pith. sign in

arxiv: 2606.19611 · v1 · pith:SGQKOCRKnew · submitted 2026-06-17 · 🧮 math.NA · cs.NA· math.AP

Bregman-projected mirror methods for regularized stationary mean-field games

Hussain Al Abdulaziz , Yuri Ashrafyan , Yeva Gevorgyan , Diogo Gomes This is my paper

Pith reviewed 2026-06-26 19:47 UTC · model grok-4.3

classification 🧮 math.NA cs.NAmath.AP
keywords mean-field gamesBregman projectionmirror descentvariational inequalitiesregularizationconvergence analysisnumerical methodsstationary problems
0
0 comments X

The pith

A constrained two-step mirror method converges strongly to the solution of low-order regularized stationary mean-field games.

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

The paper formulates low-order Laplacian regularization of stationary mean-field game systems as a variational inequality in mixed Lebesgue-Sobolev spaces when Hamiltonians are separable with specified growth. It introduces a Bregman geometry matched to those spaces and studies a constrained two-step mirror iteration that freezes the operator evaluation. A one-step Bregman inequality is derived for the exact iteration, establishing strong convergence to the unique solution for each fixed regularization parameter under summability conditions on the step sizes. Numerical tests on one- and two-dimensional problems confirm residual decay with mesh refinement and practical gains from the two-step scheme.

Core claim

For separable Hamiltonians of the stated growth type, the regularized stationary MFG system becomes a variational inequality on L^β(T^d) × W^{1,γ}(T^d); the exact constrained Bregman-projected mirror iteration satisfies a one-step Bregman inequality that yields strong convergence to its unique solution whenever the step sizes obey the natural summability requirement.

What carries the argument

The one-step Bregman inequality for the exact constrained two-step mirror method with frozen operator evaluation, which bounds progress in the Bregman divergence adapted to the mixed Banach-space geometry of the regularized variational inequality.

If this is right

  • Strong convergence holds in the natural Banach space for every fixed regularization parameter ε>0.
  • The same one-step inequality applies directly to the exact constrained iteration under the given summability condition on step sizes.
  • Residual decay is observed under mesh refinement in one- and two-dimensional discretizations.
  • The two-step implementation exhibits improved practical performance compared with the one-step version in the tested cases.

Where Pith is reading between the lines

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

  • The frozen-operator two-step structure may reduce per-iteration cost when the MFG operator is expensive to evaluate repeatedly.
  • Similar Bregman geometries matched to mixed Sobolev-Lebesgue spaces could be examined for other low-order regularizations of variational inequalities arising in game theory.
  • The summability condition on step sizes is the precise requirement that converts the one-step inequality into a convergent telescoping sum.

Load-bearing premise

The Hamiltonians take the separable form H(x,p,m)=H0(x,p)−g(m) with quadratic or super-quadratic growth and linear or super-linear density couplings.

What would settle it

A sequence of step sizes whose reciprocals sum to a finite value for which the exact constrained iteration fails to converge strongly in the mixed Lebesgue-Sobolev norm to the unique regularized solution.

Figures

Figures reproduced from arXiv: 2606.19611 by Diogo Gomes, Hussain Al Abdulaziz, Yeva Gevorgyan, Yuri Ashrafyan.

Figure 1
Figure 1. Figure 1: overlays the numerical solution on the exact one at N = 64, and [PITH_FULL_IMAGE:figures/full_fig_p031_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Convergence for the exact test solution (d = 1). Errors ∥mh − m⋆∥L3 and |uh − u ⋆ |W1,3 versus h on a log-log scale, with a first-order reference line. 7.4. One-dimensional results. We solve the stationary MFG on T = R/Z with drift b(x) = cos(2πx), potential V (x) = sin(2πx), and grid sizes N = 64, 128, 256, 512, 1024 [PITH_FULL_IMAGE:figures/full_fig_p032_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a) shows the residual RN (zn) versus iteration number on a semilog scale for all five grid sizes. The convergence curves are nearly mesh-independent over the tested range, in agreement with the iteration counts in [PITH_FULL_IMAGE:figures/full_fig_p033_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: shows the density m(x) and value function u(x) at N = 64 and 128, which overlap closely. The density is a smooth modulation of the uniform state m = 1. 0 0.5 1 0.5 1 1.5 x m (a) density m(x) 0 0.5 1 1.3 1.4 1.5 x u (b) value function u(x) [PITH_FULL_IMAGE:figures/full_fig_p033_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Exact test solution (d = 2, 64 × 64 grid). Numerical density mh and value function uh (points) overlaid on the exact surfaces m⋆ , u⋆ (shaded); the points lie on the surfaces [PITH_FULL_IMAGE:figures/full_fig_p035_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Convergence of the two-step mirror descent versus iteration for the two-dimensional experiment on the 16 × 16, 32 × 32, and 64×64 grids (semilog): (a) residual RN (zn); (b) Bregman increment DΦ(zn, zn−1). (a) density m(x1, x2) (b) value function u(x1, x2) [PITH_FULL_IMAGE:figures/full_fig_p036_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Two-dimensional equilibrium on the 64×64 grid. Left panel shows the density m(x1, x2), ranging in [0.004, 2.00]. Right panel shows the value function u(x1, x2). the bounded admissible set K. Together, these yield strong convergence of the iter￾ates for each fixed ε > 0. The numerical experiments, including validation against exact test solutions, are consistent with the theory and show mesh-robust residual… view at source ↗
read the original abstract

We develop and analyze a Bregman-projected mirror iteration for low-order regularizations of stationary mean-field game (MFG) systems in their natural Banach space setting. For separable Hamiltonians of the form \(H(x,p,m)=H_0(x,p)-g(m)\), with quadratic or super-quadratic Hamiltonian growth and linear or super-linear density couplings, we formulate a low-order \(\bar\gamma\)-Laplacian regularization of the stationary MFG system as a variational inequality on \(L^{\bar\beta}(\mathbb T^d)\times W^{1,\bar\gamma}(\mathbb T^d)\). To approximate solutions of this regularized variational inequality, we introduce a Bregman geometry matched to the mixed Lebesgue--Sobolev exponents of the problem and analyze a constrained two-step mirror method with frozen operator evaluation. For the exact constrained iteration and each fixed regularization parameter \(\epsi>0\), we derive a one-step Bregman inequality and use it to prove that the constrained iteration converges strongly to the unique solution of the regularized variational inequality under natural summability conditions on the step sizes. Numerical experiments on one- and two-dimensional models, validated against exact test solutions, illustrate residual decay under mesh refinement and suggest improved practical performance of the two-step implementation in the tested discretizations.

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

0 major / 2 minor

Summary. The manuscript develops and analyzes Bregman-projected mirror methods for low-order regularized stationary mean-field game systems posed as variational inequalities in the natural Banach spaces L^β(T^d) × W^{1,γ}(T^d). For separable Hamiltonians H(x,p,m)=H0(x,p)−g(m) with quadratic/super-quadratic growth and linear/super-linear couplings, it introduces a constrained two-step mirror iteration with frozen operator evaluation, derives a one-step Bregman inequality for the exact constrained iteration (fixed ε>0), and proves strong convergence to the unique regularized solution under summability conditions on the step sizes α_k. Numerical experiments on 1D/2D models with exact test solutions illustrate residual decay under mesh refinement.

Significance. If the one-step Bregman inequality and resulting convergence hold under the stated growth and coupling assumptions, the work supplies a technically matched Bregman geometry and convergence theory for mirror methods applied to low-order regularizations of stationary MFGs. This addresses a gap in handling the mixed Lebesgue–Sobolev structure without higher-order smoothing and provides a foundation for practical two-step implementations, which is a meaningful contribution to numerical analysis of MFGs.

minor comments (2)
  1. The abstract and introduction distinguish the convergence analysis (exact constrained iteration) from the two-step practical scheme, but the manuscript should clarify in §3 or §4 whether any convergence guarantee extends to the frozen-operator two-step version or if it remains heuristic.
  2. Notation for the regularization parameter (ε vs. arγ) and the precise definition of the Bregman distance should be cross-checked for consistency between the variational inequality formulation and the iteration analysis.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive and detailed summary of our work on Bregman-projected mirror methods for regularized stationary mean-field games. We appreciate the recognition that the one-step Bregman inequality and convergence theory address a relevant gap for low-order regularizations in the mixed Lebesgue-Sobolev setting. No major comments were provided in the report. We will implement the recommended minor revision and any additional editorial suggestions in the revised manuscript.

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The central claim derives a one-step Bregman inequality directly from the exact constrained iteration applied to the regularized variational inequality (under the separable Hamiltonian and growth assumptions), then invokes standard summability conditions on step sizes to obtain strong convergence to the unique solution. This is a standard direct proof in the Banach-space setting and does not reduce to any fitted input, self-definition, or load-bearing self-citation chain. Numerical experiments are validated against independent exact test solutions, confirming the analysis remains independent of its own outputs.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim depends on domain assumptions about Hamiltonian separability and growth, the choice of regularization, and the specific function spaces; step sizes are free parameters subject to summability.

free parameters (2)
  • regularization parameter ε
    Fixed positive value for each regularized problem; controls the strength of the γ-Laplacian term.
  • step sizes α_k
    Sequence must satisfy natural summability conditions for the convergence proof to hold.
axioms (2)
  • domain assumption Hamiltonians are separable H(x,p,m)=H0(x,p)−g(m) with quadratic or super-quadratic growth and linear or super-linear density couplings
    Required to formulate the regularized system as a variational inequality on the stated Banach spaces.
  • domain assumption The low-order γ-Laplacian regularization yields a well-posed variational inequality on L^β(T^d)×W^{1,γ}(T^d)
    This is the setting in which the Bregman geometry and mirror iteration are defined and analyzed.

pith-pipeline@v0.9.1-grok · 5773 in / 1347 out tokens · 18524 ms · 2026-06-26T19:47:20.731945+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

25 extracted references

  1. [1]

    Achdou and I

    Y. Achdou and I. Capuzzo-Dolcetta. Mean field games: Numerical methods.Siam Journal On Numerical Analysis, 48(3):1136–1162, 2010

    2010

  2. [2]

    Almulla, R

    N. Almulla, R. Ferreira, and D. Gomes. Two Numerical Approaches to Stationary Mean-Field Games.Dynamic Games and Applications, 7(4):657–682, Dec. 2017

    2017

  3. [3]

    Aubin-Frankowski, A

    P.-C. Aubin-Frankowski, A. Korba, and F. L´ eger. Mirror descent with relative smoothness in measure spaces, with application to Sinkhorn and EM. InAdvances in Neural Information Processing Systems, 2022. arXiv:2206.08873

    arXiv 2022

  4. [4]

    Barles and P

    G. Barles and P. Souganidis. Convergence of approximation schemes for fully nonlinear second order equations.Asymptotic Analysis, 4(3):271–283, 1991

    1991

  5. [5]

    Beck and M

    A. Beck and M. Teboulle. Mirror descent and nonlinear projected subgradient methods for convex optimization.Operations Research Letters, 31:167–175, 2003

    2003

  6. [6]

    L. M. Brice˜ no-Arias, J. Deride, S. L´ opez-Rivera, and F. J. Silva. A primal-dual partial inverse algorithm for constrained monotone inclusions: Applications to stochastic programming and mean field games.Applied Mathematics & Optimization, 87:21, 2023

    2023

  7. [7]

    Ferreira and D

    R. Ferreira and D. Gomes. Existence of weak solutions to stationary mean-field games through variational inequalities.Siam Journal On Mathematical Analysis, 50(6):5969–6006, 2018

    2018

  8. [8]

    Ferreira, D

    R. Ferreira, D. Gomes, and T. Tada. Existence of weak solutions to first-order stationary mean-field games with Dirichlet conditions.Proceedings of the American Mathematical Society, 147(11):4713–4731, 2019

    2019

  9. [9]

    Ferreira, D

    R. Ferreira, D. Gomes, and T. Tada. Existence of weak solutions to time-dependent mean-field games.Nonlinear Analysis, 212:Paper No. 112470, 31, 2021

    2021

  10. [10]

    Ferreira, D

    R. Ferreira, D. Gomes, and M. Ucer. Solving Mean-Field Games with Monotonicity Methods in Banach Spaces, June 2025

    2025

  11. [11]

    Gevorgyan and D

    Y. Gevorgyan and D. Gomes. A monotonicity-based numerical method for price formation in mean field games. Preprint, 2026

    2026

  12. [12]

    D. V. Hieu and S. Reich. Two Bregman projection methods for solving variational inequalities. Optimization, 71(7):1777–1802, 2022

    2022

  13. [13]

    Huang, P

    M. Huang, P. E. Caines, and R. P. Malhame. Large-Population Cost-Coupled LQG Problems With Nonuniform Agents: Individual-Mass Behavior and Decentralized$\varepsilon$-Nash Equilibria.IEEE Transactions on Automatic Control, 52(9):1560–1571, Sept. 2007

    2007

  14. [14]

    Huang, R

    M. Huang, R. P. Malham´ e, and P. E. Caines. Large population stochastic dynamic games: Closed-loop McKean-Vlasov systems and the Nash certainty equivalence principle.Communi- cations in Information and Systems, 6(3):221–251, 2006

    2006

  15. [15]

    Izuchukwu, S

    C. Izuchukwu, S. Reich, and Y. Shehu. One-step Bregman projection methods for solving variational inequalities in reflexive Banach spaces.Optimization, 73(5):1519–1549, 2023

    2023

  16. [16]

    Lasry and P.-L

    J.-M. Lasry and P.-L. Lions. Jeux ` a champ moyen. I. Le cas stationnaire.Comptes Rendus Mathematique. Academie des Sciences. Paris, 343(9):619–625, 2006

    2006

  17. [17]

    Lasry and P.-L

    J.-M. Lasry and P.-L. Lions. Jeux ` a champ moyen. II – Horizon fini et contrˆ ole optimal.Comptes Rendus. Math´ ematique, 343(10):679–684, 2006

    2006

  18. [18]

    Lasry and P.-L

    J.-M. Lasry and P.-L. Lions. Mean field games.Japanese Journal of Mathematics, 2(1):229–260, Mar. 2007. 38 HUSSAIN AL ABDULAZIZ, YURI ASHRAFYAN, YEVA GEVORGYAN, AND DIOGO GOMES

    2007

  19. [19]

    Lauri` ere, L

    M. Lauri` ere, L. Musik, and A. Perrin. A mirror descent approach for mean field control. hal- 03972660, 2023

    2023

  20. [20]

    A. S. Nemirovsky and D. B. Yudin.Problem Complexity and Method Efficiency in Optimization. John Wiley & Sons, 1983

    1983

  21. [21]

    Nurbekyan

    L. Nurbekyan. A note on the convergence of the monotone inclusion version of the primal-dual hybrid gradient algorithm, Nov. 2023

    2023

  22. [22]

    Nurbekyan, S

    L. Nurbekyan, S. Liu, and Y. T. Chow. Monotone inclusion methods for a class of second-order non-potential mean-field games, Mar. 2024

    2024

  23. [23]

    P´ erolat, S

    J. P´ erolat, S. Perrin, R. Elie, M. Lauri` ere, G. Piliouras, and M. Geist. Scaling up mean field games with online mirror descent. InProceedings of the 21st International Conference on Au- tonomous Agents and Multiagent Systems (AAMAS), 2022. arXiv:2103.00623

    arXiv 2022

  24. [24]

    V. V. Semenov. A version of the mirror descent method to solve variational inequalities.Cy- bernetics and Systems Analysis, 53:234–243, 2017

    2017

  25. [25]

    Z. Wu, M. Lauriere, S. J. C. Chua, M. Geist, O. Pietquin, and A. Mehta. Population-aware Online Mirror Descent for Mean-Field Games by Deep Reinforcement Learning, Mar. 2024. King Abdullah University of Science and Technology (KAUST), CEMSE Division, Thuwal 23955-6900, Saudi Arabia Email address:hussain.abdulaziz.1@kaust.edu.sa King Abdullah University of...

    2024