{"id":"47bc989f-718a-446d-bf02-85ce3b6c2a92","arxiv_id":"2606.09815","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Normalized N-player α_N-potential functions converge in value and minimizers to an MFC problem whose objective is a potential for the limiting MFG precisely when lim α_N=0.","lead":"Finite-player α-potential games converge, after normalization, to potential mean-field games as the player count grows. The limit yields both a mean-field control problem and a practical route to construct potential MFGs from finite games via vanishing α.","discovery_kind":"unification","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The manuscript supplies a self-contained limit theory that rigorously connects finite-player α-potential games to potential MFGs via vanishing α_N. The technical core—new path-integral construction of α_N-potentials, propagation of chaos for the normalized potentials onto a lifted MFC problem, Poincaré lemma on Wasserstein space, and the resulting identification of the MFC objective as an MFG potential—is supported by complete proofs and concrete examples (3.1–3.3, 4.1–4.3). The only substantive modelling restriction is the non-degeneracy/compactness package in Assumption 3.1(iii), which is standard for measure-valued-control compactness and is already caveated by the authors. Because the theorems are correctly stated under these hypotheses and no logical gap appears, the reader’s ACCEPT verdict with high confidence stands. The suggested concrete test merely checks whether the non-degeneracy constant can be relaxed; a positive outcome would strengthen rather than retract the paper.","tokens_in":47656,"tokens_out":525,"duration_ms":5548,"concrete_test":"Verify that the O(1/N) discrepancy between Φ^N and ˜Φ^N in the decomposition (5.1)–(5.2) remains uniform under the linear-growth bounds of Lemma 5.1 when the non-degeneracy constant θ is allowed to approach 0 while keeping A compact and coefficients bounded; if the relative-compactness argument of [17] still produces the same limit points, the non-degeneracy hypothesis can be weakened without changing the claims.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claims (Theorems 3.1, 4.1–4.3, Corollary 4.1) rest on standard compactness/non-degeneracy assumptions (Assumption 3.1(iii)) that the paper itself flags in Remark 3.1 as removable when the running cost depends only on the state law. The Poincaré lemma (Proposition 5.1 / Theorems 4.1–4.2) and the identification of the lifted MFC objective as an MFG potential are proved in detail under the stated regularity (C^{0,2,2} costs, Lipschitz coefficients). No internal inconsistency, hidden circularity, or gap in the chain from vanishing α_N to potential MFG structure appears. The reader’s weakest-assumption note is accurate but does not undermine the theorems as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper studies the large-population limit of N-player α-potential stochastic differential games. It constructs an explicit α_N-potential via path integrals on the joint state-control space (Theorem 2.1), proves that the normalized potentials and their approximate minimizers converge to a mean-field control problem with measure-valued controls on a lifted canonical space (Theorem 3.1), and shows that lim α_N=0 is equivalent to the classical closedness/symmetry conditions for potential MFGs (Theorems 4.1–4.2). Under that condition the limiting MFC objective is itself a potential for an associated MFG with measure-valued controls (Theorem 4.3), yielding propagation of chaos from α_N-Nash equilibria to mean-field equilibria for controlled diffusions with common noise and non-separable control interactions (Corollary 4.1). A technical cornerstone is a Poincaré lemma / Green formula on Wasserstein space (Proposition 5.1).","tokens_in":47850,"tokens_out":1016,"duration_ms":19559,"significance":"If the results hold, the paper supplies a clean bridge from finite-player α-potential games to potential mean-field games, giving an explicit asymptotic construction of potential MFGs via vanishing α_N and a new route to propagation of chaos that covers common noise and non-separable state–control costs. The Poincaré lemma on Wasserstein space and the careful treatment of measure-valued controls (with concrete examples showing that lifting is necessary) are genuine technical contributions. The work therefore advances both the conceptual understanding of potential structures and the toolkit for large-population games.","major_comments":[{"comment":"Assumption 3.1(iii) (compact A, bounded coefficients, uniform non-degeneracy of idiosyncratic noise) is used for relative compactness of the empirical measures and for identifying all accumulation points of approximate minimizers of Φ_N. Remark 3.1 correctly notes that non-degeneracy can be dropped when the running cost depends only on the state law, but the main statements (Theorem 3.1, Corollary 4.1) are stated under the stronger hypothesis. A short additional remark clarifying which conclusions survive under mere Lipschitz coefficients (or citing the precise results of [18,22] that apply) would make the scope of the PoC claim more transparent without changing the theorems.","section":null},{"comment":"In the adaptation of the propagation-of-chaos arguments of [17] (proof of Theorem 3.1), the N-dependent costs F_N, G_N differ from the limiting F_∞, G_∞ by O(1/N) terms. The paper asserts that these discrepancies vanish uniformly (display (5.2)), yet the uniform-integrability estimates that justify interchanging limits under the p>2 moment assumption are only sketched. A few additional lines verifying the uniform L^{p/2} bounds on the remainder (or an explicit reference to the corresponding estimates in [17]) would close a minor but load-bearing gap in the written proof.","section":null}],"minor_comments":[{"comment":"Notation for the lifted spaces (Ω̃, Ω̃′, Λ̃, Π̃, etc.) is dense; a short table or diagram summarizing the hierarchy of measures would help the reader.","section":null},{"comment":"In Theorem 2.1 the upper bound on α_N is expressed with H^{2}-norms of state-control pairs; Corollary 2.1 then converts it into a more explicit constant. The dependence of C on the Lipschitz constants of (b,σ,γ) could be written out once for completeness.","section":null},{"comment":"Examples 3.1–3.2 are illuminating but lengthy; the key message (that the barycentric projection loses the nonlinear cost) could be highlighted in a single sentence before the calculations.","section":null},{"comment":"Typographical: “derivarive” (p. 2), “Poincar´e” inconsistently accented, and occasional missing spaces around “N-player”.","section":null},{"comment":"References [3,5,16] to the authors’ earlier α-potential work are appropriate; a one-sentence comparison of the new path-integral construction with the sensitivity-process construction of [3] would orient readers familiar with that literature.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is technically solid and fits well in a top optimization/control journal. The self-citation density to the authors’ α-potential series is noticeable but not excessive given the logical dependence. I see no novelty or priority issues. Minor revision is sufficient; the two major points are clarifications rather than conceptual flaws."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This paper does what it claims: it takes the finite-player α-potential construction, normalizes by N, and shows that both values and approximate minimizers converge to a mean-field control problem with measure-valued controls; when α_N \to 0 the limit is a potential for the corresponding MFG, and you get propagation of chaos for controlled diffusions with common noise and non-separable interactions.\n\nWhat is new is concrete. First, they build the α_N-potential by a path integral on the joint state-control space (Theorem 2.1), which stays finite-dimensional and avoids sensitivity processes. Second, they prove a Poincaré lemma / Green formula on Wasserstein space under the C^{2,2} regularity that MFG people actually use (Prop. 5.1, Thms 4.1–4.2); that is the technical heart and it looks carefully done. Third, they identify the limiting MFC objective itself as the MFG potential via the interpolation maps κ_ε (Thm 4.3) and get the PoC corollary. The examples in 3.4 make the necessity of the triple hierarchy of measures clear.\n\nThe soft spots are the usual ones and the paper flags them. Assumption 3.1(iii) needs compact actions, bounded coefficients, and non-degenerate idiosyncratic noise for the compactness arguments that identify all accumulation points of approximate minimizers. Remark 3.1 notes that non-degeneracy can be dropped when the running cost depends only on the state law; that is honest. The proofs adapt Djete-style propagation-of-chaos arguments and extend Green’s theorem; I do not see circularity or a load-bearing gap under the stated hypotheses. Self-citations to the authors’ earlier α-potential papers are appropriate.\n\nThis is for people who work on potential games, MFGs of controls, or PoC with common noise. It is pure theory, self-contained, and the math is solid enough that a serious editor should send it to referees. I would bring it to reading group and expect to cite the Poincaré lemma and the limit identification.","headline":"Solid limit theory that cleanly turns vanishing α_N into potential MFGs and gives PoC under common noise and non-separable costs.","tokens_in":48493,"tokens_out":557,"would_cite":true,"duration_ms":8228,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60Fxx","91A06","91A14","91A15","91A16"],"pacs":[],"model":"grok-4.5","headline":"Normalized N-player α-potential functions converge to a mean-field control problem whose objective is itself a potential for the limiting MFG when α_N vanishes.","keywords":["α-potential game","potential mean field game","mean field control","measure-valued control","Poincaré lemma","Wasserstein space","propagation of chaos"],"falsifier":"Construct an explicit family of N-player costs whose Hessians remain asymmetrically large so that α_N stays bounded away from zero, yet the normalized potentials still converge to an MFC objective that is a potential for the limiting MFG; any such example would break the claimed equivalence.","tokens_in":48538,"feed_emoji":"∞","tokens_out":667,"duration_ms":6112,"temperature":0.7,"pith_summary":"The paper shows that α-potential games, which turn approximate Nash search into minimization of one function, become potential mean-field games in the large-population limit. After dividing by N, both the optimal values and the approximate minimizers of the finite-player α_N-potential functions converge to those of a mean-field control problem that uses measure-valued controls. The condition that α_N itself tends to zero is equivalent to the classical closedness conditions that make an MFG potential; under that condition the limiting control objective serves as a potential function for the MFG. The technical backbone is a Poincaré lemma on Wasserstein space that reconstructs the potential by path integrals of the cost derivatives. The same limit also yields propagation of chaos for controlled diffusions that may include common noise and non-separable control interactions, giving a systematic route from finite-player games to potential MFGs.","feed_headline":"N-player α-potentials limit to potential mean-field games","feed_subtitle":"When α_N vanishes, the finite-player potential becomes the MFG potential and chaos propagates","key_machinery":"The Poincaré lemma on Wasserstein space (Theorems 4.1–4.2 and Proposition 5.1): closed differential forms built from the cost Hessians are exact, so the mean-field potential is recovered by path integrals of the cost derivatives along absolutely continuous curves of measures.","core_discovery":"Both the optimal values and the minimizers of the normalized N-player α_N-potential functions converge to those of a mean-field control problem with measure-valued controls; moreover lim α_N = 0 is equivalent to the standard closedness conditions for potential MFGs, and the limiting MFC objective is itself a potential for the corresponding MFG.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["N-player α-potentials converge to mean-field control problems","Vanishing α_N yields potential MFGs via MFC limits","α-potential minimizers pass to measure-valued MFG controls","Limiting MFC objective is a potential for the MFG","Poincaré lemma on Wasserstein space closes potential MFG gap"],"cache_read_input_tokens":32896,"weakest_assumption_plain":"The idiosyncratic noise must be uniformly non-degenerate and the action set compact with bounded coefficients; without that non-degeneracy the lifted measure-valued controls may miss some accumulation points of the finite-player minimizers.","fun_headline_variants_meta":{"raw":{"variants":["N-player α-potentials converge to mean-field control problems","Vanishing α_N yields potential MFGs via MFC limits","α-potential minimizers pass to measure-valued MFG controls","Limiting MFC objective is a potential for the MFG","Poincaré lemma on Wasserstein space closes potential MFG gap"]},"model":"grok-4.5","effort":"low","cost_usd":0.003486,"raw_usage":{"total_tokens":1137,"prompt_tokens":786,"num_sources_used":0,"completion_tokens":74,"cost_in_usd_ticks":34860000,"prompt_tokens_details":{"text_tokens":786,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":277,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":786,"tokens_out":74,"duration_ms":3208,"temperature":1.0,"reasoning_tokens":277,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-14T18:07:44.018030+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Construct an explicit family of N-player costs whose Hessians remain asymmetrically large so that α_N stays bounded away from zero, yet the normalized potentials still converge to an MFC objective that is a potential for the limiting MFG; any such example would break the claimed equivalence.","supporting_citations":[],"review_version":2}