstrained importance sampling,

    O. Bardou, N. Frikha, and G. Pagès, “Computing VaR and CVaR using stochastic approximation and adaptive unconstrained importance sampling,” Monte Carlo Methods and Applications, vol. 15, no. 3, Jan. 2009. [Online]. A vailable: https://www.degruyter.com/document/ doi/10.1515/MCMA.2009.011/html

    work page doi:10.1515/mcma.2009.011/html 2009

  24. [24]

    Optimizing the CVaR via Sampling

    A. Tamar, Y. Glassner, and S. Mannor, “Optimizing the CVaR via Sampling,” Nov. 2014, arXiv:1404.3862 [stat]. [Online]. A vailable: http://arxiv.org/abs/1404.3862

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  25. [25]

    Cutting plane algorithms for mean-CVaR portfolio optimization with nonconvex transaction costs,

    Y. Takano, K. Nanjo, N. Sukegawa, and S. Mizuno, “Cutting plane algorithms for mean-CVaR portfolio optimization with nonconvex transaction costs,” Computational Management Science, vol. 12, no. 2, pp. 319–340, Apr. 2015. [Online]. A vail- able: http://link.springer.com/10.1007/s10287-014-0209-7

    work page doi:10.1007/s10287-014-0209-7 2015

  26. [26]

    Sample average approximation of conditional value-at-risk based variational inequalities,

    A. Cherukuri, “Sample average approximation of conditional value-at-risk based variational inequalities,” Optimization Letters, vol. 18, no. 2, pp. 471–496, Mar. 2024. [Online]. A vail- able: https://link.springer.com/10.1007/s11590-023-01996-9

    work page doi:10.1007/s11590-023-01996-9 2024

  27. [27]

    Monte Carlo bounding techniques for determining solution quality in stochastic programs,

    W.-K. Mak, D. P. Morton, and R. Wood, “Monte Carlo bounding techniques for determining solution quality in stochastic programs,” Operations Research Letters, vol. 24, no. 1-2, pp. 47–56, Feb. 1999. [Online]. A vailable: https: //linkinghub.elsevier.com/retrieve/pii/S0167637798000546

    work page 1999

  28. [28]

    Bundle Method for Non-Convex Minimization with Inexact Subgradients and Function Values,

    D. Noll, “Bundle Method for Non-Convex Minimization with Inexact Subgradients and Function Values,” in Computational and Analytical Mathematics, D. H. Bailey, H. H. Bauschke, P. Borwein, F. Garvan, M. Théra, J. D. Vanderwerff, and H. Wolkowicz, Eds. New York, NY: Springer New York, 2013, vol. 50, pp. 555–592, series Title: Springer Proceedings in Mathemat...

    work page doi:10.1007/978-1-4614-7621-4_26 2013

  29. [29]

    Stochastic Optimization with Optimal Importance Sampling,

    L. Aolaritei, B. P. G. V. Parys, H. Lam, and M. I. Jordan, “Stochastic Optimization with Optimal Importance Sampling,” Apr. 2025, arXiv:2504.03560 [math]. [Online]. A vailable: http://arxiv.org/abs/2504.03560

    work page arXiv 2025

  30. [30]

    A general framework of importance sampling for value-at-risk and conditional value-at-risk,

    L. Sun and L. J. Hong, “A general framework of importance sampling for value-at-risk and conditional value-at-risk,” in Proceedings of the 2009 Winter Simulation Conference (WSC). Austin, TX, USA: IEEE, Dec. 2009, pp. 415–

    work page 2009

  31. [31]

    A vailable: http://ieeexplore.ieee.org/document/ 5429348/

    [Online]. A vailable: http://ieeexplore.ieee.org/document/ 5429348/

    work page