Linear-quadratic stochastic nonzero-sum differential games between graphon teams

De-xuan Xu; Nan-jing Huang; Zhun Gou

REVIEW 3 major objections 5 minor 1 cited by

Linear-quadratic stochastic nonzero-sum differential games between graphon teams

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper establishes that a nonzero-sum stochastic game between two graphon-interacting teams has a feedback Nash equilibrium whenever a coupled pair of operator-valued Riccati equations admits a solution, and proves such solutions…

desk verdict Novel two-team graphon game setting, but the main existence theorem currently rests on an unverified equivalence between the coupled Riccati system and the operator equation that is actually solved. read the letter →

arxiv 2506.11468 v1 pith:I245CQZH submitted 2025-06-13 math.OC

De-xuan Xu , Zhun Gou , Nan-jing Huang This is my paper · ORCID

classification math.OC MSC 91A2391A1593E2049N1060H15

keywords stochasticdifferentialgamesgraphonNashequilibriumoperator-valuedRiccatiequationsdynamicprogramminginfinite-dimensionalcontrollinear-quadraticnonzero-sum

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

The paper studies a nonzero-sum stochastic differential game in which two large teams of agents, each internally coupled through graphons (symmetric kernels on $[0,1]^2$ that encode interaction strengths), choose controls to minimize their own social costs. It establishes that the continuum team game is exactly an infinite-dimensional two-agent game, and then derives a feedback Nash equilibrium in the form $\bar u_i(t)=-(2/R_{ii})B_i^*\Pi_i(t)x_t$, where the operators $\Pi_1,\Pi_2$ solve a coupled operator-valued Riccati system. The main existence theorem states that this Riccati system has a solution on the whole time horizon whenever the inter-team coupling parameter is sufficiently small, and a local solution for full coupling. A sympathetic reader should care because this gives a concrete path from heterogeneous many-agent team competition to a small set of operator equations whose solutions directly yield equilibrium strategies and costs.

What carries the argument

The central machinery is the graphon operator $(T_M\varphi)(\alpha)=\int_0^1 M(\alpha,\beta)\varphi(\beta)d\beta$, which is self-adjoint and compact on $L^2[0,1]$. It converts the continuum dynamics (1) into the Hilbert-space stochastic evolution (4) on $H^2=L^2[0,1]^2$, making the two-team game equivalent to an infinite-dimensional two-agent linear-quadratic game. Dynamic programming then transforms the Nash conditions into the coupled operator Riccati equations (10); the existence proof rewrites these as the single nonlinear operator equation (12) in the Banach space $D$ of block-diagonal self-adjoint operators, proves local solvability by a contraction argument, and uses a perturbation theorem from the decoupled case $\epsilon=0$ to extend to small $\epsilon$.

What would settle it

Take two non-identical graphons with non-commuting operators, choose a nonzero but small $\epsilon$, and solve the coupled Riccati system (10) on $[0,T]$ after projecting onto a finite basis of $L^2[0,1]$. If the smallness hypotheses of Theorem 3.2 hold and yet no finite-dimensional solution reaches $t=0$, the perturbation theorem is wrong; if a solution does reach $t=0$, compare the implied feedback laws against the original continuum dynamics to test the Nash property and the missing well-posedness step.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that Problem 1.1 admits a feedback Nash equilibrium for two graphon-interacting teams whenever the coupled operator-valued Riccati-type equations (10) have a solution. The paper proves such a solution exists on all of $[0,T]$ for every sufficiently small coupling $|\epsilon|$ (Theorem 3.2) and on a short interval near the terminal time even when $\epsilon=1$ (Theorem 3.1). At $\epsilon=0$ the coupled system splits into two independent operator Riccati equations, one per team, and the perturbation theorem Theorem 3.3 carries this solution to nonzero small $\epsilon$. Thus the paper's two-team Nash game is solved by reducing it to a solvable pair of operator equations.

Load-bearing premise

The derivation depends on the unproved assertion that the infinite-dimensional stochastic evolution (4) admits a unique strong solution and that the Itô formula used in the verification argument is valid; the paper refers to a theorem [47] for this, but that reference is absent from the bibliography.

Editorial extensions

If this is right

For $|\epsilon|$ small enough, Theorem 3.2 gives an equilibrium in feedback form, so each team's optimal control at time $t$ depends only on the current aggregate state through the operator $\Pi_i(t)$.
The value functions are quadratic, $J_i=\langle\Pi_i(0)x_0,x_0\rangle_{H^2}+q_i(0)$, so the equilibrium payoffs are computable from the same operator pair.
At $\epsilon=0$ the coupled system decouples into two independent operator Riccati equations, recovering single-population graphon linear-quadratic control as the zero-coupling limit.
Theorem 3.1 provides a local-in-time solution near $T$ even at full coupling $\epsilon=1$, so the feedback-equilibrium construction is not restricted to weak inter-team coupling near the terminal time.

Reading between the lines

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

An extension the authors leave implicit is spectral truncation: because graphon operators are compact, expanding states in their eigenbases turns the coupled operator equations (10) into a finite system of Riccati ODEs, giving a numerical route to test the theorem.
The same perturbation argument should extend to small heterogeneity in the graphons themselves, not only small cross-team coupling, since the contraction estimates depend on operator norms rather than on the specific kernels.
The missing well-posedness lemma is the main bridge between the formal Riccati derivation and a fully rigorous theorem; supplying a proof of unique mild solutions of (4) and the required Itô formula interchange would complete the paper's program.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The setting is new: continuous-time nonzero-sum LQ game between two graphon teams, solved by dynamic programming into coupled operator Riccati equations, with existence via a small-epsilon perturbation argument. The paper is honest about what it does, cites the relevant discrete-time and zero-sum precedents, and proves the perturbation theorem (Thm 3.3) in full detail. Credit is due for the formal setup and for isolating the right base case at epsilon = 0.

The soft spot is load-bearing. The Nash equilibrium in Theorem 2.1 is defined by the coupled system (10), but the existence proof, Theorem 3.2, proves existence only for the single operator equation (12). The paper says "Clearly, we can check" and gives block definitions that are never reconciled with the products in (12). On the natural reading of the definitions, the quadratic terms do not match: the (1,1) block at epsilon = 0 contains -Π1 S1 Π1 - Π1 S2 Π1 - Π2 S2 Π1, while (10) requires -Π1 S1 Π1 - Π1 S2 Π2 - Π2 S2 Π1. These agree only under an extra condition like S2 Π1 = S2 Π2. If the intended block structure is different, it is not stated. Since Theorem 3.2 is the only existence argument for (10), the main theorem is currently unproved. This is fixable if the reduction can be made correct, but it is not a cosmetic issue.

Other gaps: well-posedness of the Hilbert SDE (4) cites [47], which is not in the bibliography; Theorem 3.1's proof is omitted, though it is local and not used for the main small-epsilon claim. These are minor beside the equivalence problem.

Bottom line: the problem is worth solving and the machinery is standard. As it stands, the paper needs a correct, explicit derivation of (12) from (10) before the main result can be trusted. I would send it to a referee—the setting is novel and the gap is narrow—but the referee should be asked to verify the block calculation and the reduction step.

Referee Report

3 major / 5 minor

Summary. The paper studies a finite-horizon linear-quadratic nonzero-sum stochastic differential game between two teams of interacting agents, with intrateam interactions described by graphon operators. The authors reformulate the two-team problem as an infinite-dimensional two-player game, derive coupled operator-valued Riccati-type equations (10) from a dynamic programming verification theorem, and then prove small-coupling existence of solutions to a single operator Riccati equation (12). From this they conclude the existence of a feedback Nash equilibrium for the original game. The main novelty is the continuous-time two-team graphon setting and the perturbation argument for the coupled Riccati system.

Significance. If the connection between the coupled system (10) and the single operator equation (12) is made fully rigorous, the paper is a meaningful contribution to graphon games and infinite-dimensional stochastic differential games: it is, to the authors' knowledge, the first continuous-time treatment of nonzero-sum team-against-team games with heterogeneous intrateam interactions. The Banach-space perturbation theorem (Theorem 3.3) is proved in detail, the epsilon=0 base case is constructed from classical Riccati theory rather than assumed, and no parameters are fitted, so the existence argument is not circular. The contribution is moderately significant and fits the journal's scope, but the current presentation leaves the central reduction between the two Riccati formulations unverified.

major comments (3)

[Section 3, Eq. (12) and preceding displayed definitions] The reduction from the coupled system (10) to the single operator equation (12) is asserted with "Clearly, we can check" but is never shown. The displayed definitions are not self-consistent as written: K_epsilon, S, S0, I_{2x2}, J, Qbar, and G are written as column vectors, and as written they do not define elements of L(H4) without an implicit diagonal convention. If the intended convention is that a column (X;Y) denotes the block-diagonal operator diag(X,Y) and that J is the swap on H2 oplus H2, then a direct block computation does reproduce the quadratic and epsilon-coupling terms of (10); in particular the (1,1)-block at epsilon=0 is -Pi1 S1 Pi1 - Pi1 S2 Pi2 - Pi2 S2 Pi1, which matches the first line of (10). However, this convention is never stated, and no derivation of the equivalence is given. Since Theorem 3.2 is proved only for (12) while Theorem 2.1 is stated for (10), the main existence claim depends on this unstated and unverified reduction. Please state the block conventions explicitly and provide the full verification of the equivalence.
[Section 2, after Eq. (4)] The existence and uniqueness of a strong (mild) solution to the stochastic evolution equation (4) is asserted by reference to "[47, Th.3.14]", but no reference [47] appears in the bibliography. The Itô-formula computations in (5)-(7) and the verification theorem depend on the well-posedness of the state equation. Please supply a correct citation with the precise hypotheses, or include a short proof adapted to the present setting.
[Section 3, Theorem 3.1] Theorem 3.1 is stated as a local existence result for (12) at epsilon=1, but its proof is omitted with the sentence "The proof is similar to the one of [19, Th 2.1] and so we omit it here." Since this is one of the paper's stated existence results, the reader cannot check whether the cited theorem applies to the operator Riccati equation. If Theorem 3.1 is not needed for the main small-coupling argument, this should be stated explicitly and the claim should be proved or moved to a remark; otherwise a full proof is required.

minor comments (5)

[Section 2, Eq. (4)] The noise coefficient Sigma is displayed as a column (sigma1 I; sigma2 I), but in (4) it must act as the block-diagonal operator diag(sigma1 I, sigma2 I) on the H2-valued Wiener process; please clarify the intended block-matrix convention.
[Section 3, Lemma 3.1 and Theorem 3.3] The symbol D is used both for the Banach space of block-diagonal operators in Lemma 3.1 and for the solution domain in Theorem 3.3; please rename one of them to avoid ambiguity.
[Section 2, Eq. (5)] The expectation of the Itô integral is dropped without comment; please add a remark on the integrability of the stochastic integral or a localization argument to justify the martingale property.
[Proposition 2.1, conditions (iii)-(iv)] The conditions are written as "0 = ... <= ...", which conflates the HJB equality at the equilibrium strategy with the inequality for arbitrary controls; separating these two statements would improve readability.
[Introduction and metadata] There is a typo "hetergeneous" that should be "heterogeneous", and the 2020 Mathematics Subject Classification field is left blank.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: equilibrium is derived from Riccati existence that itself rests on external baseline [20] and a proved perturbation theorem; only unproved-equivalence/omitted-proof gaps remain, which are correctness risks, not circular dependence.

full rationale

No significant circularity. The Nash equilibrium formula in Theorem 2.1 is obtained by solving the Bellman/verification inequalities in Proposition 2.1; its load-bearing input is the existence of coupled Riccati solutions (10). That existence is addressed independently of any fitted data: the epsilon=0 baseline is taken from the external monograph [20], the perturbation-in-epsilon argument (Theorem 3.3) is proved in the paper from external textbook results [18,19], and the final solution is then inserted back into the verification theorem. The only self-reference is [15], a background citation for the graphon modeling framework; it does not supply the uniqueness/existence result on which the equilibrium theorem rests, so it is not load-bearing. The manuscript does however contain non-circular completeness/gap risks: the asserted equivalence between (10) and the vectorized equation (12) is stated as 'Clearly, we can check' without a derivation, the block definitions of K_epsilon, S, J, etc. are not reconciled with the products in (12), Theorem 3.1's proof is omitted, and the well-posedness citation [47] for (4) is absent from the bibliography. These are correctness and reproducibility concerns, not instances where a prediction reduces to its own inputs.

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

The central claim rests on standard infinite-dimensional control results plus two paper-specific unproved assertions: well-posedness of the infinite-dimensional SDE and the equivalence of the Riccati system to a single operator Riccati equation. No free parameters are fitted, and no new physical or mathematical entities are introduced.

assumptions (4)

domain assumption Well-posedness of the Hilbert-space SDE (4): existence and uniqueness of a strong (mild) solution.
Invoked when rewriting the graphon game as an infinite-dimensional two-agent game and in the It\^o formula in Proposition 2.1; justified only by a missing citation [47].
standard math Existence and uniqueness of solutions to the scalar operator Riccati equations for \epsilon=0.
Cited from [20, Part IV, Chapter 1, Th 2.1] and used to construct the base solution in Theorem 3.2.
standard math Local existence and continuation theorems for ODEs in Banach spaces ([19, Th 2.1, 2.2]).
Used for Theorem 3.1 and Theorem 3.3; not proved in the paper.
ad hoc to paper The coupled Riccati system (10) is equivalent to the operator Riccati equation (12) on the space D.
Stated as 'Clearly, we can check' with no derivation. The existence theorems target (12), so this equivalence transfers their conclusions to (10).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Linear-quadratic stochastic nonzero-sum differential games between graphon teams." pith.science (2026). https://pith.science/paper/I245CQZH

@misc{pith2026250611468,
  author       = {Pith},
  title        = {Pith review of: Linear-quadratic stochastic nonzero-sum differential games between graphon teams},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I245CQZH}},
  note         = {Machine review of arXiv:2506.11468}
}

read the original abstract

We study a class of nonzero-sum stochastic differential games between two teams with agents in each team interacting through graphon aggregates. On the one hand, in each large population group, agents act together to optimize a common social cost function. On the other hand, these two groups compete with each other, forming a Nash game between two graphon teams. We note that the original problem can be equivalently formulated as an infinite-dimensional two-agent Nash game. Applying the dynamic programming approach, we obtain a set of coupled operator-valued Riccati-type equations. By proving the existence of solutions to the equations mentioned above, we obtain a Nash equilibrium for the two teams.

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

Infinite-Dimensional LQ Mean Field Games with Common Noise: Small and Arbitrary Finite Time Horizons
math.OC 2026-01 conditional novelty 6.0 of 10

Infinite-dimensional linear-quadratic mean field games with common noise have unique equilibria for small time horizons and, under deterministic common-noise diffusion, for arbitrary finite time horizons.

Reference graph

Works this paper leans on

20 extracted references · 18 canonical work pages · cited by 1 Pith paper

[1]

Medvedev, The Nonlinear Heat Equation on W-Random Graphs, Archive for Rational Mechanics and Analysis, 212:781-803, 2014

G.S. Medvedev, The Nonlinear Heat Equation on W-Random Graphs, Archive for Rational Mechanics and Analysis, 212:781-803, 2014

work page 2014
[2]

Medvedev, The continuum limit of the Kuramoto model on sparse random graphs, Communications in Mathematical Sciences, 17(4):883-898, 2019

G.S. Medvedev, The continuum limit of the Kuramoto model on sparse random graphs, Communications in Mathematical Sciences, 17(4):883-898, 2019

work page 2019
[3]

Kaliuzhnyi-Verbovetskyi and G.S

D. Kaliuzhnyi-Verbovetskyi and G.S. Medvedev, Sparse Monte Carlo method for nonlocal diffusion problems, SIAM Journal on Numerical Analysis, 60(6):3001-3028, 2022

work page 2022
[4]

Medvedev and G

G.S. Medvedev and G. Simpson, A numerical method for a nonlocal diffusion equation with additive noise, Stochastics and Partial Differential Equations: Analysis and Computations, 11(4):1433-1469, 2023

work page 2023
[5]

Sanjari, N

S. Sanjari, N. Saldi and S. Y¨ uksel, Nash equilibria for exchangeable team-against-team games, their mean-field limit, and the role of common randomness, SIAM Journal on Control and Optimization, 62(3):1437-1464, 2024

work page 2024
[6]

uz Zaman, A

M.A. uz Zaman, A. Koppel, M. Lauri` ere and T. Ba¸ sar, Independent RL for cooperative-competitive agents: A mean-field perspective, arXiv preprint arXiv:2403.11345, 2024

arXiv 2024
[7]

Carmona, K

R. Carmona, K. Hamidouche, M. Lauri` ere and Z. Tan, Policy optimization for linear-quadratic zero-sum mean-field type games, In 2020 59th IEEE Conference on Decision and Control (CDC) (pp. 1038-1043). IEEE, 2020, December

work page 2020
[8]

Coppini, A note on Fokker-Planck equations and graphons, Journal of Statistical Physics, 187(2):15, 2022

F. Coppini, A note on Fokker-Planck equations and graphons, Journal of Statistical Physics, 187(2):15, 2022

work page 2022

Show all 20 references

[9]

Fouque and Z

J.-P. Fouque and Z. Zhang, Mean field game with delay: a toy model, Risks, 6(3):90, 2018

2018
[10]

Liu and D

H. Liu and D. Firoozi, Hilbert space-valued LQ mean field games: An infinite-dimensional analysis, arXiv:2403.01012, 2024

2024 arXiv
[11]

Ghilli and M

D. Ghilli and M. Ricciardi, Linear quadratic Nash systems and master equations in Hilbert spaces, arXiv preprint arXiv:2504.00637, 2025

2025 arXiv
[12]

Federico, F

S. Federico, F. Gozzi and A. ´Swiech, On mean field games in infinite dimension, arXiv preprint arXiv:2411.14604, 2024

2024 arXiv
[13]

Gao and P.E

S. Gao and P.E. Caines, Graphon control of large-scale networks of linear systems, IEEE Transactions on Automatic Control, 65(10):4090-4105, 2019

2019
[14]

Gao and P.E

S. Gao and P.E. Caines, Spectral representations of graphons in very large network systems control, In 2019 IEEE 58th conference on decision and Control (CDC) (pp. 5068-5075). IEEE, 2019

2019
[15]

D.X. Xu, Z. Gou, N.J. Huang, S. Gao, Linear-quadratic graphon mean field games with common noise, arXiv preprint, arXiv:2401.09030, 2024

2024 arXiv
[16]

Lov´ asz, Large Networks and Graph Limits, vol

L. Lov´ asz, Large Networks and Graph Limits, vol. 60, American Mathematical Society, Providence, RI, 2012. 12

2012
[17]

Ichikawa, Dynamic programming approach to stochastic evolution equations, SIAM Journal on Con- trol and Optimization, 17:152-174, 1979

A. Ichikawa, Dynamic programming approach to stochastic evolution equations, SIAM Journal on Con- trol and Optimization, 17:152-174, 1979

1979
[18]

Tikhonov, A.B

A.N. Tikhonov, A.B. Vasil’eva and A.G. Sveshnikov, Differential equations, Springer-Verlag, New York, New York, 1985

1985
[19]

Guo, Nonlinear functional analysis, 2nd ed, (in Chinese), Ji’nan: Shandong Science and Technology Press, 2001

D.J. Guo, Nonlinear functional analysis, 2nd ed, (in Chinese), Ji’nan: Shandong Science and Technology Press, 2001

2001
[20]

Bensoussan, G

A. Bensoussan, G. Da Prato, M.C. Delfour and S.K. Mitter, Representation and control of infinite- dimensional systems, Vol. II. Systems & Control: Foundations & Applications, Birkh¨ auser Boston, Inc., Boston, MA, 1993. 13

1993

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Medvedev, The Nonlinear Heat Equation on W-Random Graphs, Archive for Rational Mechanics and Analysis, 212:781-803, 2014

G.S. Medvedev, The Nonlinear Heat Equation on W-Random Graphs, Archive for Rational Mechanics and Analysis, 212:781-803, 2014

work page 2014

[2] [2]

Medvedev, The continuum limit of the Kuramoto model on sparse random graphs, Communications in Mathematical Sciences, 17(4):883-898, 2019

G.S. Medvedev, The continuum limit of the Kuramoto model on sparse random graphs, Communications in Mathematical Sciences, 17(4):883-898, 2019

work page 2019

[3] [3]

Kaliuzhnyi-Verbovetskyi and G.S

D. Kaliuzhnyi-Verbovetskyi and G.S. Medvedev, Sparse Monte Carlo method for nonlocal diffusion problems, SIAM Journal on Numerical Analysis, 60(6):3001-3028, 2022

work page 2022

[4] [4]

Medvedev and G

G.S. Medvedev and G. Simpson, A numerical method for a nonlocal diffusion equation with additive noise, Stochastics and Partial Differential Equations: Analysis and Computations, 11(4):1433-1469, 2023

work page 2023

[5] [5]

Sanjari, N

S. Sanjari, N. Saldi and S. Y¨ uksel, Nash equilibria for exchangeable team-against-team games, their mean-field limit, and the role of common randomness, SIAM Journal on Control and Optimization, 62(3):1437-1464, 2024

work page 2024

[6] [6]

uz Zaman, A

M.A. uz Zaman, A. Koppel, M. Lauri` ere and T. Ba¸ sar, Independent RL for cooperative-competitive agents: A mean-field perspective, arXiv preprint arXiv:2403.11345, 2024

arXiv 2024

[7] [7]

Carmona, K

R. Carmona, K. Hamidouche, M. Lauri` ere and Z. Tan, Policy optimization for linear-quadratic zero-sum mean-field type games, In 2020 59th IEEE Conference on Decision and Control (CDC) (pp. 1038-1043). IEEE, 2020, December

work page 2020

[8] [8]

Coppini, A note on Fokker-Planck equations and graphons, Journal of Statistical Physics, 187(2):15, 2022

F. Coppini, A note on Fokker-Planck equations and graphons, Journal of Statistical Physics, 187(2):15, 2022

work page 2022