Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Beyond separability: convergence rate of vanishing viscosity approximations to mean field games via FBSDE stability

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 For mean field games with nonlocal, non-separable Hamiltonians, adding small noise of size β² yields solutions that converge at rate O(β) in value, gradient, and player distribution.

desk verdict Real improvement over [59] for non-separable nonlocal MFGs, but the headline O(beta) rate for u^beta rests on an unstated Wasserstein differentiability assumption that needs repair. read the letter →

arxiv 2505.18529 v3 pith:P2MBPBJC submitted 2025-05-24 math.OC math.APmath.PR

Winston Yu , Qiang Du , Wenpin Tang This is my paper · ORCID
classification math.OCmath.APmath.PR MSC 49N8035Q8960H1049L25
keywords meanfieldgamesvanishingviscositynon-separableHamiltonianFBSDEstabilityconvergencerateWassersteindistancepolicyiterationN-player
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

This paper establishes that the vanishing viscosity approximation to mean field games in $\mathbb{R}^d$ is first-order accurate in the noise parameter: with diffusivity $\beta^2/2$, the value function, its gradient, and the equilibrium player distribution all converge at a rate of $O(\beta)$. The result holds for Hamiltonians that are nonlocal and possibly non-separable in the measure argument, going beyond the separable structure on which earlier quantitative results relied. A sympathetic reader should care because deterministic (first-order) mean field games are numerically unstable to solve directly, while adding a small idiosyncratic noise stabilizes them; the paper supplies a quantitative control of the error this stabilization introduces. The same $O(\beta)$ rate matches the classical rate for vanishing viscosity in Hamilton–Jacobi equations, so the coupling with the Fokker–Planck equation does not degrade the order of accuracy.

What carries the argument

The proof runs through the forward-backward stochastic differential equation (FBSDE) representation of the mean field game, in which $(X_t^\beta,Y_t^\beta)$ solves $dX_t=\nabla_p H(X_t,Y_t,\rho_t)dt+\beta dB_t$ and $dY_t=-\nabla_x H(X_t,Y_t,\rho_t)dt+\beta Z_t dB_t$, with terminal condition $Y_T=-\nabla_x g(X_T,\rho_T)$. The key object is the decoupling field: a theorem of the well-posedness theory guarantees $Y_t^\beta=-\nabla u_\beta(t,X_t^\beta)$, so the forward equation for $X^\beta$ can be analyzed independently, and the uniform semiconcavity bound $K=\sup_\beta \|\nabla^2_{xx}u_\beta\|_{L^\infty}<\infty$ converts differences in gradients into differences in trajectories. A Grönwall estimate (Lemma 3.2) bounds $\mathbb{E}|X^\beta-X|^2$ by $\beta^2$ plus the $L^\infty$ gradient error; a second FBSDE stability lemma (Lemma 3.4) removes the bounded-initial-condition restriction; and a general quantitative stability theorem for Hamilton–Jacobi equations (Theorem 4.1) turns the measure-error and gradient-error bounds into the uniform-on-compacts rate for $u_\beta$.

What would settle it

A concrete way to test the claim is to compute $\sup_{t\in[0,T]}W_2(\rho^\beta_t,\rho_t)$ and $\|u_\beta-u\|_{L^\infty([0,T]\times K)}$ for a nonlocal, non-separable Hamiltonian satisfying Assumptions 2.1–2.3 but whose measure coupling is only $W_1$-Lipschitz and not Wasserstein differentiable, such as $H(x,p,\mu)=\Gamma|p|^2+\gamma(p)F(x,\rho\ast\varphi)$ with $F$ chosen to be non-differentiable in the measure argument. If the observed slope in $\beta$ exceeds one, or the error fails to vanish linearly as $\beta\to0$, the Taylor expansion at equations (4.8)–(4.10) would be the broken link.

Watch

Extended reading notes

Core claim

The central claim is Theorem 4.2: under Assumptions 2.1–2.5, for any compact set $K\subseteq\mathbb{R}^d$ and small $\beta>0$, $$\|u_\$\beta$-u\|_{L^\infty([0,T]\times K)} \le C(1+\operatorname{diam}(K)^2)\$\beta$,$$ where $u_\beta$ is the value function of the second-order mean field game and $u$ is the first-order limit. The same linear rate is obtained in the 2-Wasserstein metric for the measure flow, $W_2(\rho^\beta_t,\rho_t)\le C\beta$ (Corollary 3.6), and in $L^\infty([0,T]\times U)$ for $\nabla u_\beta\to\nabla u$ on bounded domains (Theorem 3.3), with an $L^2(\rho_t)$ version in Corollary 3.7. The result covers nonlocal, non-separable Hamiltonians and does not impose Lasry–Lions or displacement monotonicity beyond what is needed for well-posedness. This improves the previous $O(\beta^{1/2})$ rate in $L^1(\mathbb{T}^d)$ obtained for separable Hamiltonians. The paper also derives consequences for $N$-player approximations, mean field control, and the iteration complexity of policy iteration for first-order mean field games.

Load-bearing premise

The load-bearing premise is Assumption 2.5, that the mean field game is well posed for every $\beta\ge 0$ and supplies a uniform Hessian bound on the value function, because that assumption produces the FBSDE decoupling field and the semiconcavity estimate; the Taylor step in the measure argument additionally needs a Wasserstein differentiability property not explicitly listed in Assumptions 2.1–2.3.

Editorial extensions

If this is right

  • For any compact set $K\subseteq\mathbb{R}^d$, the value function converges uniformly on $[0,T]\times K$ with $\|u_\beta-u\|_{L^\infty([0,T]\times K)}\le C(1+\operatorname{diam}(K)^2)\beta$.
  • The equilibrium distribution of players converges in $W_2$ at the same linear rate, $\sup_{t\in[0,T]}W_2(\rho^\beta_t,\rho_t)\le C\beta$.
  • The gradient of the value function converges at rate $O(\beta)$ both uniformly on bounded domains and in $L^2(\rho_t)$ uniformly in time.
  • In the $N$-player approximation, choosing noise intensity $\beta\asymp\varepsilon$ and $N\gg\varepsilon^{-(d+8)}$ approximates the first-order mean field flow within accuracy $\varepsilon$ in $W_1$.
  • Policy iteration for first-order mean field games, run on the $\beta$-viscous approximation with $\beta\asymp\varepsilon$, requires on the order of $\varepsilon^{-2}$ iterations to reach accuracy $\varepsilon$.

Reading between the lines

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

  • The proof already notes the likely extension to $L^p$ and $W_p$ metrics for $p\in[1,2)$ when the initial measure has only a finite $p$-th moment; this is a natural next step that the existing arguments appear to support.
  • The Taylor expansions of $H$ and $g$ in the measure argument at equations (4.8)–(4.10) require a Wasserstein differentiability property that is not explicitly listed among Assumptions 2.1–2.3, so the true structural threshold for the $O(\beta)$ rate may be Wasserstein differentiability of the couplings, not separability.
  • For couplings that are only $W_1$-Lipschitz but not Wasserstein differentiable, the rate could drop below $O(\beta)$; this gives a concrete testable boundary for the theorem's assumptions.
  • The FBSDE-decoupling technique, rather than the dual-equation method used for separable Hamiltonians, is what enables the non-separable case; the same template may apply to other perturbations of first-order mean field games, such as entropy regularization.
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.

Referee Report

3 major / 5 minor

Summary. The paper studies the vanishing viscosity approximation of first-order mean field games on R^d with nonlocal and possibly non-separable Hamiltonians. The main claims are an O(β) convergence rate for the value function in L^∞ on compact sets, for the player distribution in the 2-Wasserstein metric, and for the gradient of the value function in L^∞ on bounded sets and in L^2(ρ_t), where β^2 is the diffusivity. The proof combines an FBSDE representation of the MFG, stability estimates for the associated forward-backward system, and a general stability theorem for Hamilton-Jacobi equations with respect to coefficient and viscosity perturbations. Applications to N-player games, mean field control, and policy iteration are presented.

Significance. If the central theorem holds under the stated assumptions, the result is a substantial improvement over the previous work [59], which obtained only an O(β^{1/2}) rate in L^1 on the torus for separable Hamiltonians. The proof strategy is genuinely different from the dual-equation PDE approach of [59]: it exploits the FBSDE decoupling field and a stability lemma for FBSDEs, and it avoids any monotonicity assumption beyond what is needed for well-posedness. The paper also provides a useful standalone stability estimate for HJ equations and several applications. The numerical section gives concrete evidence for the predicted first-order rate, and the closed-form example in Section 6.1 provides a model where the rate is in fact O(β^2), appropriately identifying the limit of the general theorem.

major comments (3)
  1. [Theorem 4.2, equations (4.8) and (4.10)] The proof of the O(β) rate for u^β in Theorem 4.2(1) relies on a first-order Taylor expansion of H and g in the measure argument, writing differences such as H(y,δ_K,ρ^β_s)-H(y,δ_K,ρ_s) in terms of E[<∇_μH(y,δ_K,ρ_s,X_s), X^β_s-X_s>] plus an o(E[|X^β_s-X_s|^2]^{1/2}) error. This requires that for each fixed (y,p) the maps μ↦H(y,p,μ) and μ↦g(y,μ) are Wasserstein differentiable at μ=ρ_s with an L^2(ρ_s)-integrable gradient, and that quantities such as ||∇_μH(0,0,ρ_s,·)||_{L^1(ρ_s)} and ||∇_μg(0,ρ_T,·)||_{L^1(ρ_T)} are finite. Assumptions 2.1–2.3 only assert second derivatives in x and p and W1-Lipschitz continuity of ∇_xH, ∇_pH, and ∇_xg; they do not assert existence or integrability of ∇_μH or ∇_μg. Assumption 2.5 and [49, Theorem 4.1] supply the FBSDE representation and the uniform semiconcavity bound K, not measure differentiability of H and g. Thus the advertised O(β) rate for u^β is not justified under the hypotheses stated, since equations (4.8)–(4.10) are load-bearing for the main theorem.
  2. [Theorem 4.1, Step 5] The comparison of solutions with different viscosity coefficients, ||v_1-v_2||_{L^∞([0,T]×K)} ≤ C(1+diam(K)^2)|ν_1-ν_2|^{1/2}, is asserted by saying 'By modifying the proof of [59, Lemma 6.3]' together with 'taking Ω=K and using the linear-in-x growth of ∇u_1 and ∇u_2, and the quadratic-in-x growth of ∂_t u_1 and ∂_t u_2'. No proof is given, and the modification is not immediate because [59, Lemma 6.3] is proved on the torus T^d for a separable Hamiltonian and may rely on monotonicity-type structure. This step is needed for the |ν_1-ν_2|^{1/2} term in the stability bound (4.2), which is then used in Theorem 4.2(2) with ν_1=β^2/2 and ν_2=0. The gap should be closed by a full proof or by a precise reference with page and theorem number.
  3. [Theorem 4.2(2)] The proof of part (2), which claims the rate for all β≥0 under an additional W1-Lipschitz assumption on H in the measure argument, uses the bound ||H(·,·,ρ^β·)-H(·,·,ρ·)||_{L^∞(K×R^d×[0,T])} ≤ C W2(ρ^β_t,ρ_t) and a corresponding bound for g. The H-bound requires that H itself, not merely ∇_xH and ∇_pH, is Lipschitz in the measure argument with respect to the Wasserstein distance. The statement 'If additionally H is Lipschitz in the measure argument with respect to W1' adds this as an extra hypothesis, but the proof also implicitly needs an analogous Lipschitz or differentiability property for g in the measure argument, since it writes ||g(·,ρ^β_T)-g(·,ρ_T)||_{L^∞(K)} ≤ ||∇_μg||_∞ W2(ρ^β_T,ρ_T). The manuscript does not state or prove such a property for g under the assumptions, and the display immediately after (4.10) uses ||∇_μg||_∞ without it being among Assumptions 2.1–2.3.
minor comments (5)
  1. [Section 2.2, Assumption 2.1] The notation ∇^2_{xμ}H and ∇^2_{pμ}H is introduced as a formal shorthand for the Lipschitz constants of ∇_xH and ∇_pH with respect to W1, but the manuscript never defines these as genuine derivatives and states 'despite not specifying a measure on P_2(R^d)'. This is confusing because the symbols suggest Hessians in the measure argument. The authors should either define these objects rigorously or use distinct notation for Lipschitz constants.
  2. [Corollary 3.6, proof] The approximation argument for removing the boundedness of the initial condition uses a random variable ξ̃ = ξ·1_{|ξ|≤r} and then writes E[|ξ-ξ̃|^2] ≤ E[|ξ|^2 : |ξ|≥r] ≤ r^{-2}E[|ξ|^2]ε. The inequality as written appears to use ε in two different ways; the text should be reorganized so that the choice of r is made explicit.
  3. [Section 5.3, Claim 5.3 and Corollary 5.4] Claim 5.3 is explicitly labeled a claim and refers to future work, yet Corollary 5.4 is stated as a result depending on that unproved claim. The paper should state clearly that Corollary 5.4 is conditional on the conjectured exponential convergence of policy iteration, and should not present it as a proved theorem.
  4. [Section 4, proof of Theorem 4.2, after (4.9)] The proof says 'the final inequality being from the observation that Corollary 3.6 holds for the L^2(Ω,F_s,P) metric as well'. Corollary 3.6 is stated for W2, and while the proof of Corollary 3.6 indeed produces a bound on E[|X^β_s-X_s|^2]^{1/2} for the FBSDE coupling, this is not stated as part of the corollary. The authors should state this stronger metric explicitly in Corollary 3.6 or add a remark explaining the distinction between W2 and the L^2(Ω) coupling.
  5. [Section 6.2] The numerical example considers a local coupling term μ(x), while Theorem 4.2 is proved for nonlocal couplings. The text appropriately says this example is outside the theorem's scope, but it would be useful to add a sentence explaining whether the observed slope 1.05 is consistent with the theory or purely exploratory.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the O(β) convergence rate follows from a forward Gronwall chain built on external well-posedness and FBSDE inputs; the only self-citation is non-load-bearing.

full rationale

The derivation is self-contained in the sense relevant to circularity. Assumption 2.5 supplies well-posedness of the MFG for all β≥0, and the FBSDE representation with decoupling field Yβ_t=-∇uβ(t,Xβ_t), together with the uniform semiconcavity bound K=sup_β∥∇²_xxuβ∥∞, is imported from the external references [49, Lemma 3.4] and [49, Theorem 4.1]; neither contains the target O(β) rate. Lemma 3.2, Theorem 3.3, Corollary 3.6, and Theorem 4.2 form a forward chain of estimates with explicit β and coefficient-difference source terms, closed by Gronwall's inequality; no parameter is fitted to the target quantity and no target inequality appears as an assumption. The only citation to the authors' own prior work, [59], is used for comparison and as proof inspiration in Theorem 4.1 Step 5 ('By modifying the proof of [59, Lemma 6.3]'), not as a black-box input that forces the conclusion; moreover, [59] establishes only a weaker O(β^{1/2}) rate in L1(T^d) under separability, so it cannot be the source of the present non-separable O(β) claim. The manuscript also flags unproved material explicitly, e.g. Claim 5.3 is stated with 'We plan to prove it rigorously in the future.' A genuine correctness concern, but not a circularity, is that the proof of Theorem 4.2(1) uses first-order Wasserstein Taylor expansions of H and g at (4.8) and (4.10), requiring ∇μH and ∇μg differentiability that is not explicitly stated in Assumptions 2.1–2.3; this is a regularity gap, not a reduction of the conclusion to its inputs.

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

The central theorem is conditional: it assumes well-posedness (Assumption 2.5) and imports the FBSDE representation, decoupling field, and semiconcavity bound from [49]. No constants are fitted to data, and no new entities are introduced. The main hidden input is the Wasserstein differentiability of H and g in the measure argument used in the Taylor expansions of Theorem 4.2.

assumptions (4)
  • domain assumption The MFG system (1.1) is well-posed for all β≥0 with solutions in the sense of Definition 2.8.
    Assumption 2.5 is stated and then in force throughout; it imports existence and uniqueness from prior literature such as [49] and [4] rather than proving it here.
  • domain assumption The FBSDE (2.2) has a strong solution and -∇uβ is its decoupling field; the uniform Hessian bound K=sup_β ∥∇²xx uβ∥∞<∞ holds.
    Invoked from [49, Theorem 4.1 and Lemma 3.4] in Section 2.3; Lemmas 3.2, 3.3 and Theorem 4.2 depend on this representation and bound.
  • ad hoc to paper H and g are Wasserstein differentiable in the measure argument with L1(ρ)-integrable gradients, allowing the Taylor expansions at equations (4.8) and (4.10).
    Not listed among Assumptions 2.1-2.3; the proof of Theorem 4.2 uses ∇μH and ∇μg with error terms o(E|Xβ-X|²^(1/2)), so this regularity is a hidden assumption.
  • standard math Classical viscosity solution theory, Gronwall's inequality, and standard BSDE well-posedness hold.
    Used throughout, including the proof of Lemma 3.4 in the appendix; these are standard background results.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond separability: convergence rate of vanishing viscosity approximations to mean field games via FBSDE stability." pith.science (2026). https://pith.science/paper/P2MBPBJC

@misc{pith2026250518529,
  author       = {Pith},
  title        = {Pith review of: Beyond separability: convergence rate of vanishing viscosity approximations to mean field games via FBSDE stability},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P2MBPBJC}},
  note         = {Machine review of arXiv:2505.18529}
}
abstract

This paper studies the vanishing viscosity approximation to mean field games (MFGs) in $\mathbb{R}^d$ with a nonlocal and possibly non-separable Hamiltonian. We prove that the value function converges at a rate of $\mathcal{O}(\beta)$, where $\beta^2$ is the diffusivity constant, which matches the classical convergence rate of vanishing viscosity for Hamilton-Jacobi (HJ) equations. The same rate is also obtained for the approximation of the distribution of players as well as for the gradient of the value function. The proof is a combination of probabilistic and analytical arguments by first analyzing the forward-backward stochastic differential equation associated with the MFG, and then applying a general stability result for HJ equations. Applications of our result to $N$-player games, mean field control, and policy iteration for solving MFGs are also presented.

Figures

Figures reproduced from arXiv: 2505.18529 by the authors.

Figure 1
Figure 1. Plot of (u β , ρβ ) for β ∈ {0.1, 0.3, 0.5, 1.0} (left to right). In [59], it was proved that u β converges at a rate of O(β 1 4 ) in some weighted L 2 norm. Now by regressing log ||u β − u||∞ over log β, we find that the slope is 1.050 using all β ∈ {0.1, . . . , 1.0}, 9Our numerical results are based on the codes available at https://colab.research.google.com/drive/ 1shJWSD2MA5Fo7_rB625dAvNTdZS1a7bG?usp=sharing [… view at source ↗
Figure 2
Figure 2. Plot of ||u β − u||∞ again β. while the slope is 1.162 using the first half β ∈ {0.1, . . . , 0.5}. It is natural to expect that ||u β − u||∞ ≍ β 1+δ as β → 0, (6.7) for some 0 < δ < 1. The rate (6.7) is better than the proved O(β 1 4 )-rate for MFGs with a local Hamiltonian, and is between the O(β)-rate for MFGs with a general nonlocal Hamiltonian and the O(β 2 )-rate for the example in Section 6.1. An interesting … view at source ↗

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

  1. On the rate of the vanishing viscosity approximation for Mean Field Games with nonlocal coupling

    math.AP 2026-07 accept novelty 6.0 of 10

    The vanishing viscosity approximation for first-order MFGs with nonlocal coupling converges at rate O(ε^{1/2}) for the value function and O(ε^{1/8}) for the density in Wasserstein distance.

Reference graph

Works this paper leans on

65 extracted references · 51 canonical work pages · cited by 1 Pith paper

  1. [59]

    Tang and Y

    W. Tang and Y. P. Zhang. The convergence rate of vanishing viscosity approximations for mean field games.SIAM J. Math. Anal., 57(3):3217–3254, 2025

    work page 2025

  2. [1]

    Achdou, F

    Y. Achdou, F. J. Buera, J.-M. Lasry, P.-L. Lions, and B. Moll. Partial differential equation models in macroeconomics.Philos. Trans. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci., 372(2028):20130397, 19, 2014

    work page 2014

  3. [2]

    Achdou and M

    Y. Achdou and M. Lauri` ere. Mean field games and applications: numerical aspects. In Mean field games, volume 2281 ofLecture Notes in Math., pages 249–307. Springer, Cham, 2020

    work page 2020

  4. [3]

    Achdou and A

    Y. Achdou and A. Porretta. Mean field games with congestion.Ann. Inst. H. Poincar´ e C Anal. Non Lin´ eaire, 35(2):443–480, 2018

    work page 2018

  5. [4]

    Bansil, A

    M. Bansil, A. R. M´ esz´ aros, and C. Mou. Global well-posedness of displacement monotone degenerate mean field games master equations.SIAM J. Control Optim., 63(2):993–1021, 2025

    work page 2025

  6. [5]

    Cacace, F

    S. Cacace, F. Camilli, and A. Goffi. A policy iteration method for mean field games. ESAIM Control Optim. Calc. Var., 27:Paper No. 85, 19, 2021

    work page 2021

  7. [6]

    Calzola, E

    E. Calzola, E. Carlini, and F. J. Silva. A high-order scheme for mean field games.J. Comput. Appl. Math., 445:Paper No. 115769, 19, 2024

    work page 2024

  8. [7]

    F. Camilli. A policy iteration method for mean field games.IF AC-PapersOnLine, 55(30):406–411, 2022

    work page 2022

Show all 65 references
  1. [8]

    Camilli and Q

    F. Camilli and Q. Tang. Rates of convergence for the policy iteration method for mean field games systems.J. Math. Anal. Appl., 512(1):Paper No. 126138, 18, 2022

    2022

  2. [9]

    Cardaliaguet

    P. Cardaliaguet. Weak solutions for first order mean field games with local coupling. In Analysis and geometry in control theory and its applications, volume 11 ofSpringer INdAM Ser., pages 111–158. Springer, Cham, 2015

    2015

  3. [10]

    Cardaliaguet and J

    P. Cardaliaguet and J. P. Graber. Mean field games systems of first order.ESAIM Control Optim. Calc. Var., 21(3):690–722, 2015

    2015

  4. [11]

    Cardaliaguet and A

    P. Cardaliaguet and A. Porretta. An introduction to mean field game theory. InMean field games, volume 2281 ofLecture Notes in Math., pages 1–158. Springer, Cham, 2020

    2020

  5. [12]

    Carlini and F

    E. Carlini and F. J. Silva. A fully discrete semi-Lagrangian scheme for a first order mean field game problem.SIAM J. Numer. Anal., 52(1):45–67, 2014

    2014

  6. [13]

    Carlini, F

    E. Carlini, F. J. Silva, and A. Zorkot. A Lagrange-Galerkin scheme for first order mean field game systems.SIAM J. Numer. Anal., 62(1):167–198, 2024

    2024

  7. [14]

    R. Carmona. Applications of mean field games in financial engineering and economic theory. InMean field games, volume 78 ofProc. Sympos. Appl. Math., pages 165–219. Amer. Math. Soc., Providence, RI, 2021

    2021

  8. [15]

    Carmona and F

    R. Carmona and F. Delarue.Probabilistic theory of mean field games with applications I, volume 83 ofProbability Theory and Stochastic Modelling. Springer, Cham, 2018. Mean field FBSDEs, control, and games. 28 WINSTON YU, QIANG DU, AND WENPIN TANG

    2018

  9. [16]

    Carmona and D

    R. Carmona and D. Lacker. A probabilistic weak formulation of mean field games and applications.Ann. Appl. Probab., 25(3):1189–1231, 2015

    2015

  10. [17]

    Chaintron and S

    L.-P. Chaintron and S. Daudin. Optimal rate of convergence in the vanishing viscosity for uniformly convex Hamilton-Jacobi equations.arxiv:2506.13255, 2025

    2025 arXiv

  11. [18]

    Cirant and A

    M. Cirant and A. Goffi. Convergence rates for the vanishing viscosity approximation of Hamilton-Jacobi equations: the convex case.arxiv:2502.15495, 2025

    2025 arXiv

  12. [19]

    Cirant and A

    M. Cirant and A. R. M´ esz´ aros. Long time behavior and stabilization for displacement monotone mean field games.arXiv:2412.14903, 2024

    2024

  13. [20]

    Colombo, G

    M. Colombo, G. Crippa, M. Graff, and L. V. Spinolo. On the role of numerical viscosity in the study of the local limit of nonlocal conservation laws.ESAIM Math. Model. Numer. Anal., 55(6):2705–2723, 2021

    2021

  14. [21]

    Colombo, G

    M. Colombo, G. Crippa, and L. V. Spinolo. On the singular local limit for conservation laws with nonlocal fluxes.Arch. Ration. Mech. Anal., 233(3):1131–1167, 2019

    2019

  15. [22]

    Crandall, H

    M. Crandall, H. Ishii, and P.-L. Lions. User’s guide to viscosity solutions of second order partial differential equations.Bull. Amer. Math. Soc. (N.S.), 27(1), 1992

    1992

  16. [23]

    M. G. Crandall and P.-L. Lions. Viscosity solutions of Hamilton-Jacobi equations.Trans. Amer. Math. Soc., 277(1):1–42, 1983

    1983

  17. [24]

    Daudin, F

    S. Daudin, F. Delarue, and J. Jackson. On the optimal rate for the convergence problem in mean field control.J. Funct. Anal., 287(12):Paper No. 110660, 94, 2024

    2024

  18. [25]

    Delarue, D

    F. Delarue, D. Lacker, and K. Ramanan. From the master equation to mean field game limit theory: a central limit theorem.Electron. J. Probab., 24:1–54, 2019

    2019

  19. [26]

    Delarue, D

    F. Delarue, D. Lacker, and K. Ramanan. From the master equation to mean field game limit: large deviations and concentration of measure.Ann. Probab., 48(1):211–263, 2020

    2020

  20. [27]

    Djehiche, A

    B. Djehiche, A. Tcheukam, and H. Tembine. Mean-field-type games in engineering.AIMS Electronics and Electrical Engineering, 1(1):18–73, 2017

    2017

  21. [28]

    M. F. Djete. Extended mean field control problem: a propagation of chaos result.Electron. J. Probab., 27:1–53, 2022

    2022

  22. [29]

    Eldan, D

    R. Eldan, D. Mikulincer, and A. Zhai. The CLT in high dimensions: quantitative bounds via martingale embedding.Ann. Probab., 48(5):2494–2524, 2020

    2020

  23. [30]

    Ferreira, D

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

    2021

  24. [31]

    M. Fischer. On the connection between symmetric n-player games and mean field games. Ann. Appl. Probab., 27(2):757–810, 2022

    2022

  25. [32]

    Gangbo, A

    W. Gangbo, A. R. M´ esz´ aros, C. Mou, and J. Zhang. Mean field games master equa- tions with nonseparable Hamiltonians and displacement monotonicity.Ann. Probab., 50(6):2178–2217, 2022

    2022

  26. [33]

    Ghattassi and N

    M. Ghattassi and N. Masmoudi. Non-separable mean field games for pedestrian flow: generalized Hughes model.Math. Models Methods Appl. Sci., 35(7):1571–1607, 2025

    2025

  27. [34]

    P. J. Graber and A. R. M´ esz´ aros. On monotonicity conditions for mean field games.J. Funct. Anal., 285(9):Paper No. 110095, 45, 2023

    2023

  28. [35]

    X. Guo, H. V. Tran, and Y. P. Zhang. Policy iteration for nonconvex viscous Hamilton– Jacobi equations.arXiv:2503.02159, 2025

    2025 arXiv

  29. [36]

    Huang, X

    K. Huang, X. Di, Q. Du, and X. Chen. A game-theoretic framework for autonomous vehicles velocity control: Bridging microscopic differential games and macroscopic mean field games.Discrete Contin. Dyn. Syst. Ser.B, 25(12):4869–4903, 2020. V ANISHING VISCOSITY APPROXIMATIONS 29

    2020

  30. [37]

    Huang, P

    M. Huang, P. E. Caines, and R. P. Malham´ e. Large-population cost-coupled LQG problems with nonuniform agents: individual-mass behavior and decentralizedϵ-Nash equilibria. IEEE Trans. Automat. Control, 52(9):1560–1571, 2007

    2007

  31. [38]

    Huang, R

    M. Huang, R. Malhame, and P. Caines. Large population stochastic dynamic games: closed-loop Mckean-Vlasov systems and the nash certainty equivalence principle.Commun. Inf. Syst., 6(3):221–252, 2006

    2006

  32. [39]

    Kerimkulov, D

    B. Kerimkulov, D. Siska, and L. Szpruch. Exponential convergence and stability of Howard’s policy improvement algorithm for controlled diffusions.SIAM J. Control Optim., 58(3):1314–1340, 2020

    2020

  33. [40]

    D. Lacker. Mean field games via controlled martingale problems: existence of Markovian equilibria.Stochastic Process. Appl., 125(7):2856–2894, 2015

    2015

  34. [41]

    D. Lacker. On the convergence of closed-loop nash equilibria to the mean field game limit. Ann. Appl. Probab., 30(4):1693–1761, 2020

    2020

  35. [42]

    Lasry and P.-L

    J.-M. Lasry and P.-L. Lions. Jeux ` a champ moyen. I. le cas stationnaire.C. R. Math. Acad. Sci. Paris, 343(9):619–625, 2006

    2006

  36. [43]

    Lasry and P.-L

    J.-M. Lasry and P.-L. Lions. Jeux ` a champ moyen. II – horizon fini et contrˆ ole optimal. C. R. Math. Acad. Sci. Paris, 343(10):679–684, 2006

    2006

  37. [44]

    Lasry and P.-L

    J.-M. Lasry and P.-L. Lions. Mean field games.Jpn. J. Math., 2(1):229–260, 2007

    2007

  38. [45]

    Lauri` ere, J

    M. Lauri` ere, J. Song, and Q. Tang. Policy iteration method for time-dependent mean field games systems with non-separable Hamiltonians.Appl. Math. Optim., 87(17), 2023

    2023

  39. [46]

    Lin and E

    C.-T. Lin and E. Tadmor.L 1-stability and error estimates for approximate Hamilton- Jacobi solutions.Numer. Math., 87(4):701–735, 2001

    2001

  40. [47]

    S. Liu, M. Jacobs, W. Li, L. Nurbekyan, and S. J. Osher. Computational methods for first- order nonlocal mean field games with applications.SIAM J. Numer. Anal., 59(5):2639– 2668, 2021

    2021

  41. [48]

    J. Ma, G. Wang, and J. Zhang. Convergence analysis for entropy-regularized control problems: a probabilistic approach.arXiv:2406.10959, 2024

    2024 arXiv

  42. [49]

    A. R. M´ esz´ aros and C. Mou. Mean field games systems under displacement monotonicity. SIAM J. Math. Anal., 56(1):529–553, 2024

    2024

  43. [50]

    B. Moll. Mean field games in macroeconomics.Econ Meets Math Workshop, 2019. Avail- able athttps://benjaminmoll.com/wp-content/uploads/2020/10/MFG_macro.pdf

    2019

  44. [51]

    Mou and J

    C. Mou and J. Zhang. Mean field game master equations with anti-monotonicity condi- tions.arXiv:2201.10762. To appear in J. Eur. Math. Soc., 2024

    2024 arXiv

  45. [52]

    Resnick.A Probability Path

    S. Resnick.A Probability Path. Modern Birkh¨ auser Classics. Birkh¨ auser Boston, MA, first edition, 2014

    2014

  46. [53]

    E. Rio. Upper bounds for minimal distances in the central limit theorem.Ann. Inst. Henri Poincar´ e Probab. Stat., 45(3):802–817, 2009

    2009

  47. [54]

    P. E. Souganidis. Existence of viscosity solutions of Hamilton-Jacobi equations.J. Differ- ential Equations, 56(5):345–390, 1985

    1985

  48. [55]

    E. Tadmor. Numerical viscosity and the entropy condition for conservative difference schemes.Math. Comp., 43(168):369–381, 1984

    1984

  49. [56]

    W. Tang. Trading and wealth evolution in the proof of stake protocol. InProof-of-Stake for Blockchain Networks: Fundamentals, Challenges and Approaches, pages 135–161. IET, 2024

    2024

  50. [57]

    Tang and F

    W. Tang and F. Tang. The Poisson binomial distribution—old & new.Statist. Sci., 38(1):108–119, 2023. 30 WINSTON YU, QIANG DU, AND WENPIN TANG

    2023

  51. [58]

    W. Tang, H. V. Tran, and Y. P. Zhang. Policy iteration for the deterministic control problems—a viscosity approach.SIAM J. Control Optim., 63(1):375–401, 2025

    2025

  52. [60]

    W. Tang, Y. P. Zhang, and X. Y. Zhou. Exploratory HJB equations and their convergence. SIAM J. Control Optim., 60(6):3191–3216, 2022

    2022

  53. [61]

    Tang and X

    W. Tang and X. Y. Zhou. Regret of exploratory policy improvement andq-learning. arXiv:2411.01302, 2024

    2024 arXiv

  54. [62]

    H. V. Tran.Hamilton-Jacobi equations—theory and applications, volume 213 ofGraduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2021

    2021

  55. [63]

    H. V. Tran, Z. Wang, and Y. P. Zhang. Policy iteration for exploratory Hamilton–Jacobi– Bellman equations.Appl. Math. Optim., 91(2):Paper No. 50, 2025

    2025

  56. [64]

    H. Wang, T. Zariphopoulou, and X. Y. Zhou. Reinforcement learning in continuous time and space: A stochastic control approach.J. Mach. Learn. Res., 21:1–34, 2020

    2020

  57. [65]

    A. Zhai. A high-dimensional CLT inW 2 distance with near optimal convergence rate. Probab. Theory Related Fields, 170(3-4):821–845, 2018. AppendixA.Proof of Lemma 3.4 For convenience, we define the following quantities: δT =|(G− ˜G)(XT , ρT )|2,∆ t =|(B− ˜B, F− ˜F ,Σ− ˜Σ)(Xt, ...

    2018

Pith tools

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