Strategically Robust Linear Quadratic Dynamic Games

Boris Velasevic; Eric Mazumdar; Nicolas Lanzetti

arxiv: 2604.22318 · v1 · submitted 2026-04-24 · 🧮 math.OC · cs.GT· cs.SY· eess.SY

Strategically Robust Linear Quadratic Dynamic Games

Boris Velasevic , Nicolas Lanzetti , Eric Mazumdar This is my paper

Pith reviewed 2026-05-08 11:17 UTC · model grok-4.3

classification 🧮 math.OC cs.GTcs.SYeess.SY

keywords linear quadratic dynamic gamesstrategic robustnessRiccati equationsdynamic equilibriumrobust controlmulti-agent systemsnetwork gamesgame theory

0 comments

The pith

Linear quadratic dynamic games with policy uncertainty admit strategically robust equilibria computed as linear policies via coupled Riccati equations.

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

The paper establishes that uncertainty about other players' policies or goals in linear quadratic dynamic games can be modeled by transforming the problem into standard linear quadratic games where each player faces a fictitious adversary penalized for deviating from the others' policies. This yields a new notion of strategically robust dynamic equilibrium for which existence and uniqueness are proven, with equilibrium policies that are Markovian and linear. These policies are found efficiently by solving coupled backward Riccati equations. A sympathetic reader would care because the method supplies a tractable way to design resilient decentralized controllers for uncertain multi-agent systems such as networks or robotics teams. Simulations further indicate that the resulting robustness can sometimes raise individual utilities and social welfare without an explicit performance penalty.

Core claim

Strategically robust linear quadratic dynamic games can be rewritten as simple transformations of linear quadratic games in which each player chooses a controller facing an adversary penalized for deviating from the other players' policies. This formulation induces a strategically robust dynamic equilibrium whose policies are Markovian, linear, and efficiently computed via coupled backward Riccati equations. Existence and uniqueness of the equilibrium follow, and numerical simulations including network games illustrate benefits for robust decentralized control along with a free-lunch effect in which robustness improves utilities and welfare.

What carries the argument

The rewriting of the strategically robust game as a standard linear quadratic game with a fictitious adversary who is penalized for deviating from other players' policies.

If this is right

Equilibrium policies remain Markovian and linear despite the added robustness requirement.
Computation reduces to solving a set of coupled backward Riccati equations.
The method applies directly to network games and yields decentralized robust control schemes.
In some games robustness produces gains in both individual utilities and social welfare.

Where Pith is reading between the lines

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

The transformation idea might extend to other dynamic game classes if analogous fictitious-adversary reductions can be derived.
The observed free-lunch effect invites experiments that measure actual performance under real policy noise rather than the modeled adversary.
Connections to existing risk-averse or distributionally robust game formulations could be used to relax the exact penalty modeling assumption.

Load-bearing premise

Uncertainty about other players' policies or goals can be captured exactly by introducing a fictitious adversary who is penalized for deviating from those policies.

What would settle it

Implement the derived controllers in a simulated or physical multi-agent LQ system where opponents' actual policies vary in ways not matching the penalty model, then check whether the controllers fail to outperform standard equilibria under the realized variations.

Figures

Figures reproduced from arXiv: 2604.22318 by Boris Velasevic, Eric Mazumdar, Nicolas Lanzetti.

**Figure 2.** Figure 2: Social cost, defined as the sum of the agents’ realized costs, as a view at source ↗

**Figure 3.** Figure 3: Star graph with N = 5 nodes. TABLE II CENTRAL NODE PERFORMANCE UNDER ADVERSARIAL PERTURBATION Scenario Terminal state x1(T) Central node realized cost NE 4.32 2.39 SR 4.77 6.23 NE (adv.) 2.24 26.00 SR (adv.) 3.92 14.81 0 5 10 15 20 −2 0 2 4 6 Time State (central node) Target NE SR NE (adv.) SR (adv.) view at source ↗

**Figure 4.** Figure 4: State of the central node at the Nash equilibrium (NE), the view at source ↗

read the original abstract

We study linear quadratic dynamic games where players are uncertain about each other's control policies or goals and consequently seek to be strategically robust. Building on recent work on strategically robust and risk-averse game theory, we first formalize the problem of strategically robust linear quadratic dynamic games. We show that these can be rewritten as simple transformations of linear quadratic games in which each player chooses a controller in a fictitious game in which they are faced with an adversary who is penalized for deviating from the other players' policies. This formulation naturally induces a novel notion of dynamic equilibrium, which we call a strategically robust dynamic equilibrium. We establish existence and uniqueness of such equilibria and furthermore show that the equilibrium policies are Markovian, linear, and can be efficiently computed via coupled backward Riccati equations. Through numerical simulations, including experiments in a network game, we illustrate the benefits of strategic robustness in designing robust and resilient decentralized control schemes. Our experiments also expose a "free-lunch" phenomenon in games in which robustness does not incur a corresponding loss in performance but can yield improvements in players' utilities and social welfare.

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.

Desk Editor's Note private letter to a colleague

This paper reduces strategically robust LQ games to standard LQ games against a fictitious penalized adversary, delivering linear Markov equilibria via coupled Riccati equations plus a simulation-based free-lunch observation.

read the letter

The main contribution is the explicit rewriting of strategic robustness in linear-quadratic dynamic games as an equivalent game where each player faces a fictitious adversary penalized for deviating from the others' policies. This produces a new equilibrium notion with existence and uniqueness, Markovian linear policies, and computable solutions through the usual backward Riccati recursion under standard stabilizability and detectability conditions. The reduction stays inside classical LQ theory rather than inventing new machinery, which keeps the claims grounded and the numerics straightforward to run. The network-game example shows the approach in action and surfaces the free-lunch effect where robustness sometimes raises both individual payoffs and welfare without an obvious cost. That observation is useful for practitioners even if it remains empirical for now. The modeling step that turns policy uncertainty into the penalized-adversary term is the load-bearing assumption; if the real uncertainty does not map cleanly onto that penalty, the robustness interpretation weakens, though the paper states the transformation directly and does not hide the modeling choice. No circularity or hidden fitting appears in the derivations. The work is aimed at researchers in robust decentralized control, networked systems, and dynamic games who already use LQ methods and want a tractable way to add strategic robustness. It deserves a serious referee because the reduction is clean, the equilibrium properties follow from standard arguments, and the simulations provide concrete illustration without overclaiming generality.

Referee Report

0 major / 3 minor

Summary. The paper formalizes strategically robust linear quadratic (LQ) dynamic games in which players face uncertainty about others' policies or goals. It shows that such games admit an exact rewriting as standard LQ games against fictitious adversaries penalized for deviating from the other players' policies; this induces a notion of strategically robust dynamic equilibrium for which existence and uniqueness are established. The equilibria are proven to be Markovian and linear, and are obtained by solving a system of coupled backward Riccati equations under standard stabilizability and detectability conditions. Numerical examples, including a network game, illustrate the approach and report a 'free-lunch' phenomenon in which robustness improves both individual utilities and social welfare.

Significance. If the equivalence holds, the work supplies a computationally tractable extension of classical LQ game theory that directly incorporates strategic robustness without requiring new fixed-point machinery. The explicit transformation to fictitious-adversary games, the Markovian-linear characterization, and the Riccati recursion constitute a clean theoretical contribution; the numerical illustrations of performance gains without apparent cost are noteworthy and invite further empirical study.

minor comments (3)

§3, Definition 2: the precise penalty function on the fictitious adversary (e.g., quadratic weighting on policy deviation) should be stated explicitly rather than described only in prose, to make the transformation fully reproducible from the text alone.
§4.2, Algorithm 1: the initialization and termination criteria for the coupled Riccati iteration are not specified; adding a short pseudocode block or explicit stopping tolerance would improve clarity.
Figure 3 caption: the network topology and the precise uncertainty levels used in the 'free-lunch' experiment should be stated in the caption or a table so that the result can be replicated without consulting the main text.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive evaluation and recommendation to accept the manuscript. The referee's summary correctly captures the core contributions: the reformulation of strategically robust LQ dynamic games as standard LQ games against penalized fictitious adversaries, the resulting notion of strategically robust dynamic equilibrium, the proof of existence/uniqueness of Markovian linear equilibria via coupled Riccati equations, and the numerical illustrations including the free-lunch phenomenon.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper formalizes strategically robust LQ dynamic games and derives an exact rewriting as standard LQ games against a fictitious adversary penalized for policy deviation. This modeling transformation is presented explicitly and leads to a new equilibrium notion whose existence, uniqueness, and Markovian linear structure are obtained via standard coupled Riccati recursions under the usual stabilizability/detectability assumptions of LQ theory. No step in the equivalence, equilibrium definition, or backward recursion reduces to a fitted parameter, self-referential definition, or unverified self-citation; the numerical examples are offered only as illustrations of the free-lunch phenomenon and do not enter the derivation. The central claims therefore remain independent of their own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

The paper rests on standard existence results for Riccati equations in LQ games and introduces new modeling constructs to represent strategic robustness; no fitted numerical parameters are mentioned.

axioms (2)

standard math Standard stabilizability and detectability conditions hold so that the coupled Riccati equations admit unique positive-definite solutions
Invoked to guarantee existence and uniqueness of the equilibrium policies.
domain assumption Uncertainty about other players' policies can be represented by a fictitious adversary penalized for deviation from those policies
Core modeling step that converts the robust game into a standard LQ game.

invented entities (2)

Strategically robust dynamic equilibrium no independent evidence
purpose: Equilibrium concept that incorporates strategic robustness to policy uncertainty
New definition introduced to characterize solutions of the transformed game.
Fictitious adversary no independent evidence
purpose: Device to encode uncertainty about other players' policies inside a standard LQ game
Invented modeling construct that enables the reduction to known theory.

pith-pipeline@v0.9.0 · 5495 in / 1632 out tokens · 82993 ms · 2026-05-08T11:17:59.817272+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

17 extracted references · 5 canonical work pages

[1]

Efficient iterative linear-quadratic approximations for non- linear multi-player general-sum differential games,

D. Fridovich-Keil, E. Ratner, L. Peters, A. D. Dragan, and C. J. Tomlin, “Efficient iterative linear-quadratic approximations for non- linear multi-player general-sum differential games,” inInternational Conference on Robotics and Automation. IEEE, 2020, pp. 1475– 1481

2020
[2]

Linear quadratic differential game approach for attitude takeover control of failed spacecraft,

Y . Chai, J. Luo, N. Han, and J. Xie, “Linear quadratic differential game approach for attitude takeover control of failed spacecraft,”Acta Astronautica, vol. 175, pp. 142–154, 2020

2020
[3]

Optimal economic stabilization policies under decen- tralized control and conflicting objectives,

R. Pindyck, “Optimal economic stabilization policies under decen- tralized control and conflicting objectives,”IEEE Transactions on Automatic Control, vol. 22, no. 4, pp. 517–530, 2003

2003
[4]

Ba¸ sar and G

T. Ba¸ sar and G. J. Olsder,Dynamic noncooperative game theory. SIAM, 1998

1998
[5]

Unraveling in guessing games: An experimental study,

R. Nagel, “Unraveling in guessing games: An experimental study,” The American Economic Review, vol. 85, no. 5, pp. 1313–1326, 1995

1995
[6]

Ba¸ sar and P

T. Ba¸ sar and P. Bernhard,H-infinity optimal control and related minimax design problems: a dynamic game approach. Springer Science & Business Media, 2008

2008
[7]

Guaranteeing cost strategies for linear quadratic differential games under uncertain dynamics,

F. Amato, M. Mattei, and A. Pironti, “Guaranteeing cost strategies for linear quadratic differential games under uncertain dynamics,” Automatica, vol. 38, no. 3, pp. 507–515, 2002

2002
[8]

ε-equilibrium in LQ differential games with bounded uncertain disturbances: robustness of standard strategies and new strategies with adaptation,

M. Jimenez and A. Poznyak, “ε-equilibrium in LQ differential games with bounded uncertain disturbances: robustness of standard strategies and new strategies with adaptation,”International Journal of Control, vol. 79, no. 07, pp. 786–797, 2006

2006
[9]

On data-driven Wasserstein distributionally robust Nash equilibrium problems with heterogeneous uncertainty,

G. Pantazis, B. Franci, and S. Grammatico, “On data-driven Wasser- stein distributionally robust Nash equilibrium problems with hetero- geneous uncertainty,”arXiv preprint arXiv:2312.03573, 2023

work page arXiv 2023
[10]

Tractable multi-agent rein- forcement learning through behavioral economics,

E. Mazumdar, K. Panaganti, and L. Shi, “Tractable multi-agent rein- forcement learning through behavioral economics,” inThe Thirteenth International Conference on Learning Representations, 2025

2025
[11]

Lanzetti, S

N. Lanzetti, S. Fricker, S. Bolognani, F. Dörfler, and D. Paccagnan, “Strategically robust game theory via optimal transport,”arXiv preprint arXiv:2507.15325, 2025

work page arXiv 2025
[12]

Provably convergent actor-critic in risk- averse MARL,

Y . Zhang and E. Mazumdar, “Provably convergent actor-critic in risk- averse MARL,”arXiv preprint arXiv:2602.12386, 2026

work page arXiv 2026
[13]

Convergent Q-learning for infinite-horizon general-sum markov games through behavioral economics,

——, “Convergent Q-learning for infinite-horizon general-sum markov games through behavioral economics,” in2025 IEEE 64th Conference on Decision and Control (CDC). IEEE, 2025, pp. 5899–5904

2025
[14]

Training general- izable collaborative agents via strategic risk aversion,

C. Qu, Y . Zhang, N. Lanzetti, and E. Mazumdar, “Training general- izable collaborative agents via strategic risk aversion,”arXiv preprint arXiv:2602.21515, 2026

work page arXiv 2026
[15]

Optimal modified feedback strategies in LQ games under control imperfections,

M. Rabbani, N. Mojahed, and S. Nazari, “Optimal modified feedback strategies in LQ games under control imperfections,”arXiv preprint arXiv:2503.19200, 2025

work page arXiv 2025
[16]

Block diagonally dominant matrices and generalizations of the Gerschgorin circle theorem,

D. G. Feingold and R. S. Varga, “Block diagonally dominant matrices and generalizations of the Gerschgorin circle theorem,”Pacific Journal of Mathematics, vol. 12, no. 4, pp. 1241–1250, 1962

1962
[17]

Controllability metrics, limitations and algorithms for complex networks,

F. Pasqualetti, S. Zampieri, and F. Bullo, “Controllability metrics, limitations and algorithms for complex networks,”IEEE Transactions on Control of Network Systems, vol. 1, no. 1, pp. 40–52, 2014

2014

[1] [1]

Efficient iterative linear-quadratic approximations for non- linear multi-player general-sum differential games,

D. Fridovich-Keil, E. Ratner, L. Peters, A. D. Dragan, and C. J. Tomlin, “Efficient iterative linear-quadratic approximations for non- linear multi-player general-sum differential games,” inInternational Conference on Robotics and Automation. IEEE, 2020, pp. 1475– 1481

2020

[2] [2]

Linear quadratic differential game approach for attitude takeover control of failed spacecraft,

Y . Chai, J. Luo, N. Han, and J. Xie, “Linear quadratic differential game approach for attitude takeover control of failed spacecraft,”Acta Astronautica, vol. 175, pp. 142–154, 2020

2020

[3] [3]

Optimal economic stabilization policies under decen- tralized control and conflicting objectives,

R. Pindyck, “Optimal economic stabilization policies under decen- tralized control and conflicting objectives,”IEEE Transactions on Automatic Control, vol. 22, no. 4, pp. 517–530, 2003

2003

[4] [4]

Ba¸ sar and G

T. Ba¸ sar and G. J. Olsder,Dynamic noncooperative game theory. SIAM, 1998

1998

[5] [5]

Unraveling in guessing games: An experimental study,

R. Nagel, “Unraveling in guessing games: An experimental study,” The American Economic Review, vol. 85, no. 5, pp. 1313–1326, 1995

1995

[6] [6]

Ba¸ sar and P

T. Ba¸ sar and P. Bernhard,H-infinity optimal control and related minimax design problems: a dynamic game approach. Springer Science & Business Media, 2008

2008

[7] [7]

Guaranteeing cost strategies for linear quadratic differential games under uncertain dynamics,

F. Amato, M. Mattei, and A. Pironti, “Guaranteeing cost strategies for linear quadratic differential games under uncertain dynamics,” Automatica, vol. 38, no. 3, pp. 507–515, 2002

2002

[8] [8]

ε-equilibrium in LQ differential games with bounded uncertain disturbances: robustness of standard strategies and new strategies with adaptation,

M. Jimenez and A. Poznyak, “ε-equilibrium in LQ differential games with bounded uncertain disturbances: robustness of standard strategies and new strategies with adaptation,”International Journal of Control, vol. 79, no. 07, pp. 786–797, 2006

2006

[9] [9]

On data-driven Wasserstein distributionally robust Nash equilibrium problems with heterogeneous uncertainty,

G. Pantazis, B. Franci, and S. Grammatico, “On data-driven Wasser- stein distributionally robust Nash equilibrium problems with hetero- geneous uncertainty,”arXiv preprint arXiv:2312.03573, 2023

work page arXiv 2023

[10] [10]

Tractable multi-agent rein- forcement learning through behavioral economics,

E. Mazumdar, K. Panaganti, and L. Shi, “Tractable multi-agent rein- forcement learning through behavioral economics,” inThe Thirteenth International Conference on Learning Representations, 2025

2025

[11] [11]

Lanzetti, S

N. Lanzetti, S. Fricker, S. Bolognani, F. Dörfler, and D. Paccagnan, “Strategically robust game theory via optimal transport,”arXiv preprint arXiv:2507.15325, 2025

work page arXiv 2025

[12] [12]

Provably convergent actor-critic in risk- averse MARL,

Y . Zhang and E. Mazumdar, “Provably convergent actor-critic in risk- averse MARL,”arXiv preprint arXiv:2602.12386, 2026

work page arXiv 2026

[13] [13]

Convergent Q-learning for infinite-horizon general-sum markov games through behavioral economics,

——, “Convergent Q-learning for infinite-horizon general-sum markov games through behavioral economics,” in2025 IEEE 64th Conference on Decision and Control (CDC). IEEE, 2025, pp. 5899–5904

2025

[14] [14]

Training general- izable collaborative agents via strategic risk aversion,

C. Qu, Y . Zhang, N. Lanzetti, and E. Mazumdar, “Training general- izable collaborative agents via strategic risk aversion,”arXiv preprint arXiv:2602.21515, 2026

work page arXiv 2026

[15] [15]

Optimal modified feedback strategies in LQ games under control imperfections,

M. Rabbani, N. Mojahed, and S. Nazari, “Optimal modified feedback strategies in LQ games under control imperfections,”arXiv preprint arXiv:2503.19200, 2025

work page arXiv 2025

[16] [16]

Block diagonally dominant matrices and generalizations of the Gerschgorin circle theorem,

D. G. Feingold and R. S. Varga, “Block diagonally dominant matrices and generalizations of the Gerschgorin circle theorem,”Pacific Journal of Mathematics, vol. 12, no. 4, pp. 1241–1250, 1962

1962

[17] [17]

Controllability metrics, limitations and algorithms for complex networks,

F. Pasqualetti, S. Zampieri, and F. Bullo, “Controllability metrics, limitations and algorithms for complex networks,”IEEE Transactions on Control of Network Systems, vol. 1, no. 1, pp. 40–52, 2014

2014