pith. sign in

arxiv: 2604.05294 · v2 · submitted 2026-04-07 · 🧮 math.OC

Mean Field Games and Control on Large Expander Graphs

Tao Zhang , Peter E. Caines This is my paper

Pith reviewed 2026-05-10 19:55 UTC · model grok-4.3

classification 🧮 math.OC
keywords mean field gamesexpander graphsgraphexon measuresaveraging operatorsTuring instabilitynetwork stabilitylinear quadratic controlsparse networks
0
0 comments X

The pith

Large expander graphs allow mean field games to be formulated in a continuous limit with stability conditions for spatial instabilities.

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

The paper proves that empirical graphexon measures on finite large expander graphs converge weakly to a continuous limit measure, and that the corresponding discrete averaging operators converge strongly to a continuous operator. This convergence justifies setting up a linear-quadratic mean field game in which agents are distinguished by their position in a continuous space and interact only through a neighborhood average. The analysis then derives algebraic conditions guaranteeing global asymptotic stability of the resulting closed-loop system and locates parameter thresholds at which a Turing-type topological instability appears, keeping the overall mean stable while spatial deviations grow. Such results matter for anyone modeling strategic interactions or control on very large but sparse networks such as communication or social graphs.

Core claim

In the graphexon framework the sequence of empirical measures defined on finite expander graphs converges weakly to a limit graphexon measure on a continuous state space X, while the associated discrete averaging operators converge strongly to a continuous operator G. These limits permit the formulation of a linear-quadratic mean field game in which each agent is identified by a spatial label alpha in X and interacts exclusively with the neighborhood average produced by G. Algebraic conditions are then established for global asymptotic stability of the closed-loop system, together with explicit parameter thresholds that produce a Turing-type topological instability in which the homogeneous 4

What carries the argument

The graphexon framework that supplies the limit topology and the continuous averaging operator G that encodes neighborhood interactions for the mean-field game.

If this is right

  • The linear-quadratic mean field game admits a well-posed continuous formulation.
  • Global asymptotic stability of the closed-loop dynamics can be checked by algebraic tests on the system matrices and the spectrum of G.
  • Critical parameter values exist at which the homogeneous equilibrium loses stability in the spatial directions while remaining stable in the mean.
  • The same convergence properties support mean-field control problems on the same class of graphs.

Where Pith is reading between the lines

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

  • Analogous convergence results might be obtainable for other families of sparse graphs that admit a graphexon-type limit.
  • The instability thresholds could guide the design of network topologies that either suppress or encourage spatial pattern formation in agent states.
  • Computational methods developed for the continuous model may scale better than direct simulation on the original large discrete graphs.

Load-bearing premise

The large expander graphs possess limit topologies that are faithfully described by the graphexon framework, so that the discrete-to-continuous convergence of measures and operators holds.

What would settle it

A sequence of finite expander graphs for which the empirical graphexon measures fail to converge weakly or for which the closed-loop trajectories on the finite graphs do not approach the predicted continuous behavior as the graph size grows.

Figures

Figures reproduced from arXiv: 2604.05294 by Peter E. Caines, Tao Zhang.

Figure 1
Figure 1. Figure 1: (a) A Cayley Graph with two generators ( [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Spatial distribution of agent states on the [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
read the original abstract

This paper investigates mean field games and control on sparse networks. In the case of large expander graphs, the limit topologies are analyzed using the graphexon framework, which characterizes both dense network limits and sparse connections. We prove that the sequence of empirical graphexon measures defined on finite graphs converges weakly to a limit graphexon measure on a continuous state space. Furthermore, the associated sequence of discrete averaging operators converges strongly to a continuous operator. These properties enable the formulation of a linear-quadratic mean field game in which each agent is identified by a spatial network label $\alpha \in X$ and only interacts with the neighborhood average defined by the operator $\mathcal{G}$ characterized by large expander graphs. In Section 5, algebraic conditions for the global asymptotic stability of the closed-loop system are established. The analysis identifies parameter thresholds that gives rise to a Turing-type topological instability, where the homogeneous mean state remains stable while the spatial deviation field diverges over the continuous spectrum of the limit operator.

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 / 2 minor

Summary. The paper investigates mean field games and control on large expander graphs using the graphexon framework. It proves that empirical graphexon measures on finite graphs converge weakly to a limit graphexon measure on a continuous state space X, and that the associated discrete averaging operators converge strongly to a continuous operator G. These limits enable formulation of a linear-quadratic mean-field game in which agents labeled by spatial positions α ∈ X interact only through neighborhood averages defined by G. In Section 5, algebraic conditions are derived for global asymptotic stability of the closed-loop system, identifying parameter thresholds that produce a Turing-type topological instability in which the homogeneous mean state remains stable while spatial deviations grow according to the spectrum of the limit operator G.

Significance. If the convergence statements and the passage of stability thresholds to the limit hold, the work supplies a rigorous discrete-to-continuous bridge for mean-field games on sparse expander networks, extending the graphexon formalism to control problems and furnishing explicit algebraic criteria for a topological instability. The combination of weak measure convergence, strong operator convergence, and the resulting LQ-MFG stability analysis is a substantive contribution to networked control and mean-field theory on non-dense graphs.

major comments (2)
  1. [Section 5] §5 (Turing instability analysis): The algebraic stability conditions and the critical parameter thresholds for Turing-type instability are obtained from the spectrum of the limit operator G. The manuscript establishes only strong convergence of the discrete operators G_n to G. Strong convergence alone does not guarantee convergence of spectra or eigenvalues; without additional arguments (e.g., norm convergence of G_n, uniform resolvent bounds, or spectral gap preservation for expanders), the instability thresholds derived for the continuous system need not apply to the finite-graph systems whose behavior the paper ultimately seeks to describe.
  2. [Convergence theorems] Convergence theorems (presumably §3–4): The weak convergence of empirical graphexon measures and the strong convergence of averaging operators are load-bearing for the entire mean-field formulation. The paper should state explicit quantitative conditions on the expander sequence (expansion constant, degree growth, or n → ∞ rate) and supply error bounds or rates that justify passing the LQ-MFG and stability analysis to the limit; the abstract’s reference to “large expander graphs” is too vague for the claims that follow.
minor comments (2)
  1. [Abstract] Abstract, last sentence: grammatical error (“parameter thresholds that gives rise” should read “that give rise”).
  2. [Notation] Notation section: the graphexon measure and the precise definition of the averaging operator G should be introduced with a self-contained paragraph before the convergence statements, to aid readers unfamiliar with the graphexon literature.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback on our manuscript. The comments highlight important points regarding the passage of stability properties to the limit and the need for more precise statements on the expander assumptions. We address each major comment below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Section 5] §5 (Turing instability analysis): The algebraic stability conditions and the critical parameter thresholds for Turing-type instability are obtained from the spectrum of the limit operator G. The manuscript establishes only strong convergence of the discrete operators G_n to G. Strong convergence alone does not guarantee convergence of spectra or eigenvalues; without additional arguments (e.g., norm convergence of G_n, uniform resolvent bounds, or spectral gap preservation for expanders), the instability thresholds derived for the continuous system need not apply to the finite-graph systems whose behavior the paper ultimately seeks to describe.

    Authors: We agree that strong operator convergence alone does not guarantee spectral convergence in general, and the referee's observation is correct on this point. The stability thresholds in Section 5 are derived for the limiting continuous LQ-MFG, which the paper positions as the asymptotic description of the finite-graph systems. To strengthen the bridge, we will add a remark and supporting argument in the revised Section 5 that exploits the uniform spectral gap of the expander sequence to obtain convergence of the relevant eigenvalues (specifically, those determining the instability) in the Hausdorff sense. This uses the fact that the limit operator G inherits a spectral gap from the discrete expanders, yielding uniform resolvent bounds outside a neighborhood of the spectrum and allowing the algebraic stability conditions to pass to the limit for large n. We will revise the manuscript to include this clarification and argument. revision: yes

  2. Referee: [Convergence theorems] Convergence theorems (presumably §3–4): The weak convergence of empirical graphexon measures and the strong convergence of averaging operators are load-bearing for the entire mean-field formulation. The paper should state explicit quantitative conditions on the expander sequence (expansion constant, degree growth, or n → ∞ rate) and supply error bounds or rates that justify passing the LQ-MFG and stability analysis to the limit; the abstract’s reference to “large expander graphs” is too vague for the claims that follow.

    Authors: We accept that the assumptions on the expander sequence should be stated more explicitly to avoid vagueness. The manuscript relies on a sequence of d-regular expander graphs with expansion constant bounded below by a fixed positive number independent of n, with the graphexon limit existing under the standard conditions of the framework (degree growth sublinear in n). We will revise the abstract, introduction, and the statements of the convergence theorems in Sections 3 and 4 to list these conditions explicitly. Regarding error bounds or rates, the current proofs establish qualitative weak convergence of measures and strong convergence of operators without quantitative rates; deriving explicit rates would require additional quantitative tools (such as concentration bounds on the empirical measures) that are not developed here. We will add a remark acknowledging this and noting that the asymptotic results hold as n → ∞ under the stated assumptions, while the LQ-MFG formulation and stability analysis are justified in the limit. revision: partial

Circularity Check

0 steps flagged

No circularity: convergence proofs and algebraic conditions form an independent chain

full rationale

The paper establishes weak convergence of empirical graphexon measures and strong convergence of discrete averaging operators to a continuous limit operator via stated theorems. These are then used to formulate the LQ-MFG on the continuous space X with interaction via the limit operator G. Section 5 derives algebraic stability conditions and parameter thresholds for Turing-type instability directly from the spectrum of the limit operator. No quoted equations or claims reduce by construction to fitted inputs, self-definitions, or load-bearing self-citations; the derivation is presented as self-contained mathematical analysis rather than tautological renaming or parameter fitting.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper depends on the graphexon framework (presumably prior literature) to define limit topologies and on unstated technical assumptions that guarantee weak convergence of empirical measures and strong convergence of averaging operators on expander graphs. No explicit free parameters are named in the abstract, but the stability thresholds are algebraic conditions whose precise form is not supplied.

axioms (2)
  • domain assumption Large expander graphs admit a graphexon limit that simultaneously captures dense and sparse connections
    Invoked to justify the continuous-state formulation and operator convergence
  • domain assumption The sequence of discrete averaging operators converges strongly to a continuous operator G
    Required for the mean-field interaction term in the linear-quadratic game

pith-pipeline@v0.9.0 · 5464 in / 1517 out tokens · 35606 ms · 2026-05-10T19:55:36.239335+00:00 · methodology

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

36 extracted references · 36 canonical work pages

  1. [1]

    Jeux à champ moyen

    Jean-Michel Lasry and Pierre-Louis Lions. Jeux à champ moyen. I–le cas stationnaire.C. R. Math., 343(9):619–625, 2006

    work page 2006

  2. [2]

    Mean field games.Jpn

    Jean-Michel Lasry and Pierre-Louis Lions. Mean field games.Jpn. J. Math., 2(1):229–260, 2007

    work page 2007

  3. [3]

    Malhamé, and Peter E

    Minyi Huang, Roland P. Malhamé, and Peter E. Caines. Large population stochastic dynamic games: Closed-loop McKean-Vlasov systems and the Nash certainty equiva- lence principle.Commun. Inf. Syst., 6(3):221–252, 2006

    work page 2006

  4. [4]

    Caines and Minyi Huang

    Peter E. Caines and Minyi Huang. Graphon mean field games and their equations.SIAM J. Control Optim., 59(6): 4373–4399, 2021

    work page 2021

  5. [5]

    Tchuendom, and Peter E

    Shuang Gao, Rinel F. Tchuendom, and Peter E. Caines. Linear quadratic graphon field games.Commun. Inf. Syst., 21(3):341–369, 2021

    work page 2021

  6. [6]

    AMS, 2012

    László Lovász.Large Networks and Graph Limits, vol- ume 60. AMS, 2012

    work page 2012

  7. [7]

    Chayes, Henry Cohn, and Nina Holden

    Christian Borgs, Jennifer T. Chayes, Henry Cohn, and Nina Holden. Sparse exchangeable graphs and their limits via graphon processes.J. Mach. Learn. Res., 18(210):1–71, 2018

    work page 2018

  8. [8]

    Victor Veitch and Diana M. Roy. The class of random graphs arising from exchangeable random measures. arXiv preprint arXiv:1512.03099, 2015

    work page Pith review arXiv 2015

  9. [9]

    Recurrence of distri- butional limits of finite planar graphs.Electron

    Itai Benjamini and Oded Schramm. Recurrence of distri- butional limits of finite planar graphs.Electron. J. Probab., 6:1–13, 2001

    work page 2001

  10. [10]

    Local weak convergence for sparse networks of interacting pro- cesses.Ann

    Daniel Lacker, Kavita Ramanan, and Ruoyu Wu. Local weak convergence for sparse networks of interacting pro- cesses.Ann. Appl. Probab., 33(2):843–888, 2023

    work page 2023

  11. [11]

    Peter E. Caines. Embedded vertexon-graphons and embed- ded GMFG systems. InProc. 61st IEEE Conf. Decision and Control (CDC), pages 5550–5577. IEEE, 2022

    work page 2022

  12. [12]

    Caines and Minyi Huang

    Peter E. Caines and Minyi Huang. Sparse network mean field games: Ring structures and related topologies. In Proc. 63rd IEEE Conf. Decision and Control (CDC), pages 2584–2590, Milan, Italy, 2024

    work page 2024

  13. [13]

    Caines and Minyi Huang

    Peter E. Caines and Minyi Huang. Mean field games on sparse network limits: Laplexion dynamics and ϵ-Nash equilibria. InProc. 64th IEEE Conf. Decision and Control (CDC), pages 1299–1305, Rio de Janeiro, Brazil, 2025

    work page 2025

  14. [14]

    Caines, and Minyi Huang

    Tao Zhang, Peter E. Caines, and Minyi Huang. Mean field games on embedded anisotropic networks. Submitted to Applied Mathematics & Optimization (AMO), 2026. Preprint– MeanFieldGames andControl onLargeExpanderGraphs11

    work page 2026

  15. [15]

    Ex- pander graphs and their applications.Bull

    Shlomo Hoory, Nathan Linial, and Avi Wigderson. Ex- pander graphs and their applications.Bull. Amer. Math. Soc., 43(4):439–561, 2006

    work page 2006

  16. [16]

    Ramanujan graphs.Combinatorica, 8(3):261–277, 1988

    Alexander Lubotzky, Ralph Phillips, and Peter Sarnak. Ramanujan graphs.Combinatorica, 8(3):261–277, 1988

    work page 1988

  17. [17]

    Grigorii Aleksandrovich Margulis. Explicit group- theoretical constructions of combinatorial schemes and their application to the design of expanders and concentra- tors.Problemy peredachi informatsii, 24(1):51–60, 1988

    work page 1988

  18. [18]

    Ultrafast consensus in small-world networks

    Reza Olfati-Saber. Ultrafast consensus in small-world networks. InProceedings of the 2005, American Control Conference, 2005., pages 2371–2378. IEEE, 2005

    work page 2005

  19. [19]

    Akers and Balakrishnan Krishnamurthy

    Sheldon B. Akers and Balakrishnan Krishnamurthy. A group-theoretic model for symmetric interconnection net- works.IEEE transactions on Computers, 38(4):555–566, 2002

    work page 2002

  20. [20]

    Pastry: Scalable, de- centralized object location, and routing for large-scale peer- to-peer systems

    Antony Rowstron and Peter Druschel. Pastry: Scalable, de- centralized object location, and routing for large-scale peer- to-peer systems. InIFIP/ACM International Conference on Distributed Systems Platforms and Open Distributed Processing, pages 329–350. Springer, 2001

    work page 2001

  21. [21]

    Xpander: Towards optimal-performance dat- acenters

    Asaf Valadarsky, Gal Shahaf, Michael Dinitz, and Michael Schapira. Xpander: Towards optimal-performance dat- acenters. InProceedings of the 12th International on Conference on emerging Networking EXperiments and Technologies, pages 205–219, 2016

    work page 2016

  22. [22]

    Explicit constructions of linear size superconcentrators

    Ofer Gabber and Zvi Galil. Explicit constructions of linear size superconcentrators. InProc. 20th Annu. Symp. Found. Comput. Sci. (FOCS), pages 364–370. IEEE, 1979

    work page 1979

  23. [23]

    Expander graphs in pure and ap- plied mathematics.Bull

    Alexander Lubotzky. Expander graphs in pure and ap- plied mathematics.Bull. Amer. Math. Soc., 49(1):113–162, 2012

    work page 2012

  24. [24]

    Springer, 1994

    Alexander Lubotzky.Discrete Groups, Expanding Graphs and Invariant Measures, volume 125. Springer, 1994

    work page 1994

  25. [25]

    Groups that produce expander graphs.arXiv preprint arXiv:2511.16479, 2025

    Luca Sabatini. Groups that produce expander graphs.arXiv preprint arXiv:2511.16479, 2025

    work page arXiv 2025

  26. [26]

    Caines and Minyi Huang

    Peter E. Caines and Minyi Huang. Graphon mean field games and the GMFG equations: ϵ-Nash equilibria. In Proc. 58th IEEE Conf. Decision and Control (CDC), pages 286–292. IEEE, 2019

    work page 2019

  27. [27]

    Shuang Gao and Peter E. Caines. Graphon control of large- scale networks of linear systems.IEEE Trans. Autom. Control, 65(10):4090–4105, 2019

    work page 2019

  28. [28]

    Perturbation analysis with Q-noise in LQG graphon mean field games

    Tao Zhang, Peter E Caines, and Shuang Gao. Perturbation analysis with Q-noise in LQG graphon mean field games. InProc. 64th IEEE Conf. Decision and Control (CDC), pages 2218–2225. IEEE, 2025

    work page 2025

  29. [29]

    Zimmer.Ergodic Theory and Semisimple Groups

    Robert J. Zimmer.Ergodic Theory and Semisimple Groups. Springer, 2013

    work page 2013

  30. [30]

    Conway.A Course in Functional Analysis

    John B. Conway.A Course in Functional Analysis. Springer, 2019

    work page 2019

  31. [31]

    Ivan N. Sanov. A property of a representation of a free group. InDokl. Akad. Nauk SSSR, volume 57, pages 657– 659, 1947

    work page 1947

  32. [32]

    Cambridge university press, 2008

    Bachir Bekka, Pierre de La Harpe, and Alain Valette.Kazh- dan’s property (T), volume 11. Cambridge university press, 2008

    work page 2008

  33. [33]

    Cor- nell Univ., 1958

    Harry Kesten.Symmetric Random Walks on Groups. Cor- nell Univ., 1958

    work page 1958

  34. [34]

    Caines, and Minyi Huang

    Shuang Gao, Peter E. Caines, and Minyi Huang. LQG graphon mean field games: Analysis via graphon-invariant subspaces.IEEE Trans. Autom. Control, 68(12):7482– 7497, 2023

    work page 2023

  35. [35]

    The chemical basis of morphogen- esis.Bull

    Alan Mathison Turing. The chemical basis of morphogen- esis.Bull. Math. Biol., 52(1):153–197, 1990

    work page 1990

  36. [36]

    Mikhailov

    Hiroya Nakao and Alexander S. Mikhailov. Turing patterns in network-organized activator–inhibitor systems.Nat. Phys., 6(7):544–550, 2010

    work page 2010